Digest of 38 Interviews: 16 Fund Managers, 11 Execs & Founders, 5 Economists, 6 Analysts
Bill Ackman, Tyler Cowen, & execs of Anthropic, OpenAI, AMD, Boing, Airbus, Honeywell, IMAX discuss private credit bubble, Kimi K3, AI race & export/import controls, Odyssey, SpaceX IPO, & more
Roundup
Investor Spotlight
Bill Ackman joins Money Rehab with Nicole Lapin to break down the opportunities he’s seeing in the market, which companies are in his portfolio, and what could trigger the next 2008. Bill walks Nicole through his investing playbook, the biggest mistake new investors make, and whether we’re in an AI bubble. Bill gives his takes on gold, Bitcoin, Chipotle, Starbucks, T-bills, Trump, and Mamdani. Bill also gets personal about his inheritance plans for his four daughters, what he actually thinks makes someone successful in business, and whether he would run for office.
07.20.26 | 46 min | YouTube | Apple | Spotify
The AI Trade (Bull vs. Bear)
Tom Thornton, President at Hedge Fund Telemetry, joins Bloomberg Businessweek Daily to discuss current market sentiment, specifically addressing the surge in bearish bets against US equities, risks associated with AI capital expenditure, and his perspective on China and Korean markets.
07.20.26 | 8 min | YouTube
Dan Niles, Niles Investment Management, joins 'Closing Bell Overtime' to talk the state of play in the AI trade as renewed competition out of China rattles US markets.
07.20.26 | 4 min | YouTube | Apple | Spotify
Katty Kay talks with Professor Gerry Tsoukalas, who recently co-wrote a paper with Professor Brett Falk titled "The AI Layoff Trap." They built a mathematical model to game out just how bad an AI layoff wave could get — and whether anything can actually stop it. According to their model, many companies may see this firing-your-customers trap coming and walk into it anyway. If companies eventually replace too many workers with AI, will there be enough people left with a paycheck to continue buying goods and keep the economy going?
07.20.26 | 10 min | YouTube
In this mid-year market review, Cambria CEO and CIO Meb Faber introduces his new book, Investing in America: The Rise of A 250-Year Bull Market, provides a comprehensive outlook on the current investment landscape, discusses the high valuations of US equities, the benefits of global diversification, highlights opportunities in foreign and emerging markets, and the tactical role of shareholder yield strategies.
07.20.26 | 50 min | YouTube
George Noble, Managing Partner of Noble Capital Advisors, joins David Lin to discuss the risks facing the AI and technology boom, the case for energy and gold, and why investors should focus on valuations, earnings, and market cycles rather than momentum.
07.20.26 | 50 min | YouTube | Apple | Spotify
Today on Prof G Markets, Scott Galloway and Ed Elson unpack OpenAI's turbulent week and assess the biggest threats facing the company. They discuss why China's latest AI models could pose a serious challenge and why Scott thinks there might be a leadership shake-up at OpenAI. Then, they examine the growing narrative that the market is becoming more diversified, arguing that AI's influence extends far beyond Big Tech, and identify the sectors they believe are least exposed to the AI trade.
07.20.26 | 1 hr 5 min | YouTube | Apple | Spotify
Sam Huszczo, founder & CIO of SGH Wealth Management, joins Bloomberg ETF IQ to discuss the resilience of the current bull market, attributing its durability to robust earnings growth rather than speculative fervor. He cautions against the rise of overly complex, greed-driven financial products and critiques the overuse of the term democratization, which he argues is often used to justify risky retail exposure to private credit. He identified small-cap stocks as a notably overlooked opportunity for investors.
07.20.26 | 7 min | YouTube
George Seay, founder and chairman of Anandale Capital, joins The Street, where he identifies a major market pivot point where investors should rotate away from chasing AI winners toward a more diversified approach. He remains bullish on long-term earnings growth while advocating for patience, advising investors to prepare for volatility in the fall by rebalancing portfolios and targeting high-quality franchises like Microsoft, Netflix, and HCA at more attractive entry points.
07.20.26 | 22 min | YouTube
David Bailin, the founder of the CIO Group, appeared on the Schwab Network to discuss how investors are overestimating earnings expectations, why major hyperscalers remain attractively valued, and why the AI infrastructure buildout and adoption cycle still has significant long-term potential.
07.19.26 | 7 min | YouTube
John Freeman of Ravenswood Partners and Kai Wu of Sparkline Capital appeared on the Schwab Network to discuss the risks AI poses to the software sector, the future demand for memory storage, and how increased AI coding capabilities may impact company valuations.
07.20.26 | 9 min | YouTube | Apple | Spotify
China AI Race
Tyler Cowen joins TBPN to discuss why he believes banning Chinese open source AI is impossible, the rise of "AI maniacs," why Europe is quietly becoming an AI powerhouse, the future of jobs in an AI economy, why data centers are good for local communities, AI-generated religions, the gambling boom, economic "vibecession," and much more.
07.20.26 | 35 min | YouTube | Apple | Spotify
China's Moonshot has reignited the AI race with its powerful Kimi K3 model. Crosslink Capital partner Mary D'Onofrio joins Ed Ludlow on Bloomberg Tech to discuss whether more efficient AI models are changing the economics of artificial intelligence, what it means for OpenAI and Anthropic, and where venture investors see the biggest opportunities across AI infrastructure and agentic software.
07.20.26 | 8 min | YouTube | Apple | Spotify
Peter H. Diamandis and crew are joined by Emad Mostaque, CEO of Stability AI, on ep. 272 of Moonshots to discuss the AI Sputnik moment sparked by the release of the Chinese model Kimi K3; the rapid acceleration of frontier models; the potential for recursive self-improvement in artificial intelligence; the implications of open- versus closed-source AI development and model sovereignty; the transformative impact of model quantization, which enables high-performance models to run on mobile and edge devices; and the broader economic and societal shifts driven by the exponential growth of intelligence.
07.19.26 | 2 hr 7 min | YouTube | Apple | Spotify
Bloomberg's Peter Elstrom joins Bloomberg Open Interest to discuss how the release of Moonshot AI's powerful Kimi K3 model has surprised Wall Street. Chinese AI models are emerging as serious, low-cost competitors to US frontier labs like OpenAI and Anthropic. These Chinese advancements are creating pricing pressure that could impact the valuation goals of major US tech companies. There is an ongoing policy debate about whether the US should restrict these foreign models to protect domestic firms or allow them to prevent hurting American businesses that rely on cost-effective AI tools.
07.20.26 | 4 min | YouTube
Invesco fund manager Fiona Yang appeared on the Merryn Talks Money podcast to discuss the volatility of the AI-driven market in Korea, her investment outlook on China and India, and the potential for companies using AI to become the next generation of investment winners.
07.19.26 | 38 min | YouTube | Apple | Spotify
Byron Deeter, a partner at Bessemer Venture Partners, joined CNBC’s Squawk on the Street to discuss why concerns over Chinese AI models are overstated, the importance of corporate trust and data security in model selection, and his perspective on the future pricing dynamics of leading AI models.
07.20.26 | 4 min | YouTube
Arthur Mensch, co-founder and CEO of French AI startup Mistral, joins The Economist’s AI writer Alex Hern to explain why he believes open source gives Europe its strongest chance to stay competitive. He sees the moment as critical for Europe. He warns that without building its own AI industry, the continent could become dependent on the U.S. for the technology’s future. The 33-year-old former Google DeepMind researcher has helped make Mistral Europe’s top AI company and one of the biggest open-source AI labs outside China.
07.14.26 | 43 min | YouTube (7 min) | Apple (Paid) | Spotify (Paid) | The Economist (Paid)
Deutsche Bank’s CIO of Emerging Markets, Jacky Tang, appeared on CNBC to discuss the structural drivers for China’s financial markets, the role of domestic Chinese AI models compared to US rivals, and the importance of investing across the entire AI value chain.
07.20.26 | 5 min | YouTube
AI pioneer Kai-Fu Lee appeared on Bloomberg Technology to discuss his startup 01.AI’s pivot toward enterprise applications, the company’s outlook for profitability and a 2027 IPO, and the competitive dynamics of the AI landscape in China versus the United States.
07.20.26 | 13 min | YouTube
Goldman Sachs analyst Ronald Keung appeared on CNBC International to explain how Chinese AI firms are leveraging open-weight models to drive revenue sharing, using API feedback loops for self-improvement, and navigating potential risks like data security and computing constraints in global markets.
07.20.26 | 4 min | YouTube
George Chen, a partner and digital practice chair at The Asia Group, appeared on “Bloomberg: The Asia Trade” to discuss the impact of Moonshot AI’s Kimi K3 model, the limitations of current US export controls on advanced AI chips, and the shifting competitive landscape between China and the US in the global AI race.
07.20.26 | 7 min | YouTube
AI Infrastructure & Power
On episode 69 of The Real Eisman Playbook, Steve Eisman sits down with Ben Kallo, sustainable energy and mobility analyst at Baird, to explore what may be the most important and least understood constraint on the entire AI trade: the U.S. power grid, which needs to add roughly 30 gigawatts of new capacity per year over the next decade. Ben walks through his top picks in the space, including GE Vernova and Tesla, where the autonomous vehicle and energy storage businesses are the real long-term story despite years of Elon's unkept promises.
07.20.26 | 50 min | YouTube | Apple | Spotify
Wendell Huang, the CFO of TSMC, appeared on CNBC International to explain the company’s increased investment in Arizona, discuss the reasons for profit margin dilution, and highlight the strategic benefits to the U.S. semiconductor ecosystem.
07.20.26 | 7 min | YouTube
AMD data center and AI leadership appeared on CNBC to discuss the launch of the Helios rack-scale AI system, the competitive strategy against Nvidia, and how the company is managing manufacturing and supply chain constraints to support major enterprise customers.
07.20.26 | 18 min | YouTube
Macro & Credit
Bloomberg News Credit Reporter Scott Carpenter and Bloomberg's Chief Correspondent for Private Capital Davide Scigliuzzo discuss an innovative financial strategy being proposed by UBS to help private credit and equity funds access new liquidity using A2 bonds.
07.20.26 | 7 min |YouTube
Mohamed El-Erian, The Wharton School professor and Allianz chief economic advisor, joins 'Squawk Box' to discuss his thoughts on prediction markets, latest market trends, inflation outlook, state of the AI boom, and more.
07.20.26 | 6 min |YouTube
Private credit has ballooned to roughly a trillion dollars, but Nick Nemeth of Mispriced Assets argues the danger isn't the banking system — it's insurance. In this Monetary Matters interview with Jack Farley, Nemeth lays out how private-equity-owned insurers have become highly leveraged holders of private credit and CLOs, why he thinks annuity surrenders could spark a run with no federal backstop, and how adjusted EBITDA, layered leverage, and lax loan ratings mirror the setup before 2008 — except, in his view, the scale looks more like 1929. He closes with contrarian rankings of Apollo, Ares, Blackstone, and Blue Owl.
07.14.26 | 1 hr 14 min | YouTube | Apple | Spotify
In Episode 490 of Hidden Forces, Demetri Kofinas speaks with Patrick Boyle, founder of the quantitative hedge fund Palomar Capital Management, and professor of finance at King's College London, about the SpaceX IPO and what it reveals about the integrity of America's financial markets, the cult of personality filling the vacuum left by the nation's collapsing institutions, the commercial and fiscal challenges facing Europe's largest economies, and the housing crisis and generational wealth divide driving the appeal of left-wing populism across the developed world.
07.20.26 | 53 min | YouTube | Apple | Spotify
Industrials & Aerospace
Boeing’s CEO Kelly Ortberg joins Bloomberg’s Guy Johnson at the Farnborough Airshow to discuss plans to ramp up production, demand and the outlook for artificial intelligence and robotics in aircraft manufacturing.
07.20.26 | 10 min | YouTube
Boeing CEO Kelly Ortberg appeared on CNBC to discuss the company’s progress on 737 Max certification, the commitment to meeting the schedule for the complex next-generation Air Force One program, and his strategy for rebuilding regulatory trust while investing in future production capacity.
07.20.26 | 5 min | YouTube
IMAX CEO Rich Gelfond appeared on Bloomberg Talks to discuss the record-breaking success of Christopher Nolan’s The Odyssey, the strategic challenges of distributing films in IMAX format, and his positive outlook for the company’s 2026 box office projections.
07.20.26 | 8 min | YouTube | Apple | Spotify
Airbus CEO Guillaume Faury discussed with Bloomberg at the Farnborough International Air Show the company’s robust demand, order backlog, strategic goal to introduce a next-generation single-aisle aircraft by 2030, and the company’s role in the future of European defense and jet fighter development.
07.20.26 | 7 min | YouTube | Apple | Spotify
Airbus CEO Guillaume Faury appeared on CNBC to discuss the company’s strong global aircraft demand, its progress in meeting production delivery targets despite engine supply chain challenges, and long-term plans for launching a next-generation single-aisle aircraft.
07.20.26 | 7 min | YouTube
Honeywell Technologies CEO and Chair Vimal Kapur joined the Leaders with Francine Lacqua podcast to discuss his career journey from engineer to executive, his strategic decision to split Honeywell into three independent companies, and his belief that artificial intelligence will primarily reshape work by augmenting human productivity rather than replacing it.
07.19.26 | 25 min | YouTube | Apple | Spotify
AI Founders & Executives
OpenAI chairman and Sierra co-founder, Bret Taylor, joins 'Squawk Box' to discuss the state of the AI boom, AI tokenmaxxing, ROI on AI spending, state of AI competition, Apple lawsuit, and more.
07.20.26 | 8 min | YouTube | Apple | Spotify
Boris Cherny, the head of Claude Code at Anthropic, appeared on the Odd Lots podcast to discuss the origins and safety-focused development of the coding agent, the evolving roles of software engineers in an era of AI-assisted development, and the future of agentic tools acting as collaborative coworkers.
07.20.26 | 1 hr 10 min | YouTube | Apple | Spotify
Venture Capital
A16Z General Partner and defense investor, Connor Love, appeared on Bloomberg Tech to discuss the expansion of Andreessen Horowitz’s American Dynamism strategy, the importance of building a robust domestic industrial and defense base, and the transformative potential of low-cost autonomous technology in modern warfare.
07.20.26 | 7 min | YouTube | Apple | Spotify
Deep Dive
Investor Spotlight Deep Dive
Bill Ackman, Pershing Square Founder & CEO, on AI Bubble Fears, His Bullish and Bearish Calls, and Building a Legacy | Money Rehab with Nicole Lapin
The State of the Economy and the AI Story
“I think the big story is really AI. It’s driving an enormous amount of entrepreneurship, and it’s giving access to intelligence to a very broad group of people.” — Bill Ackman
Ackman opened by framing AI as the defining force in the current economy, not just a market narrative. He argued it’s already a driver of entrepreneurship and job growth, but only for people who learn to use the tools rather than fear being replaced by them. Asked for his broadest read on the economy, he pushed back on the idea that anything currently feels uniquely “off,” noting markets always carry some degree of unease.
Is This an AI Bubble?
“In private markets there’s more risk of valuations being massively above where they should be — but you’re also seeing businesses able to go from zero to real scale incredibly fast.” — Bill Ackman
Pressed on whether AI valuations have gotten ahead of themselves, Ackman didn’t dismiss the concern outright. He pointed to a startup he’d looked at that raised a seed round at a $2 billion valuation with no product or revenue, now raising again at $5 billion — evidence, in his view, that private-market pricing has decoupled from fundamentals in places. His own hedge, rather than picking a winner among the AI labs, is to stay in dominant public businesses with durable moats.
Anthropic, OpenAI, and the Race for AI Supremacy
“At a certain point in time OpenAI was the leading company, and then Google was the leading frontier model company... there’s an enormous amount of capital out there looking for the next Anthropic.” — Bill Ackman
Ackman’s read on the frontier AI race is that leadership keeps rotating, which makes it a bad place for a concentrated bet. He named Anthropic as currently holding the lead but stressed how quickly that has changed hands already between OpenAI and Google. That instability is exactly why Pershing Square avoids trying to pick the AI labs directly and instead looks for picks-and-shovels-style businesses that benefit regardless of which model wins.
Inside the Pershing Square Portfolio
“You want to go to the Uber platform to order your car because you want the lowest-cost car that’s going to get you from place A to B in the shortest period of time.” — Bill Ackman
Ackman walked through several of Pershing Square’s roughly dozen concentrated holdings. He’s bullish on Uber, arguing the market is wrong to price in disruption from Tesla’s robotaxi ambitions, since consumers will chase the cheapest, fastest option regardless of platform. He cited Amazon’s dominance in convenience (using a pharmacy run in New York City as his foil) as the kind of moat he looks for. On recent portfolio moves, he confirmed buying roughly $2 billion of Microsoft and trimming Alphabet to fund it — a swap driven purely by relative price versus quality, not a bearish call on Google.
Advice for New Investors and the Power of Compounding
“You get into trouble when you try to make money really quickly — options, leverage, very speculative businesses — especially when you’re starting with a relatively small amount of capital.” — Bill Ackman
His core advice to new investors is to resist the urge to get rich fast. He used Berkshire Hathaway’s $48 share price in the 1960s as an illustration of how compounding, not speculation, is what actually builds wealth over decades. The lesson, in his framing, is patience: buy businesses you believe will compound at a high rate and let time do the work.
Why Ackman Won’t Borrow Against His Portfolio
“The key to being a successful long-term investor is being able to survive those challenging market periods when everyone else is panicking — you want to have money to invest when others are panicking.” — Bill Ackman
Asked why he doesn’t simply borrow against his holdings instead of selling, Ackman explained the mechanics of margin risk: a leveraged portfolio can trigger a margin call in a downturn, forcing a sale at the worst possible time. He said a meaningful share of Pershing Square’s profits over roughly twenty years have come specifically from having cash available to buy during periods when the market was in distress.
Bullish or Bearish: Gold, Bitcoin, Chipotle, Starbucks, T-bills, Trump, and Mamdani
“It’s an asset that’s only worth what people tell you they’re prepared to pay for it, and it doesn’t pay you any yield in the meantime.” — Bill Ackman, on gold
Nicole Lapin ran Ackman through a rapid-fire bullish-or-bearish round covering a wide mix of assets and public figures:
Gold — No real view either way. He called it an asset with no yield that’s worth only what someone else is willing to pay, and joked he’d rather advise husbands to buy their wives jewelry than own GLD.
Bitcoin — Owns none. He called Satoshi Nakamoto “a genius” and said he admires the construct, but has no idea whether Bitcoin is worth $5,000 or a trillion dollars, and doesn’t need an opinion to skip owning it.
Chipotle — One of Pershing Square’s most successful investments; the firm helped recruit Brian Niccol as CEO. Neutral on where the stock trades today.
Starbucks — Also tied to Niccol, who left Chipotle to run it; the turnaround has been tougher there, but Ackman believes the company is well positioned long-term.
T-bills — Fine for keeping cash safe, but he’d still rather own high-quality compounding businesses over the long run.
Trump — Bullish. Ackman said he backed him relatively early and is “optimistic about the last couple of years of his term,” adding that a president not worried about reelection can make decisions without regard to politics.
Zohran Mamdani — Bearish on his impact on New York City. Ackman called him smart, charismatic, and good at politics, but said his policy decisions are bad for the city’s finances and are discouraging business investment.
Fixing the Retirement Crisis
“Wages grow at a much slower rate than stocks over a long period of time.” — Bill Ackman
Ackman flagged that roughly 40% of the U.S. workforce — gig and service workers without employer retirement plans — lacks an easy way to start investing for the future. He pointed to the new “Trump savings accounts,” which let people begin with $1,000 and contribute annually, as a step toward closing that gap, arguing that market exposure matters precisely because wage growth alone won’t build wealth over time.
Family, Inheritance, and Whether He’d Run for Office
“You’re never going to inherit anything from me — so you’ve got to make it on your own.” — Ackman’s father, as he recalled it
Ackman traced his own drive back to a lesson from his father, a commercial mortgage broker who told him early on not to expect an inheritance. He’s tried to give his four daughters the opposite kind of security without removing their motivation — helping them afford to live somewhere like New York City if that’s what they choose, but deliberately not steering them toward a specific “safe” career. On whether he’d run for mayor of New York, he didn’t rule it out entirely, saying it’s “something I could do someday,” but suggested he could have more impact helping the right candidate get elected than running himself.
His Formula for Success
“The value of a business is the present value of the cash it generates over its life.” — Bill Ackman
Ackman closed by returning to first principles: durable, understandable businesses that compound over time beat chasing speculative trends, in AI or anywhere else. His parting tip for listeners was the same one that’s guided Pershing Square for two decades — invest in something built to withstand the test of time, and keep the discipline to hold it through the periods when everyone else is selling.
The AI Trade (Bull vs. Bear) Deep Dive
Scott Galloway & Ed Elson, Prof G Markets, on Sam Altman’s OpenAI Exit Odds, China’s AI Price War, and Why “Market Diversification” Is a Myth | Prof G Markets
Is Sam Altman on the “Green Mile” at OpenAI?
“He’s an innovator, not an operator. They need an adult in the room right now.” — Scott Galloway
Galloway opened with a bold prediction: OpenAI installs a new CEO within roughly six months, and his pick for the job is board chairman Bret Taylor. His case rests on a string of what he called executional missteps from Altman — struggling to manage employees who want to leave, an ill-fated hardware push that echoes a prior partnership, and the collapse of OpenAI’s expected role powering Siri after Apple chose Google instead. Ed Elson pushed back gently, noting the underlying business problem — compute costs far outrunning revenue — isn’t really an execution failure so much as an industry-wide open question, comparable to Amazon or Uber’s early years of heavy losses before profitability.
China’s AI Price War Threatens the Whole Industry
“If we are as dependent on AI as we are, it is clear that this is the fastest way to bring the industry to its knees.” — Ed Elson
The two debated how seriously to take Chinese AI labs undercutting US frontier models on price, calling it the “elephant in the room” for the entire American AI sector. Elson raised the possibility that sustained Chinese price competition could eventually force a reckoning at OpenAI or Anthropic. Galloway was more sanguine, arguing there’s room for both ends of the market the way there’s room for Emirates and Delta, or LVMH and mass retail — premium, better-IP frontier models can coexist with cheaper alternatives rather than lose to them outright.
Is the Market Really Diversifying Away From AI?
“Buying quote-unquote AI-adjacent stocks and calling it broadening is like ordering a Diet Coke with your double-double — you still bought a cheeseburger.” — Scott Galloway
Both hosts pushed back hard on the narrative that 2026’s market rally has broadened beyond Big Tech. They pointed to small caps as an example: some of the Russell 2000’s best performers this year are direct AI beneficiaries, including semiconductor company MaxLinear (up 424% year-to-date) and cloud computing firm Rackspace (up 567%), with roughly 24% of the entire Russell 2000 now carrying some AI exposure.
The same pattern shows up in sectors that look like diversification on the surface. Utilities are up 7% year-to-date, but the gains are concentrated in AI power providers like Dominion and Bloom Energy (up 143%). In real estate, the winning REITs aren’t homebuilders — they’re data-center operators like Digital Realty Trust and Iron Mountain. The conclusion: what looks like a broadening market is, on closer inspection, still just the AI trade wearing different clothes.
Wealth Inequality and the Mamdani Debate
“Capitalism works as long as it rests on a bed of empathy, and you keep reinvesting in the middle class with redistribution of income.” — Scott Galloway
The conversation drifted into a broader debate about wealth concentration, prompted by New York mayoral politics. Galloway argued that unchecked capital accretion at the top — helped along by a tax structure he called regressive — is what fuels support for more radical redistribution politics in the first place. Elson pushed back that Mamdani’s actual policy record so far has been more moderate than the “democratic socialist” label suggests, with Galloway conceding some of his moves have been “surprisingly reasonable” even as he remains skeptical of the broader movement.
Meb Faber, Cambria CEO & CIO, on His 60/30/10 Portfolio Default, the REIT Surprise of 2026, and Why Institutions Keep Overpaying for Bonds | Mid-Year Market Review
Rebalancing the “Default” Portfolio: 60/30/10
“I love Vanguard, John Bogle, the GOAT, to be on Mount Rushmore. However, I love to tease them...” — Meb Faber
Faber used his mid-year review to restate his baseline allocation framework: starting from the classic Bogle-style split of roughly two-thirds US equities to one-third international, he rounds it to an easier 60% US, 30% foreign developed, and 10% emerging markets as his personal starting point before any tilts. He’s quick to needle his “Boglehead” fans even while praising Bogle as one of the greats, arguing that pure indexers can be too rigid when market conditions shift.
The REIT Surprise Nobody Called
“If you would’ve asked me at the beginning of the year, would you expect global REITs to be up 12%? I would have said no. But here we are.” — Meb Faber
One of the mid-year’s biggest surprises for Faber has been real estate. Global REITs are up 12% this year, a result he says he wouldn’t have predicted given his fund’s value tilt — a reminder, in his framing, that diversification keeps paying off in places investors don’t expect.
Why Institutions Keep Overpaying for Bonds
“Will you buy bonds with no margin of safety? They’re like, ‘sure.’ ... All of a sudden they’re like, ‘Oh, hallelujah, they yield five now.’” — Meb Faber
Faber drew a pointed contrast between how investors treat expensive stocks versus expensive bonds. Ask someone if they’d buy a stock trading at 100 times revenue and they’ll immediately call it reckless, he said, but the same investors happily buy corporate bonds with essentially no margin of safety the moment yields tick up off the floor. His read: a lot of institutional buying is fundamentally price-insensitive, driven by liabilities that need matching rather than genuine value.
Updating the Ivy Portfolio and Leaning Into Trend-Following
“No one else admits to 50% in trend. They’ll run an optimizer, ask how much you should put in trend, and it says half — then they say, ‘well, obviously we can’t do that, it’s way too much,’ so they constrain the optimizer.” — Meb Faber
Responding to a listener who’s run his Ivy Portfolio strategy for two decades, Faber defended running a much larger trend-following allocation than most advisors are comfortable with — up to half a portfolio, in some cases — arguing that firms routinely override what their own optimization models tell them because the output looks too aggressive on paper, even when the underlying math supports it.
Where to Park “Safe” Money
Asked what actually counts as the safest asset to hold, Faber said Cambria doesn’t limit its own cash management to T-bills alone, spreading safe-money exposure across a broader set of short-duration, low-risk holdings rather than concentrating it in a single instrument.
George Seay, Anandale Capital Founder & Chairman, on the SpaceX Wake-Up Call, Buying the AI Dip, and Why Netflix Is His Top Pick | The Street
A Market Pivot Point: Rotating Away From the AI Winners
“For the last few years, the playbook has been pretty simple: AI winners. But now we’re starting to see leadership broaden out.” — Seay’s opening framing, put to him by his interviewer
Seay pointed to the SpaceX IPO as a genuine turning point for how investors are pricing the AI and data-center trade. The stock priced at roughly 80 times sales, and has since fallen below its IPO price — a signal, in his view, that investors are starting to take a harder look at how much they’ve bid up securities tied to AI infrastructure. He argued the market playbook is shifting from chasing a narrow set of winners toward a more diversified approach.
Why $80–85 Oil Is the Sweet Spot for the U.S.
Seay made the case that higher oil prices are now a net positive for the U.S. economy rather than a drag, given how much the domestic oil and gas industry has grown over the past two decades. He noted that 2008’s $148-a-barrel spike would be roughly $250 in today’s dollars once adjusted for inflation, versus current prices around $80–85 — a range he called ideal, since the U.S. benefits from oil staying elevated without tipping into the $95-plus territory that starts to hurt consumers.
Buying the Dip in Micron and Nvidia
Despite the broader caution on AI valuations, Seay said he’s still actively buying weakness in top chip names. He singled out Micron, which has been in a sharp sell-off, as a stock with “enormous” current earnings power that he’d add to on further declines, grouping it with Nvidia among the names he considers must-owns in this market — bought specifically when they get knocked down, not chased on strength.
How to Get International Exposure Without Overloading on China
For most investors, Seay recommended low-cost ETFs and index funds over stock-picking abroad, but with a specific warning: many broad international indexes are heavily concentrated in China, and investors can end up with 40% of their international exposure in a single country without realizing it. His advice is to look for cheap, broadly diversified international funds and check the underlying country weightings before buying.
The Case for Netflix as a Value Pick
“This is a premier franchise. They own their marketplace. They’re the gorilla in the space.” — George Seay
Netflix stood out as one of Seay’s top individual picks despite a rough stretch — the stock has been hit hard even after in-line earnings failed to reassure investors. He expects a wave of tax-loss selling in September and October to create a better entry point, framing Netflix’s post-2022 rebound (when it also traded down to mid-single-digit multiples before multiplying in value) as a preview of how the current dip could play out.
Disciplined Rebalancing Over Market Timing
Rather than trying to time an AI-trade top, Seay advocated for strict, data-driven rebalancing rules. His example: an investor targeting a 70% stocks / 30% bonds-and-cash mix who drifts to 80% stocks should trim back in phases — by 5% to as much as 10% — to restore the original allocation, regardless of what the market is doing at that moment.
Professor Gerry Tsoukalas, Co-Author of “The AI Layoff Trap,” on Why Mass AI Layoffs Could Backfire on the Companies Doing the Firing | BBC News
The Firing-Your-Customers Trap
“Who’s going to be left to buy products if everyone gets automated and replaced?” — Gerry Tsoukalas
Tsoukalas, co-author with Professor Brett Falk of a paper titled “The AI Layoff Trap,” laid out the core mechanism: as AI becomes capable of doing more jobs, individual companies have a clear incentive to cut costs by replacing workers. The problem is that those same workers are also consumers. If enough companies automate simultaneously, the aggregate result is a shrinking pool of people with income to spend — undermining the very demand every company depends on.
Why It’s a Dominant Strategy, Not a Choice
“No matter what you do, no matter what the other companies are doing, your best strategy is to adopt as much of this technology as possible.” — Gerry Tsoukalas
What makes the trap dangerous, in Tsoukalas’s telling, is that it isn’t a case of executives simply failing to see the risk. He described automation as a “dominating strategy” in game-theory terms — in a competitive market with many firms, any single company that holds back on automating risks losing to rivals who don’t, so the rational move for each individual company is to automate regardless of the collective consequences. He explicitly compared the setup to the prisoner’s dilemma, where individually rational choices produce a worse outcome for everyone.
The Odysseus Solution: Self-Restraint
“Companies need to slow down with the firing or replacement of human workers with AI. We’re calling for self-restraint.” — Gerry Tsoukalas
Tsoukalas’s proposed remedy draws on the Odyssey: rather than trust individual willpower in the moment, Odysseus had his sailors physically tie him to the mast before sailing past the sirens, so he couldn’t act on temptation even if he wanted to. He’s calling for companies — or policymakers — to build in similar pre-committed restraints on how fast they replace workers with AI, rather than relying on executives to voluntarily slow down. His interviewer was openly skeptical that kind of discipline is emerging in the tech industry, describing companies as behaving more like sailors racing straight toward the sirens than crews tying themselves down.
Tom Thornton, Hedge Fund Telemetry President, on Korea’s Leveraged AI Unwind, Why Oracle’s 2030 Goals Are in Question, and Going Long China Tech | Bloomberg Businessweek Daily
Korea’s Leveraged AI Trade Unwinds
“Investors were cashing out of their life insurance policies to put money into SK Hynix... that gets a little crazy.” — Tom Thornton
Thornton pointed to Korea as the clearest warning sign of excess in the AI trade. A 2x-leveraged SK Hynix ETF is down 73% from its high, the Kospi has fallen roughly 30% in the past month, and Bloomberg data show investors sold a record $30.5 billion of Korean equities in June with July selling already nearing $8 billion. His read: retail investors piled into leverage on one side of the boat, and the boat tipped over — a dynamic he flagged as early as late May.
Is This the Start of AI Capex Cuts?
“It’s not a matter of if, but a matter of when there’s going to be capex cuts.” — Tom Thornton
Thornton sees early cracks in the AI infrastructure story, pointing to Meta and SpaceX leasing out unused compute capacity as a signal that frontier models may not be consuming resources as fast as hyperscalers planned for. He compared the setup to the 2000 tech bubble, when telecom companies quietly stopped increasing equipment orders before the broader collapse followed — Q3 2000, in his telling, was the real inflection point, well before headlines caught up.
Bullish on Alphabet, Skeptical of Oracle’s 2030 Ambitions
Thornton singled out Oracle’s long-term growth targets heading into 2030 as increasingly likely to come under pressure, while staying bullish on Alphabet specifically — citing Gemini’s distribution advantage through Samsung, Android, and iPhone, plus Alphabet’s other income-generating businesses giving it more room to sustain capex than peers. He sees Anthropic’s strength concentrated in enterprise for now, distinct from Alphabet’s consumer reach.
Buying China Tech at Half Off
“I told everybody, buy these with two hands, and basically it was crickets. Nobody wanted them. And now they’re up 20%.” — Tom Thornton
Thornton has been positioned long China and Hong Kong tech — Alibaba, KraneShares CSI China Internet Fund, and iShares China Large-Cap ETF — after they fell roughly 50% from their highs. He cited Alibaba’s diversified approach (chips, AI models, cloud, consumer AI) plus a likely Chinese stimulus package and Alibaba’s 35% stake in Moonshot AI as reasons the trade has quickly turned around after being universally unloved just weeks earlier. He also flagged that cheaper, more efficient open-source Chinese models could pressure expensive U.S. token pricing from OpenAI and Anthropic over the next six months.
Short Tesla Into Earnings
Thornton disclosed a short position in Tesla ahead of earnings, citing three years of negative growth, an aging two-model lineup, and a long list of unfulfilled promises from the company’s CEO — specifically calling out the still-unanswered timelines for robotaxi and Optimus scaling.
Dan Niles, Niles Investment Management Founder, on Why Chinese AI Models Are a Bigger Threat Than Wall Street Realizes | Closing Bell Overtime
Chinese Models Prove They Can Compete
“There’s a saying that necessity is the mother of invention. They’re not getting the latest and greatest chips, so they’re asking: what innovations can we do to get around that, versus just throwing more hardware at the problem?” — Dan Niles
Niles argued markets are underestimating what Chinese AI labs can do without access to top-tier chips. He traced the pattern back to DeepSeek’s early breakthroughs with techniques like mixture-of-experts, and pointed to Moonshot AI’s Kimi K3 and Alibaba’s Qwen as proof that lower-cost, lower-chip-tier models can now genuinely compete with the best U.S. frontier models rather than just serve the low end.
Three Vectors of the AI Trade
Niles broke the AI trade down into three interacting forces: the largest AI spenders starting to cut their bills, token production climbing rapidly as generative AI usage grows, and the price companies pay per token falling as cheaper alternatives like Kimi and Qwen enter the market. His view is that the interplay of those three factors — not any single one — will determine how the trade plays out from here.
Why He’s Betting Infrastructure Over Models
“I think all these frontier models eventually become commoditized. So the play is better on the infrastructure side versus the model side, where every week a new model comes out and jumps to the top of the ranks.” — Dan Niles
Given how quickly leadership among frontier models keeps changing hands, Niles said his own positioning favors the infrastructure layer of the AI trade over the model layer, since picking whichever lab is out front in any given week is close to impossible.
Coinbase Cuts Its AI Bill in Half While Uber Blows a Year’s Budget in Four Months
“When you’ve got the top 1% of companies accounting for probably more than 50% of AI spend, if they’re trying to cut back, can the other 99% make up for it? I’m not sure that’s the case near term.” — Dan Niles
As a near-term warning sign, Niles cited Coinbase’s CEO publicly stating the company cut its AI bill by nearly 50% using open-source models even as token usage kept rising, contrasted with Uber disclosing it blew through its entire annual token budget in just four months. He sees a mismatch forming between a handful of dominant AI spenders potentially pulling back and everyone else being unable to offset it.
Not Oversold Yet, But Starting to Wade Back In
Niles cautioned against long-range predictions in this cycle, recalling that rosy multi-year forecasts made in 2000 preceded a 78% peak-to-trough crash in the Nasdaq — and pushed back specifically on the idea that memory chips are “no longer cyclical,” calling it a claim he’s heard before with the same bad outcome. Even so, after semiconductor names fell more than 14% in a month, he said the sector is down a lot but not yet technically oversold, and that he’s beginning to selectively add back into infrastructure names rather than waiting for a bottom signal.
Sam Huszczo, SGH Wealth Management Founder & CIO, on Why the Bull Market Isn’t as Fragile as the Headlines Suggest | Schwab Network
Don’t Fight the Bull: Why P/E Compression Isn’t Bearish
“Why can’t we just look at it and say, hey, be thankful that earnings are growing faster than the stock market?” — Sam Huszczo
Huszczo pushed back on constant “this rally is due to end” commentary, arguing the market’s resilience is being driven by earnings growth outpacing price gains, which has actually brought valuations down to a P/E near 20.5 — elevated, in his view, but not extreme. He referenced Greenspan’s “irrational exuberance” remark as a cautionary example, noting the market kept climbing another 105% after that warning before the next real bear market arrived.
“We Have Beat Up the Word Democratization”
“We’ve watered it down to nothing... it’s hawking private credit to retail people that don’t have the time horizon.” — Sam Huszczo
Huszczo singled out the private markets as one of the clearest signs of excess “greed mode” in the current cycle, arguing the word “democratization” has been stretched to justify selling illiquid, long-horizon products like private credit to retail investors who don’t have the matching time horizon to hold them.
Portfolio Construction: Pairing Uncorrelated Strategies
Walking through his own positioning, Huszczo defended JMOM, a momentum ETF with roughly 90% annual turnover that he likened to a Porsche 911 — fast, but with real crash risk. He noted the fund has recently rotated out of concentrated AI momentum names like AMD and Micron and into healthcare, telecom, and discretionary. His broader thesis is that real diversification comes from pairing two strategies that each work independently but move differently from each other — he cited a -0.52 correlation between momentum and equal-weight exposure as the kind of pairing that lets a portfolio hold up even when a previously hot factor, like the “Mag Seven,” cools off.
The Leveraged ETF Excess Getting “Out of Hand”
Huszczo flagged South Korea’s move to halt new listings of leveraged single-stock ETFs as a clear signal of excess spilling into U.S. markets too, comparing some of the newer leveraged and parlay-style ETF products to absurd combined bets (his example: a product paying out 10x if a stock rises 2% and the Fed cuts rates a quarter point). His advice isn’t to abandon models, but to stay aware when greed is creeping into product design rather than chase every new leveraged vehicle.
Small Caps: Outperforming in Every Sector
“Out of all 11 GICS sectors, every single one of them — small cap is outperforming large cap. That’s a factor play, not a sector play.” — Sam Huszczo
Huszczo highlighted small caps as an overlooked opportunity, noting they’ve had their fastest start to a year in two decades, with roughly two-thirds of issuances trading above their 200-day moving averages and heavily shorted names underperforming — a sign, in his view, that quality rather than just size is driving the move. He specifically pointed to strength in regional banks as a genuine positive read on the broader economy, since their performance tracks ordinary consumers paying their bills.
David Bailin, CIO Group Founder, on Why Amazon, Google, and Microsoft Are Undervalued, and Who Really Loses as AI Costs Fall | Schwab Network
Why Amazon, Google, and Microsoft Look Reasonably Valued
“The markets have definitely overreacted to a bunch of AI fears.” — David Bailin
Bailin argued the sell-off in AI-adjacent names has gone too far, noting the buildout of data centers and their supporting energy infrastructure is still only roughly one-third complete, making it premature to judge utilization. Among the major hyperscalers — Amazon, Microsoft, and Google, which he separates from Oracle — he sees valuations as relatively reasonable given current earnings, meaning the real market concern isn’t profitability but the trajectory of their capital spending.
The CapEx Hurdle: 110% Growth Wasn’t Enough
Bailin laid out the market’s actual expectations bar: infrastructure and chip spending was expected to rise 110% year-over-year in the most recent period, with another roughly 60% increase built into estimates for next year. That steep hurdle is why chip-related earnings have been so sensitive to any sign hyperscalers might slow capex growth — and why DRAM and other commodity chip makers are the most exposed if that growth rate comes in below plan.
China’s Cheaper Models Are a Pricing Story, Not a Demand Story
“It’s the volume that people are missing... utilization is nowhere near what it’s going to be two years, three years, four years from now.” — David Bailin
Asked about cheaper Chinese models pressuring the AI trade, Bailin reframed the concern: falling token prices are a natural and expected part of the market maturing, since expensive frontier models will increasingly be reserved for high-value uses like code generation while cheaper models handle common tasks. What he thinks investors are underweighting is the sheer growth still ahead in AI utilization as robotics, logistics, and call centers get automated over the coming years — volume growth that should outpace the price declines.
Rotating Into Private Equity and Healthcare
Beyond technology, Bailin flagged private equity as a particular favorite after a rough prior twelve months, pointing to rising deal activity and improving bank earnings as early signs of a turnaround. He also highlighted healthcare, which has been under pressure for close to two years, as showing signs of brightening — a source of sustained growth he sees as largely independent of how the AI trade plays out.
Winners and Losers as AI Costs Fall
Bailin’s framework for the next phase of the AI trade: the winners are hyperscalers and the broader universe of companies that use AI and benefit as its cost falls, while the losers are hardware makers whose pricing power depends on scarcity — chip and component manufacturers whose margins compress as supply catches up with demand over the next six to eighteen months.
John Freeman (Ravenswood Partners) & Kai Wu (Sparkline Capital) on the “SaaS Apocalypse,” Chipflation, and Whether AI Models Become Commodities | Schwab Network
Is the “SaaS Apocalypse” Overblown?
“If you give them five years, ten years, they’re going to be able to replicate most of enterprise software.” — John Freeman
Freeman called the recent software sell-off an overreaction in the near term, even as he acknowledged the longer-run threat is real: AI coding tools may not be able to replicate most enterprise software today, but a ten-to-fifteen-year horizon changes the calculus on what counts as defensible “terminal value” for a software business. He carved out exceptions like ServiceNow, which he sees building AI-agent management tools substantial enough to offset whatever revenue the company might lose to the same trend.
Which Software Moats Actually Survive AI?
“You have to ask: for which of these companies was the moat just the code?” — Kai Wu
Kai Wu framed the current software drawdown through the lens of past disruption cycles — brick-and-mortar retail, newspapers — where the consistent pattern wasn’t uniform destruction but increased dispersion between survivors and casualties. He pointed to the New York Times surviving the newspaper collapse and Walmart surviving the retail apocalypse as precedent: companies whose advantage was purely technical (just being able to write good code) are vulnerable now that AI coding tools commoditize that skill, but companies whose moat rests on customer relationships, brand equity, or network effects are better positioned to survive even a 60-70% stock decline, which he noted has already hit names like ServiceNow, Adobe, Salesforce, and Intuit.
Memory Chips: “Chipflation” and Why This Cycle Might Break the Pattern
“Memory companies who normally do 30-40% gross margin in a good year are doing 80% gross margin plus — and yet they’re priced like cyclical companies.” — John Freeman
Freeman addressed SK Group’s chairman publicly warning that memory prices are abnormally high and that supply must increase to prevent further “chipflation.” His counterpoint: markets are pricing memory makers as purely cyclical even though margins have exploded well beyond historical norms, and AI capex is consuming memory at a pace that could climb another five-to-tenfold if large models shift toward training on video rather than text. He compared the setup to semiconductor equipment makers like Lam Research and ASML, which broke out of their old cyclicality around 2010 and never fully returned to it — arguing memory stocks trading at six to seven times earnings look attractive on the same logic.
China’s Open-Weight Models and the Jevons Paradox Debate
Kai Wu addressed the rotation out of the AI capex trade following Moonshot’s Kimi release and other competitive open-weight Chinese models, acknowledging the legitimate risk that less compute per model could mean the broader infrastructure buildout has been overdone — a pattern he compared to the telecom overbuild in the dot-com era and the railroads a century before that. His counterargument invoked the Jevons paradox: if the cost of running AI drops, elastic demand could mean more total consumption, not less, opening up use cases that were previously too expensive to justify. In his view, chip makers benefit either way, regardless of whether the winning models end up open- or closed-source.
Where Value Flows if AI Models Become Commodities
Kai Wu closed on a bigger-picture risk facing the closed-source labs specifically: as competition intensifies from both Chinese open-weight models and well-funded U.S. rivals like Meta and Google, it’s possible AI genuinely transforms society while the underlying models themselves become commoditized utilities — meaning the economic value could flow to other parts of the stack rather than concentrate in the labs currently commanding trillion-dollar valuations, even before any of them have gone public.
George Noble, Noble Capital Advisors Managing Partner, on Why This Is a Bigger Bust Than Dot-Com, His $30 SpaceX Target, and Going Long Energy, Long Gold | The David Lin Report
“One of the Biggest Bubbles in History” Is Unwinding
“If it’s not tech, it’s dreck.” When someone says something like that, it kind of tells you where you are in the cycle. — Nancy Nierman, Warburg Pincus, February 2000, as recalled by George Noble
Noble opened with a stark call: he believes markets are witnessing the unwinding of one of the biggest bubbles in history, comparable to the dot-com crash but larger. He cited strategist Julian Garran’s calculation that current malinvestment in AI infrastructure is roughly 17 times what occurred during the dot-com era, driven by the fact that this boom is far more asset-intensive and involves much larger sums relative to the size of the economy. He thinks both Oracle and OpenAI could ultimately go bankrupt.
Korea’s Margin Call Crisis
“1.2 million South Korean trading accounts got margin called — that’s about 10% of all brokerage accounts in the entire country, just in the last two weeks.” — George Noble
Noble pointed to South Korea as the leading edge of the unwind. SK Hynix has dropped roughly 70% in recent weeks, and major U.S. semiconductor names — AMD, Qualcomm, Nvidia, and Intel — have all rolled over from peaks between May and early July. Despite the sharp declines, he noted that roughly $25 billion has still flowed into the leading semiconductor ETFs since the June peak, which he reads as evidence that positioning hasn’t actually turned bearish yet, even as prices have.
Valuing Semiconductors Like Shipping Stocks
“Anyone who values a shipping stock at a price-earnings ratio is financially illiterate. That’s exactly where we are with semiconductor stocks.” — George Noble
Noble’s central framework for semiconductors is that they’re a capital-intensive, asset-heavy commodity business — much like shipping — where temporary supply-demand imbalances produce gross margins (he cited memory names running 75%+ versus a 25% long-term average) that get arbitraged away as soon as new capacity comes online. His view is that current earnings are a cyclical peak being mistakenly capitalized as if they were sustainable, and that the honest way to value these businesses is against asset or book value, not trailing earnings.
The SpaceX Short: Index Inclusion, Float Unlocks, and a $30 Price Target
“This is not in the public interest. I don’t care if nothing illegal is done — if it’s not illegal, it should be illegal.” — George Noble, on SpaceX’s index-driven rally
Noble said his firm was among the first to flag SpaceX’s IPO valuation as excessive (roughly 11 times sales), but didn’t recommend shorting it immediately because they expected — correctly — that index inclusion would force buying regardless of fundamentals. FTSE Russell (with roughly $12 trillion indexed to it) added SpaceX while S&P did not, and Noble’s team called the top of the resulting squeeze to subscribers around $145-150 on the way down from a peak of $225. He argues the real story now is the float: only about 5% of shares were listed at IPO, with another 20% unlocking imminently and roughly 7% more every few weeks after that, meaning early private investors who bought at $10-50 a share are about to be able to sell into a stock he values at closer to $30. He was sharply critical of regulators and bank executives who enabled the listing, and dismissed Elon Musk’s claim that SpaceX could eventually be worth more than the rest of Earth combined.
Also Short Tesla, and Betting Oracle Could Go Bankrupt
Noble disclosed a Tesla short as well, arguing the stock trades at roughly 14 times sales for what is fundamentally an auto company that should trade closer to one times sales — implying a fair value in the $30-50 range versus its current price, and noting Tesla has generated only $38 billion in cumulative profit across its entire history against a valuation near $1.5 trillion. He separately floated the possibility that Oracle could go bankrupt, calling it potentially “the next Cisco” given its debt load and its financial ties to OpenAI, which he also expects to eventually fail. He additionally warned that a wave of pending IPOs — SpaceX, Anthropic, OpenAI — could flood the market with new supply at a scale large enough to pressure valuations independent of fundamentals, given how much of the market’s recent strength has rested on light issuance and heavy buybacks.
Long Energy, Short Tech
Noble has been positioned long energy and short tech since turning bullish on energy in December, citing a decade of underinvestment (real capex down roughly 70%) against depletion rates north of 5% a year. He acknowledged a rough patch for energy stocks even as oil prices held up, which he attributes to temporary, one-off factors — Chinese oil import cuts, SPR drawdowns, and a rebound in Strait of Hormuz shipping traffic — that he expects to reverse, alongside near-record-high speculative short positioning in oil among retail traders.
Why He’s Bullish on Gold Again
“Rising rates in a country that can’t afford rising rates are bullish for gold.” — Luke Gromen, quoted by George Noble
Noble admitted his prior bullish gold call missed the recent pullback, driven by a stronger dollar and rising real yields — but argues dollar strength will prove short-lived. His longer-term thesis rests on the idea that a debt load this large (he cited $40 trillion in federal debt and $125 trillion in off-balance-sheet liabilities) makes a genuine rate-driven economic slowdown, and eventual easing, more likely than not — and that the U.S. is heading toward what he called a “soft default” via currency debasement rather than any formal default, which is ultimately the reason to hold gold for wealth preservation in real terms.
Lessons From Peter Lynch: “Know What You Own”
Reflecting on his time at Fidelity under Peter Lynch, Noble said the core lesson was simple: know what you own, follow price and trend rather than stories, and be skeptical of anyone using jargon nobody in the room actually understands. He warned that today’s market has become dominated by retail momentum-chasing where valuation was irrelevant on the way up and, in his view, will be equally irrelevant on the way back down. He closed by noting that market cycles have humiliated even legendary investors before — Julian Robertson closed his fund, Fidelity’s George Vanderheiden lost his mandate, and Merrill Lynch’s Chuck Clough was pushed out as chief strategist — not because they became bad investors, but because, in his words, the market temporarily lost its mind.
China AI Race Deep Dive
Emad Mostaque, Stability AI Founder, and the Moonshots Crew on Kimi K3’s “Sputnik Moment” and Why Frontier Intelligence Just Became a Perishable Asset | Moonshots
Kimi K3’s “Sputnik Moment”
“They didn’t just close the gap, they jumped the fence.” — Peter Diamandis
The panel convened an emergency episode after Moonshot AI released Kimi K3, a 2.8 trillion-parameter open-weight model that jumped straight to the top of multiple leaderboards — including first place on the front-end coding arena, surpassing Anthropic’s latest Claude model, and landing as the third point on the cost-versus-performance frontier alongside the leading U.S. labs. The panel’s host called it an AI “Sputnik moment,” and co-host Alex noted Moonshot’s models have arguably held state-of-the-art among open-weight models for nine of the past twelve months, largely unnoticed until this release forced the comparison.
No Magic, Just Engineering
“What are the American frontier labs spending their money on?” — Alex, Moonshots panelist
A recurring theme was that Kimi K3’s published architecture contains no secret breakthrough — it’s a recognizable transformer with well-understood refinements to mixture-of-experts routing and attention linearization, not some undisclosed post-transformer method some assumed OpenAI or Anthropic might be sitting on. Emad Mostaque compared the achievement to Chinese EVs outcompeting legacy automakers: not a new invention, but disciplined engineering and manufacturing execution squeezing more out of known ingredients, including running the model efficiently on older-generation Nvidia H800 chips rather than the latest hardware.
Frontier Intelligence Is Now “A Totally Perishable Asset”
“The shelf life is weeks now. Any enterprise interested in the very latest cutting-edge model doesn’t have time to evaluate it, do an RFP, look at other models — by the time they decide, they’re three generations behind anyway.” — Salem, Moonshots panelist
The panel argued this shifts where value accrues in AI: not in owning any single frontier model, since leadership now changes every few weeks, but in the interface and orchestration layer that can swap between models as they leapfrog each other. One panelist floated that this makes frontier labs increasingly resemble utilities rather than defensible platforms.
What Happens to OpenAI and Anthropic’s Trillion-Dollar Valuations
Moonshot’s own valuation was pegged around $20 billion versus roughly $1 trillion apiece for OpenAI and Anthropic. One panelist estimated that if these were public companies, the news would have wiped out perhaps 30% of frontier lab valuations in a single move, and offered a rough guess that frontier labs broadly might now be worth closer to a quarter of their valuation from three months prior — reasoning that few enterprises will keep paying premium API prices once a comparable open-weight model can be run in-house for a fraction of the cost, similar to how U.S. drug prices subsidize R&D that’s later available elsewhere as cheap generics.
The Chip Embargo Backfired
“The embargo of chips on China was totally harebrained — it was enough to irritate but not enough to actually work.” — Dave, Moonshots panelist
The panel argued U.S. export controls on advanced Nvidia chips didn’t stop Chinese AI progress — they accelerated it, forcing labs like Moonshot to pursue efficiency breakthroughs (quantization, mixture-of-experts refinements, stripped-down training data) that Western labs, flush with compute, had less incentive to prioritize. The consensus was that those efficiency gains are now permanent and transferable, regardless of future chip policy.
Quantization: Squeezing Frontier Intelligence Onto a Smartphone
The panel also covered Bonsai, a 27-billion-parameter model from Caltech-linked startup Prism ML that runs entirely on a smartphone by compressing model weights down to roughly one effective bit per parameter (from a typical 16-bit baseline) with only modest accuracy loss — alongside similar compression work from Tencent’s former Wizard LM team. The implication, as the panel framed it, is frontier-class intelligence becoming available fully offline, on-device, at a fraction of the size and cost previously assumed necessary.
The Immigration Angle: Why Moonshot’s Founder Isn’t American
The panel dug into Moonshot AI founder Yang Xilin’s background — a Carnegie Mellon PhD who ultimately built his company in China — as a jumping-off point for a broader complaint about U.S. immigration policy failing to retain top AI talent. One panelist noted research showing that while the majority of Indian-born PhDs stay in the U.S., roughly 80% of Chinese-born PhDs return home, largely because China’s startup ecosystem offers a more viable path to building a company than trying to do so from abroad — a dynamic they argued the U.S. could address by automatically granting green cards to STEM PhD graduates.
Frontier Model Releases Are Accelerating Toward Daily Drops
The panel highlighted that frontier model releases have compressed from one every 60 days in 2024 to one every 10 days recently, and that extrapolating the trend implies near-daily frontier releases by January — a pace they suggested could make individual model announcements less newsworthy than the applications and use cases built on top of them.
Arthur Mensch, Mistral CEO & Co-Founder, on Why the AI Race Won’t Have a Single Winner and How Europe Builds Sovereign AI | The Economist
Why the AI Race Won’t Have a Single Winner
“You don’t have a single energy provider in the world that is monopolistic. AI should be compared to electricity.” — Arthur Mensch
Mensch pushed back on the framing of AI as a race with one ultimate winner, arguing the technology’s addressable surface is simply too large — spanning coding, physics, life sciences, audio processing, and symbolic reasoning — for any single company or country to dominate every domain. He said Mistral has deliberately chosen where to compete hardest, citing strength in areas like document processing, audio, and symbolic mathematics, often in combination with Europe’s historical strength in manufacturing.
The Trillion-Euro Problem: AI as a Geopolitical Trade Imbalance
“If all of this flows back to the US, you’re increasing the commercial imbalance to a point where this creates enormous instabilities... the commercial balance of the world is also the reason why we live at peace.” — Arthur Mensch
Mensch laid out the economic stakes with a rough calculation: European wages total around €9 trillion, and if AI eventually displaces even 10% of that labor, that’s roughly €1 trillion a year in value. If that value flows entirely to a single foreign provider rather than being captured domestically, he argued, the resulting trade imbalance becomes destabilizing at a geopolitical level — not just for Europe, but for India, Africa, and other regions dependent on a single AI supplier.
Comparing AI to Energy, Not Software
Mensch’s core analogy throughout was energy rather than traditional software: no country wants to be fully dependent on a single external electricity provider, both because the market is too large for one monopolist and because of basic continuity risk if that supplier gets cut off. He expects AI to follow the same import-export-produce pattern as energy — Europe will keep importing stronger models where they exist (citing manufacturing-related exports back to the U.S.), while also building and exporting its own sovereign technology in areas of strength, rather than the market consolidating around one or two global providers.
Selling Sovereignty to European Citizens, Not Fear
Mensch said the realization that “AI is a sovereign technology” — economically and militarily — is spreading faster at the national level than inside EU institutions, which he said are structurally built for consensus rather than industrial policy, making them slow to act even as individual countries move quickly. He named France, Luxembourg, Greece, Sweden, and Spain as governments actively partnering with Mistral. His approach to building public buy-in deliberately avoids leading with job-loss anxiety, instead equipping governments with sovereign AI tools that visibly improve public services — job search, legal guidance, social security, and tax administration — so citizens experience tangible benefits rather than only hearing threats to their livelihoods.
AI as the Answer to Europe’s Aging Workforce
Mensch pointed to Europe’s aging population and shrinking civil service workforce as a second practical argument for AI adoption: rather than replacing government workers, AI can make existing staff meaningfully more productive, helping states absorb the wave of retirements without a corresponding collapse in public service capacity.
Fiona Yang, Invesco Asia Dragon Fund Manager, on Korea’s Memory Supercycle, Cheap Chinese Models, and Overlooked Winners in India | Merryn Talks Money
Korea’s AI-Driven Volatility and the “Value-Up” Story
Yang described Korea as a front-row seat to the AI trade and its volatility in equal measure — the Kospi has swung roughly 25% off its June peak, with highly leveraged retail investors taking real losses on margin debt. She credited Korean regulators with a genuine “value-up” corporate governance push, boosting shareholder returns and minority protections in a way she compared favorably to Japan’s earlier reform cycle, though she noted investors chasing the AI narrative have largely overlooked that structural improvement.
The Memory Supercycle: How Long Can It Last?
“Even a fantastic earnings cycle cannot meet that expectation.” — Fiona Yang, on Samsung Electronics’ post-earnings share-price drop despite 19x year-over-year profit growth
Yang traced the current memory shortage back to years of underinvestment: NAND and DRAM producers were loss-making just two to three years ago and had no incentive to expand capacity, so when generative AI demand surged, supply couldn’t respond fast enough. She expects the shortage to persist for a couple of years given how long new capacity takes to build, but cautioned that elevated share prices may not hold for that entire window since so much good news is already priced in — and a Chinese competitor’s upcoming IPO could add supply faster than the market currently expects.
Two Risks to the AI Trade: Hyperscaler ROI and Cheap Chinese Models
“[A contact] just switches between the models — certain tasks we use the 90% cheaper Chinese model, and for the most advanced tasks, maybe we use the Western model.” — Fiona Yang, relaying a conversation with a software company operator
Yang flagged two specific risks to the demand side of the AI trade: first, that current revenue run-rates at OpenAI and Anthropic make the return on hyperscalers’ capex genuinely questionable; second, that cheap Chinese large language models — built on cheap power, land, and hardware — are pulling task-by-task workloads away from premium Western models at a fraction of the cost, undermining the assumption that everyone will always pay for the best available model.
Beyond the Picks and Shovels: Bonuses, Banks, and Construction
Rather than staying concentrated in the obvious AI hardware names, Yang has been positioning in secondary and tertiary beneficiaries. She pointed to Samsung and SK Hynix employees receiving enormous profit-linked bonuses (potentially a meaningful share of national GDP if projections hold), which she expects to flow into luxury goods, travel, and housing — already visible, she said, in Korean department stores seeing queues for luxury boutiques. Her portfolio plays on that spending wave include Korean banks like KB Financial for loan growth and wealth-management demand, insurer Samsung Fire & Marine for improving shareholder returns, and Samsung E&A, a captive construction company benefiting directly from the group’s data-center buildout.
China: Cheap, Unloved, and Used as a Funding Source
Yang said China remains genuinely inexpensive but structurally unpopular, largely because investors have been selling Chinese holdings — historically the “must-own” names in any Asia allocation — specifically to fund purchases of the more exciting Korea and Taiwan AI trade. She noted China’s index is heavy with internet names now facing real questions about whether they can reinvent their business models and compete on large language models, which has weighed further on sentiment even as she still finds selective value there.
India’s Overlooked Winners: Shriram Finance and Delhivery
After a research trip to India, Yang highlighted two names she considers underappreciated despite the broader market’s continued rich valuation. Shriram Finance, a non-bank lender serving truck drivers and small businesses in rural India, recently attracted a major investment from a large Japanese financial group, which she expects to improve its credit rating and lower its funding costs as it grows alongside rural formal-banking penetration. Delhivery, a logistics company with roughly 50% share of metro parcel delivery and a growing full-truckload business, is benefiting from continued infrastructure investment that’s lowering costs and letting it take share from competitors.
Asia’s Energy Independence Push
Yang described a broader regional shift, accelerated by the Ukraine war and Iran-related disruptions, toward energy independence and onshoring rather than pure cost optimization. She’s positioned in companies tied to that build-out, including Samsung E&A’s energy and resources construction work and Worley, an Australian engineering firm that designs projects for global oil and resources majors and derives roughly half its revenue from U.S. gas-production expansion.
Why Australia Is Her Favorite Market for Corporate Governance
Yang named Australia as the strongest market she covers on corporate governance and shareholder returns, and highlighted its role as a net commodity exporter — particularly copper, which she expects to benefit from continued data-center buildout — as a structural advantage. Near-term, she flagged government housing-market interventions (including anti-negative-gearing policies) as a source of consumer-confidence softness, given how concentrated Australian household wealth is in property, but framed it as a temporary drag rather than a structural problem.
Economist Tyler Cowen on Why Banning Chinese Open-Source AI “Cannot Work,” and the Rise of Teenage “AI Maniacs” | TBPN
Why Banning Chinese Open-Source AI “Cannot Work”
“The attempt to outlaw it or ban it or use sanctions against it is going to fail miserably... I hope the Trump administration gives up on this crusade.” — Tyler Cowen
Cowen argued that open-weight Chinese models are already proliferating the way any freely copyable software does, and that major U.S. companies are quietly relying on Chinese open-source models for cost savings right now — including, he noted, competitors like Thinking Machines using one to bootstrap part of their own product development. His view is that trying to legislate this away is unworkable given how deeply it’s already embedded in enterprise supply chains, even if compute infrastructure remains a genuine point of leverage in a way pure software never was.
Open Source and America’s AI Stack Are Complements, Not Rivals
Cowen pushed back on the idea that Chinese dominance in open-source models is a slippery slope toward losing the rest of the AI stack — chips, data centers, and energy. He argued the two are actually complementary: enterprises are more willing to build on U.S. infrastructure and lock into American systems specifically because open-source alternatives exist in the background as a hedge against being cut off, reducing the fear that’s driven some of the resistance to relying entirely on closed U.S. providers.
The “Vibecession”: Why Sentiment and the Data Have Diverged
“I call it negative emotional contagion... if you look at wealth accumulation, the job market, stock prices, real wages, many other indices, they’re doing okay to fine. They’re just not doing terribly.” — Tyler Cowen
Cowen described the current gap between decent economic data and poor consumer sentiment as a self-reinforcing negative mood he doesn’t think is really about the numbers — comparable to past eras where distrust in institutions, seeded by events like COVID-era messaging, outlasted the conditions that caused it. He also revisited his earlier, now largely vindicated call that AI’s growth impact would resemble the internet’s boost to GDP in the late 1990s (roughly half a percentage point) rather than an instant economic transformation — real, but gradual.
AI Maniacs: The Next Generation of Founders
“You’re going to have a large number of companies with quite a small number of employees but pretty high revenue.” — Tyler Cowen
Cowen described a growing cohort of teenagers — he cited 13- and 17-year-olds he’d spoken with in Abu Dhabi and Finland — who have taught themselves to build with AI tools well beyond what their teachers understand, and predicted this group will increasingly compete directly in medicine, law, consulting, and banking. He said little government encouragement is needed beyond keeping models affordable and giving young people access to capital and authority, and noted he’s been surprised by Europe’s improving performance on this front, attributing it partly to agentic AI being genuinely novel there in a way chatbots weren’t.
What the Jobs Data Actually Shows
Cowen said the unemployment rate for 18-to-24-year-olds has returned to pre-large-language-model levels, which he takes as evidence AI isn’t broadly destroying entry-level jobs, even as anecdotes suggest otherwise. He attributed some of the labor market’s softer optics to a large sectoral reallocation — healthcare hiring strong, prestigious “cushy” tech-track jobs scarcer — rather than an AI-driven collapse, and separately argued that companies conducting highly public AI-driven layoffs are making a strategic mistake: firms confident in their AI advantage and business durability should be hiring aggressively, not cutting headcount for optics.
Crosslink Capital Partner Mary D’Onofrio on Why Kimi K3 Doesn’t Prove OpenAI Is Overvalued | Bloomberg Tech
Kimi K3 Sparks a Chip Stock Rebound
“China and open-weight models are closing the gap with frontier models... the productivity gains are tremendous.” — Mary D’Onofrio
D’Onofrio joined right as chip stocks were rebounding from a Friday bear-market dip triggered by Kimi K3’s release, comparing the moment to last year’s DeepSeek shock. She sees the 2.8 trillion-parameter model’s strong benchmark performance (ranking fourth by Moonshot’s own measures) as another data point that open-weight models are genuinely closing the gap with frontier labs, which she expects to be a long-term boon for AI adoption broadly, even if it’s unsettling in the short term. Her broader takeaway on competitive dynamics: models keep leapfrogging one another, so any company’s real moat has to come from data, performance, or customer relationships rather than from any single model’s temporary lead.
Why Cheaper Models Might Not Shrink Total AI Spending
D’Onofrio, who focuses on AI infrastructure and vertical AI as a venture investor (including agent infrastructure company Teleport), pushed back on the idea that a 95% drop in inference costs necessarily reduces total dollars spent on AI. Her view is that cheaper models mostly shift workloads — expensive, frontier-level reasoning tasks migrating to cheaper models that still deliver adequate performance and latency — rather than shrinking the overall AI budget. She also noted that Kimi K3’s actual token pricing ($3 per million input tokens, $15 per million output) is roughly in line with many leading models, undercutting the popular narrative that it’s dramatically cheaper across the board.
Skeptical of the “Kimi K3 Proves OpenAI Is Overvalued” Argument
“I don’t know yet, and I don’t think anybody knows yet.” — Mary D’Onofrio, on whether Kimi K3 undermines OpenAI and Anthropic’s valuations
Asked about the reaction that a $20 billion Chinese startup producing a 2.8 trillion-parameter model at this cost structure calls into question trillion-dollar valuations at OpenAI and Anthropic, D’Onofrio was cautious. She pointed out both labs already generate billions in revenue across diversified first-party and third-party businesses, and noted that “open weight” doesn’t mean free to actually run — a nuance she thinks the market is underappreciating in the rush to declare frontier labs overvalued.
What Recent AI IPOs Reveal About Investor Appetite
D’Onofrio read recent listings like SpaceX and SK Hynix’s ADR offering as a signal that public investors want AI-infrastructure exposure, but are gravitating specifically toward long-established, revenue-generating businesses rather than purely speculative newcomers — evidence, in her view, that even as the IPO window reopens around AI, investors remain comparatively risk-averse and are seeking relative safety within the theme.
Deutsche Bank’s Jacky Tang on Why Chinese AI Models Won’t Displace Their US Rivals, and China’s Structural Drivers for the Second Half | CNBC International
China’s Second-Half Drivers: AI, EVs, and Energy Security
Tang laid out several structural reasons Deutsche Bank expects Chinese equities to keep performing through the second half: momentum from the Shanghai AI conference and Kimi K3’s promising, lower-cost performance; strength in EVs and advanced technology; an energy-security theme where Asian countries diversifying their energy mix post-conflict benefit Chinese energy-infrastructure players; and domestic reforms including state-owned-enterprise restructuring and support for export leaders. His base case assumes a fragile ceasefire holds, keeping oil in a $75-85 range rather than spiking above $100, with inflation gradually easing — a broadly supportive backdrop for equities.
Why Kimi K3 Won’t Displace US Models in the West
“We don’t think that the Chinese model is going to replace the US model at this point.” — Jacky Tang
Fielding investor questions about whether Kimi K3 would drive a switch away from U.S. AI models, Tang was direct: no. He pointed to Western corporations and governments remaining wary of Chinese models on data-security grounds, noting that DeepSeek’s adoption two years ago stayed concentrated in China and select emerging markets rather than spreading internationally, reinforced by ongoing U.S. and EU restrictions on cross-border data flows and AI-compliance requirements. He read Xi Jinping’s conference remarks about AI “collaboration” as an acknowledgment that the AI world remains genuinely fragmented rather than converging around one dominant model.
Chinese AI Labs as an Export Product for Emerging Markets
Rather than viewing Chinese AI labs as household names competing directly with OpenAI and Anthropic, Tang frames them as an export product aimed primarily at emerging markets, where adoption is already meaningfully higher than in the West. He said investors shouldn’t expect major Western market share for these labs, but should still pay attention given how large the aggregate emerging-market opportunity is.
Own the Whole AI Value Chain, Not Just the Mag Seven
Tang’s broader investment strategy argues for looking past the Magnificent Seven and U.S. hyperscalers to own the full AI value chain — Korean and Taiwanese semiconductor makers, Chinese AI labs, and the construction and utility companies building supporting infrastructure — since leadership and value capture rotate quickly between different links in that chain.
Bloomberg’s Peter Elstrom on Why Kimi K3 Threatens OpenAI and Anthropic’s Trillion-Dollar IPO Ambitions | Bloomberg Open Interest
Not Another DeepSeek Moment — Something Bigger
Elstrom said U.S. tech companies believed they understood where Chinese competition was coming from — DeepSeek’s breakthrough last year, and Alibaba’s Qwen model more recently — which made Moonshot’s Kimi K3 release a genuine surprise. The 2.8 trillion-parameter model outperforms most frontier models on many benchmarks, trailing only the leading models from Anthropic and OpenAI, at a cost structure Elstrom said the market still doesn’t fully understand.
Threatens Premium Pricing and OpenAI/Anthropic’s IPO Ambitions
“It is a very serious threat for some of these frontier models in the U.S. that had been planning on charging premium prices.” — Peter Elstrom
Because Kimi K3 and similar Chinese models are open-weight and can be self-hosted — sidestepping both cost and some security concerns tied to using someone else’s servers — Elstrom sees real pricing pressure building on U.S. frontier labs. He specifically flagged the timing risk for OpenAI and Anthropic, both of which are reportedly aiming for IPO valuations above a trillion dollars each, arguing that ambition just got harder to justify.
The BYD Comparison
Elstrom drew a direct parallel to BYD and other Chinese industries — solar panels and steel among them — where Chinese companies have repeatedly proven able to match or exceed Western quality at dramatically lower prices. He noted the irony that BYD vehicles themselves aren’t available for purchase in the U.S., raising the question of whether similar restrictions might eventually target Chinese AI models.
Why Banning Chinese Models Could Backfire on American Businesses
“You may end up hurting American businesses that are not the AI leaders.” — Peter Elstrom
Asked about government appetite for barring Chinese AI models from U.S. use, Elstrom cautioned that the leading edge of affordable open-source AI right now is dominated by Chinese labs — DeepSeek, Moonshot, MiniMax, and others — meaning restrictions aimed at protecting American frontier labs could instead punish the broader universe of U.S. companies that rely on cheap, capable open-source tools to adopt AI in the first place.
Bessemer’s Byron Deeter on Why China AI Fears Are “Totally Misplaced” | CNBC Squawk on the Street
Why He Thinks China AI Fears Are “Totally Misplaced”
“This notion that Chinese models are suddenly going to crater demand and change these cost curves, I think is totally misplaced.” — Byron Deeter
With semiconductor ETF SOXX posting its worst weekly loss since September 2024 on China AI competition fears, Deeter argued the sell-off is really just profit-taking after an “epic run-up” in hardware stocks, not evidence that Chinese models are about to upend the market. He called Kimi a fine but memory-intensive, relatively slow model — solid, not category-redefining — and drew a parallel to the open-source story in cloud computing: open-weight models will take some demand and pricing pressure off parts of the market, but won’t displace frontier models for the workloads that matter most.
Three Reasons Corporate America Won’t Switch to Chinese Models
Deeter laid out his case in three parts. First, performance: once you account for full price-performance, including memory costs, and the near-certainty that these Chinese models were trained in part through distillation attacks on U.S. models, the gap isn’t as favorable as headline pricing suggests. Second, trust: he argued Fortune 1000 companies won’t put business intelligence, customer data, or employee records on Chinese AI systems any more than they’d want their most sensitive data flowing through TikTok, given how central data security and vendor trust are to enterprise relationships. Third, he expects U.S. regulation to eventually address predatory pricing, distillation attacks, and IP theft to protect the domestic AI industry’s lead.
Where the Real Pricing Pressure Will Show Up
Even without a wholesale corporate migration to Chinese models, Deeter acknowledged real pricing pressure is coming — but concentrated in low-end use cases, early-stage startups, and experimentation budgets rather than mission-critical enterprise workloads. He expects frontier labs, currently supply-constrained, to bring prices down on their own as more data-center capacity comes online, and also flagged emerging sovereign and allied-nation open-weight models — from Europe, India, and Canada — as a further competitive factor likely to carve out their own share of the ecosystem.
George Chen, The Asia Group Partner, on Whether Kimi K3 Proves US Chip Export Controls Have Failed | Bloomberg: The Asia Trade
A “Perfect Plan B”: Reigniting the Open vs. Closed Debate
“The rest of the world will also look at China as, okay, now we have a more affordable solution — it’s like a perfect plan B.” — George Chen
Chen said Kimi K3 caught both Wall Street and Silicon Valley off guard, reigniting the open-source-versus-closed-model debate globally. If an open-weight Chinese model can be produced at dramatically lower cost, he argued, the rest of the world now has a credible, affordable alternative to relying entirely on U.S. AI companies — a direct challenge to American dominance in the space.
Does This Prove Export Controls Aren’t Working?
Chen raised a pointed question for Washington: Kimi K3 is a 2.8 trillion-parameter model that reportedly relies on older H100 chips and domestic Chinese hardware rather than the restricted H200, meaning Moonshot achieved frontier-adjacent performance without the chips export controls were designed to withhold. He noted Alibaba’s competing model faces the same constraint, and expects this to force a serious rethink of export-control strategy on both sides of the U.S.-China AI dialogue — with Chinese labs candidly acknowledging they still trail Claude and other leading Western models for now, but with a five-year horizon in mind, and Huawei pursuing an integrated-systems strategy to match H200-level capacity without matching it chip-for-chip.
The US-China AI Race Is Now About Picking Allies
Chen read Xi Jinping’s announcement of a new international AI organization as a direct answer to U.S.-led coalitions, arguing the AI competition has moved beyond a purely bilateral U.S.-China contest into a broader contest for global partners and allies.
China’s Real Edge: Software, Not Hardware
Chen described a clear bifurcation: the U.S. leading in hardware, China making faster progress in software and applications — pointing specifically to AI video generation, where he said Chinese tools (citing ByteDance’s short-video AI production tools) already dominate globally, calling it “already a headache” for the U.S. the way TikTok has been. He also noted heavy state backing across China’s AI-adjacent industries — chips, robotics, and shipbuilding — describing an “at all costs” national push to develop homegrown chips matching H200-class performance within five years, with a small recently approved quota of H200 imports into China serving as more of a diplomatic gesture than a meaningful supply fix.
Wall Street’s Real Worry: The Rise of the Hybrid Model
“[OpenAI and Anthropic] consider themselves the king of the world, but now the position is being challenged.” — George Chen
Chen said the more immediate business risk for OpenAI and Anthropic isn’t wholesale customer defection, but a growing hybrid-usage pattern: companies increasingly routing basic or lower-stakes tasks to cheaper Chinese models while reserving U.S. frontier models for their most demanding work — eroding the assumption that premium labs capture all of the value regardless of task complexity. He also expects Washington’s export-control policy to tighten further in response, making it harder for Nvidia to win approval to sell more advanced chips into China.
Goldman Sachs’ Ronald Keung on How Chinese AI Labs Are Learning to Monetize Open-Weight Models | CNBC International
Chinese Models Are Hitting a “Critical Point” for Coding
Keung said Chinese large language models have reached a critical threshold of capability for global proliferation, particularly in coding, where a self-reinforcing data loop is accelerating improvement: developers running coding workloads through official API channels generate real usage data that feeds back into training the next model iteration. He linked this directly to the jump in scale from roughly 1 trillion parameters to Kimi K3’s 2.8 trillion, driven in part by strong API demand specifically for coding use cases.
From Free Open-Weight to a SaaS-Like Licensing Model
Keung described a shift underway from pure open-source releases toward “open weight” models and now what he called community licensing — MiniMax being an early example, where the model is free for research use but requires commercial users to share revenue. He framed this as Chinese AI labs evolving toward a structure that looks more like traditional SaaS economics, a response to tight compute and chip supply that makes giving away commercial usage for free increasingly unsustainable.
Blending API Revenue With Third-Party Hosting Revenue Share
Beyond revenue from their own API channels — where labs control the compute and capture the margin directly — Keung expects a second, blended revenue stream to emerge as these models get hosted on third-party platforms like AWS Bedrock and other global hyperscalers, with revenue-sharing arrangements replacing the current setup where labs get essentially nothing when their open-weight models run on someone else’s infrastructure.
Three Risks to China’s Global AI Proliferation Strategy
Keung flagged three specific risks to the current wave of strong token-consumption demand for Chinese models internationally. First, Western data-security concerns, which he compared to TikTok’s trajectory — rapid adoption first, followed by a second wave of regulatory scrutiny. Second, as Chinese models approach or reach the frontier, there’s a real question of whether the most advanced ones stay fully open to any country or business, even if the broader base of Chinese models remains open source. Third, continued access to high-end compute remains essential for Chinese labs to keep iterating at their current pace, regardless of how mature their software and licensing strategy becomes.
Kai-Fu Lee, 01.AI Chairman & CEO, on Why He’s Building Applications on Kimi K3 Instead of His Own Model, and China’s AI Monetization Playbook | Bloomberg Tech
Why 01.AI Stopped Competing on Foundational Models
“The models... three years ago, were like an average human, but now they’re smarter than all humans. So it’s such a pity if these models are only applied to simple tasks.” — Kai-Fu Lee
Lee said Kimi K3 is an excellent model that his company is already testing and building on — a deliberate strategic choice rather than a competitive threat. About a year ago, 01.AI pivoted away from fine-tuning and expanding its own foundational model in favor of building applications on top of the best available third-party models, a shift he said sharply cut expenses and put the company on a path to profitability, targeted for sometime next year.
Boss AI: Giving CEOs a “God’s Eye View”
Lee described 01.AI’s flagship product, Boss AI, as designed to give a company’s CEO full visibility across all data to drive decisions that move the financial statement — a contrast, he argued, to most enterprise AI deployments that amount to “sugar coated demonstrations” with no substantial business impact. Bookings have grown three to four times year over year, and Lee said the company’s real constraint is hiring enough multilingual consultants and staff to service demand across Asia, South America, and the Middle East, where growth is running in the triple digits.
Export Controls Have “Simply Not Worked”
“Export control was an imaginative but unsuccessful effort to contain China. It has failed.” — Kai-Fu Lee
Lee dismissed the idea that GPU export restrictions are meaningfully constraining Chinese AI development, arguing any domestic GPU shortage stems from booming global demand, business frugality, or supply limits — not from U.S. export policy specifically.
The US-China Split: Profit vs. Market Share
Lee expects the U.S. to capture more profit in foundational models and enterprise applications over the next three to five years, while China wins on market share thanks to open-source distribution and enterprises’ preference for on-premise deployment. He predicted Chinese consumer-application companies like Tencent and ByteDance will out-innovate their American counterparts, and that China maintains a hardware lead — while cautioning that low Chinese AI pricing may not be sustainable long-term, drawing a parallel to early internet and social media products that were free before monetization caught up.
Positioning Against Palantir
Lee framed 01.AI and Palantir as the only two products capable of running genuine AI transformation with results in two to three months, but said the two firms effectively avoid competing head-to-head: Palantir focuses on U.S. and allied defense and military work, while 01.AI targets Asia and Belt and Road countries Palantir largely avoids — a dynamic he expects to harden into a durable, defensible moat via first-mover advantage. He’s targeting an IPO as soon as 2027, contingent on completing the filing process.
AI Infrastructure & Power Deep Dive
Baird’s Ben Kallo on GE Vernova’s Booked-Out Backlog, the Case for a SpaceX-Tesla Merger, and Why Labor (Not Chips) Is the Next Bottleneck | The Real Eisman Playbook
America’s Grid Is Short by Up to 350 Gigawatts
“The range you could drive a truck through... could be 100 gigawatts of new capacity needed by 2035... at 350, you’re more than doubling all the electricity generation in the United States.” — Ben Kallo
Kallo, Baird’s sustainable energy and mobility analyst, told Steve Eisman that estimates for new U.S. grid capacity needed by 2035 range so widely — from 100 to 350 gigawatts against roughly 150 gigawatts of total generation today — that the uncertainty itself signals how unprepared the grid is. The country is currently adding about 30 gigawatts of base-load capacity a year, a pace Kallo said is already straining under pressures well beyond data centers: onshoring manufacturing, aging coal-plant retirements, and electrification of homes and vehicles. Curtailment days — utilities asking commercial tenants to cut air conditioning during peak demand — are becoming more frequent even before gigawatt-scale AI data centers have fully come online.
GE Vernova: Booked Out Through 2030, With Nuclear as a Late-Decade Kicker
“They’ve been able to book out until 2030 with price... something that’s booked today is not actually going to get into utility plant till 2030, 2031.” — Ben Kallo
Kallo’s top pick in the group is GE Vernova, whose gas-turbine business — the equipment utilities and data-center developers need for fast power — is sold out years in advance, giving the company unusually long earnings visibility into the mid-2030s once its service tail is included. He also flagged a fast-growing electrification segment selling transformers and grid equipment (cross-sold alongside turbines, including a recent Chevron deal) and a nascent small-modular-reactor nuclear business built with Hitachi, with a first Ontario reactor targeted for 2035 — a program Kallo said is more likely to move the earnings needle in the late 2030s than sooner, given skepticism around the industry’s aggressive SMR timelines.
Making the Case for a SpaceX-Tesla Merger
Kallo argued a SpaceX acquisition of Tesla is increasingly likely, potentially within 18 months, driven by Elon Musk’s desire for a 25% AI-controlling stake he can’t easily get inside Tesla alone, plus the capital-raising and single-board efficiency benefits of combining the companies’ overlapping projects — including a joint 100-gigawatt solar manufacturing venture and a joint chip effort (”Terrafab”) aimed at competing with Nvidia. He was clear-eyed that Tesla’s core EV business alone wouldn’t justify its valuation, with the bull case resting instead on autonomous robotaxis (which Kallo expects to scale first in Texas) and a fast-growing energy-storage (”megapack”) business that already contributes roughly 20% of Tesla’s operating income.
Solar: Bullish on the Grid Story, Cautious on the Stocks
Kallo is not recommending First Solar despite the sector’s tailwinds, citing execution risk from shifting production to the U.S. and unresolved Section 232 tariff decisions on Chinese polysilicon that leave the company unable to price contracts — though he estimated $100 of upside in the stock if tariff clarity comes through favorably. He’s more constructive on Rivian than Lucid heading into the R2 launch, and flagged skilled labor — electricians and construction crews — as the binding constraint on the entire power buildout going forward, more so than chips or capital.
AMD’s Forrest Norrod on Helios, AMD’s First Rack-Scale AI System, and the Bet That It Can Take Real Share From Nvidia | CNBC
A $5 Million, 7,000-Pound Answer to Nvidia’s Vera Rubin
AMD unveiled Helios, its first rack-scale AI system — 72 GPUs and 18 CPUs packed into a single rack, built to compete directly with Nvidia’s Grace Blackwell and newer Vera Rubin systems. Meta, OpenAI, Oracle, Microsoft, and India’s TCS have all signed on to deploy Helios this year, with Meta alone committing to up to six gigawatts of AMD GPU capacity. AMD data center lead Forrest Norrod said the system is built entirely from AMD’s own GPUs, CPUs, networking, and software stack, and that early performance and efficiency data give it real advantages in inference and memory bandwidth versus Nvidia’s offering.
Still a Single-Digit Player, But Aiming Bigger
“There’s a serious case in which AMD does great and can get to 20 and 25%. And by the way, this is hundreds of billions of dollars of revenue. You don’t play in markets to be a single-digit player.” — Forrest Norrod
Nvidia still controls more than 95% of the data-center GPU market, but Norrod said AMD’s aspiration is real share gains, not just participation. Helios is priced by third-party estimates at $5 to $5.5 million per system versus $3.5 to $4 million for Vera Rubin, with AMD arguing its pitch is lowest total cost of ownership and cost-per-token rather than sticker price — an argument Norrod said is gaining traction as customers shift focus from maximizing token generation to practical, cost-justified token utilization.
CUDA Still Wins on Ecosystem, But ROCm Is Catching Up on Openness
AMD acknowledged Nvidia’s CUDA software remains far more ubiquitous and gives Nvidia a clear ecosystem edge, but is positioning its open-source ROCm stack — backed by acquisitions and support for PyTorch, vLLM, and SGLang — as the differentiated alternative to Nvidia’s closed, proprietary approach to owning the “AI factory” end to end. Helios itself is built on more open standards than Nvidia’s rack systems, per AMD.
The Real Constraints: TSMC Capacity, Packaging, and HBM Memory
Helios’ MI455 GPUs are made on TSMC’s most advanced two-nanometer node, and AMD said CEO Lisa Su’s early, repeated trips to Taiwan helped secure wafer capacity even as Nvidia has reserved the bulk of TSMC’s advanced CoWoS packaging. AMD has since committed $10 billion to other Taiwanese packaging partners like ASE and says it has locked in relationships with all three major HBM memory suppliers — necessary given each MI455 GPU uses up to 432GB of HBM memory amid a global memory shortage. About 22.5% of AMD’s 2025 revenue came from China, and the company said it would sell Helios there if regulations and customer demand allow.
TSMC CFO on the $100 Billion Arizona Expansion, Multi-Year Chip Demand, and Why It’s Not Leaving “Any Food on the Table” | CNBC
Another $100 Billion for Arizona, on Top of the Original $100 Billion
TSMC’s CFO confirmed the company is increasing its Arizona investment by another $100 billion, on top of the $100 billion already committed earlier this year, citing continued strong multi-year demand from U.S. customers and strong support from federal, state, and city government. The Arizona site’s Phase One (N4 technology) is already running at Taiwan-level quality; Phase Two (N3) is moving in tools now and targeted for online in the second half of 2027; Phase Three (N2) construction has begun; and Phase Four, including the first U.S. advanced-packaging fab, is in early preparation.
Building Overseas Costs Four to Five Times More Than Taiwan
“It takes four to five times more in constructing a fab in the U.S. compared to in Taiwan.” — TSMC CFO
That cost gap is driving margin dilution from overseas expansion — currently 2-3 percentage points a year, widening to 3-4 points as more U.S. phases come online between 2024 and 2029. The CFO said profitability improves phase by phase but likely won’t reach Taiwan-level margins, framing the tradeoff as a deliberate one: the added investment strengthens the U.S. semiconductor supply chain and creates high-tech jobs, in exchange for government support via CHIPS Act investment tax credits and local infrastructure help with water, power, and permitting.
No Food Left on the Table
Total capital expenditure over the next three years will be significantly higher than the prior three, with Taiwan’s own 2025 capex already raised from $56 billion to $60-64 billion as the company builds out 13 fabs domestically alongside its expanding U.S. footprint. Asked directly about the scale of demand driving all of this, the CFO was blunt: TSMC does not plan to leave any capacity on the table for competitors to capture.
Macro & Credit Deep Dive
Mispriced Assets’ Nick Nemeth on Why Private Credit Is “1929, Not 2008,” and the $10 Trillion Insurance Balance Sheet at the Center of It | Monetary Matters
“The Smart Money Is the Subprime This Time”
“We’re talking about subprime taking down the economy at $1.2 trillion. There’s a trillion dollars of private credit... but really the mass of the crisis is insurance, where there’s a $10 trillion balance sheet.” — Nick Nemeth
Nemeth, who writes the Mispriced Assets newsletter, laid out a bearish thesis on private credit he’s been building for roughly three years: direct lenders are running leverage around seven times adjusted EBITDA — earnings he calls “doubly fake” once private equity firms layer in optimistic synergy assumptions that S&P data shows miss actual results by 25% to 50% of the time. He stressed he isn’t a permabear and has made money on long positions this year, but sees this cycle’s endgame concentrated specifically in insurance company balance sheets rather than banks.
Insurance, Not Banks, Is the Real Risk Transmission Point
Private equity firms — Apollo with Athene being the pioneering example — have increasingly bought or created insurance companies to access “permanent capital” for their credit funds. Unlike bank deposits, insurance policyholders face steep penalties for early withdrawal (surrender charges as high as 7% in year one), which Nemeth argues masks how thinly capitalized some insurers really are: he estimates some are levered 90 to 100 times versus the roughly 30 times leverage Lehman Brothers carried in 2008, with some carrying negative equity that regulators have exempted from standard mark-to-market rules.
Six Layers of Leverage, and Credit Ratings He Calls Compromised
“These insurers’ balance sheets are levered up in many cases more than Lehman Brothers was in 2008. We’re talking 90 times, 100 times in some cases.” — Nick Nemeth
Nemeth described a stacked structure of leverage — from underlying portfolio companies, up through BDC-level debt, fund-level leverage from sovereign wealth investors (who he says can achieve up to 20x leverage via repo agreements), general-partner financing, and finally the insurance balance sheets themselves. He was sharply critical of the ratings agencies underpinning the whole structure, arguing firms like Egan Jones and Kroll — which rate the underlying loans inside CLOs — have strong incentives to issue favorable ratings to win business, echoing dynamics he compared directly to the subprime mortgage ratings failures of 2008.
Why He Sees 1929, Not 2008
Nemeth’s central distinction: 2008 primarily hurt working-class borrowers holding subprime mortgages, while this cycle’s damage, in his view, is concentrated among white-collar workers and institutional capital funneled into private credit and PE-controlled insurance vehicles — a dynamic he compares more to the wealthy investors wiped out in 1929. He flagged rising private-credit default rates (around 6.3% in direct lending versus roughly 6% in 2008, though 2008 levels started far lower) and said a sustained run of elevated defaults — roughly six quarters by his math — would trigger forced downgrades of the CLOs sitting on insurance balance sheets, forcing capital calls the industry isn’t prepared for.
Grading the Public Alternative Asset Managers
Asked to rank the publicly traded credit managers, Nemeth called Ares “the biggest gap of brand-name aura to reality,” said Blackstone’s business is largely marketing (with CEO Jon Gray coming across as “a narrator” rather than an investor), and named Blue Owl as the most underrated despite what he called its “worst PR I’ve ever seen.” He views Apollo as a genuinely skilled underwriter — “running risk at an F-35 level,” in his words — but argued that reputation, not necessarily superior fundamentals, is what’s currently driving capital toward winners like Apollo and Ares and away from Blackstone.
Palomar Capital’s Patrick Boyle on SpaceX’s “Priced for Science Fiction” IPO and the Return of Dot-Com-Era Financial Engineering | Hidden Forces
A $135 IPO Price Elon Musk Set Himself
Boyle, founder of the now-sold Palomar Capital Management and a finance professor at King’s College London, dissected SpaceX’s IPO, which priced at roughly 100 times sales — compared to Google’s eight times sales at its IPO (growing over 200% a year) and Facebook’s roughly 10 times sales, which was itself considered aggressive at the time. Unlike a typical IPO process where investment banks build a book and set a market-clearing price, Boyle said Musk simply set the $135 share price himself — a figure that made him the “world’s first trillionaire” on paper — and took it to market with no shareholder voting rights attached.
Growing at 15% a Year Doesn’t Justify the Multiple
“If you’re paying a multiple for growth, where’s the growth?” — Patrick Boyle
Boyle noted SpaceX is reportedly burning about $5 billion a quarter, with losses expected to grow, and that unlike Google and Facebook — asset-light software businesses — SpaceX faces enormous capital expenditure needs across data centers, Nvidia chip purchases, and its core rocket and satellite business. He was skeptical of the “enterprise AI company” framing used to justify the valuation, pointing out xAI’s Grok holds roughly 3.5% AI market share and that virtually no enterprise customers he’s aware of are choosing it over Anthropic.
Engineered Index Inclusion and a Shrinking Float
Boyle detailed how SpaceX was fast-tracked into the Nasdaq-100 and Russell 1000 within days of its IPO — a highly unusual departure from typical “seasoning” periods — which he said effectively guaranteed a wave of forced buying from index funds and created a trade Wall Street banks and funds like Millennium profited from directly. He also flagged a shrinking public float (potentially under 1% of shares genuinely trading) and a wave of newly published bank price targets — some as high as $1,000 a share — that he compared to the analyst conflicts-of-interest scandals of the dot-com bubble, noting the relevant Sarbanes-Oxley-era “truth in analysis” rule was repealed roughly six months before this piece.
Institutions Weakening, Individuals Gaining Power
Boyle connected the SpaceX story to a broader theme: eroding trust in institutions since the 2008 financial crisis bailouts has empowered individuals — Musk, Trump, and others — who are skilled at commanding social-media audiences, a dynamic he said resembles the “strong individuals, weak institutions” pattern more typical of emerging markets. He also discussed the explosion of prediction markets like Kalshi and Polymarket as largely disguised sports betting fueled by “financial nihilism” among younger, economically priced-out investors, and flagged a UAE investment into Trump’s World Liberty Financial stablecoin as a case study in how the two trends — weakening institutions and unconventional capital flows — are intersecting.
Bloomberg’s Scott Carpenter on How Wall Street Is Using Insurance Wraps to Unfreeze Private Credit Cash | Bloomberg
Turning Fund Equity Into Rated Debt
Carpenter, who covers structured finance for Bloomberg, explained a deal UBS has proposed — following one partners Group already executed — that takes shares of private credit and private equity funds, pools them into special vehicles, and issues bonds against them. Layered on top is an insurance policy from Nationwide covering the majority of the transaction against losses, a “wrap” that pushes the deal’s rating up to A2 even though comparable unwrapped fund-finance deals typically land at triple-B or single-B.
Why It’s Happening Now: Funds Need Cash, Insurers Have It
“You’ve got the funds that need cash. On the other hand, you’ve got the insurance companies that have cash. You need to connect them. And that’s the financial engineering that’s happening.” — Scott Carpenter
Carpenter said private credit and private equity funds have been slow to sell existing investments and need new sources of dollars, while insurers — flush with cash from a wave of annuity sales tied to retiring Baby Boomers — need investment-grade-rated places to park that money. The insurance wrap is the mechanism connecting the two, and Carpenter noted it’s a still-rare structure in fund finance, one sophisticated potential buyers immediately recognized as unusual when they reviewed the UBS presentation.
The New Risk: Linking Two Previously Separate Parts of the Financial System
A colleague on the panel flagged the structural risk directly: if Nationwide itself were downgraded for reasons entirely unrelated to private credit, every deal it has wrapped would get downgraded in turn — creating a new contagion channel between the insurance sector and private markets that didn’t previously exist. Carpenter added that rating-agency conflicts of interest remain an open question in these deals, though ratings divisions are nominally insulated from the business-development side. As of the report, the UBS deal was still in progress, with no confirmed timeline to close.
Mohamed El-Erian on Why “The Worst of the Inflation Is Behind Us,” and Why AI Infrastructure Spending May Have No End Point | CNBC
No Rate Hikes Coming, and Inflation’s Worst Days Are Past
“I don’t think we’re going to get any rate hikes. I think the worst of the inflation is behind us.” — Mohamed El-Erian
El-Erian said tariff-driven inflation and most oil-related inflation pressure have largely worked through the system, leaving him unconcerned about the case some have made for further Fed rate hikes. He flagged AI-related inflation as a different category entirely — one he said he can live with because he believes it reflects genuine, coming productivity gains rather than simple cost-push pressure, while still keeping a close eye on regular gasoline and diesel prices as his real-time inflation gauge.
AI Buildout: An Overbuild Is Coming, Just Not Soon
El-Erian said a capacity overbuild is likely within three to four years, consistent with the pattern of every major technology buildout in history overshooting demand in its early phases — pointing to the late-1990s fiber-optic buildout as a direct parallel. But he cautioned this cycle could run much longer than fiber’s did, citing conversations with tech executives who describe AI as lacking a definable endpoint altogether, unlike fiber’s finite “lay the track and stop” buildout.
Why AI Might Not Have an End Point at All
Citing Google’s James Manyika, El-Erian described AI as “the inventor of inventions” — a general-purpose technology capable of recursive self-improvement, meaning it continuously creates new applications and demand for itself rather than reaching a natural stopping point the way earlier infrastructure buildouts did. On markets more broadly, he described a striking and “unstable equilibrium” between genuinely turbulent geopolitical news (fighting in Iran, military casualties) and calm asset prices, and said the explosive growth in prediction-market betting reflects lowered barriers to entry meeting genuine “lottery ticket” psychology — a dynamic he said has clear negative social implications even as he views broader market access as generally a good thing.
Industrials & Aerospace Deep Dive
Honeywell CEO Vimal Kapur on Why He Split Up a 100-Year-Old Conglomerate, and Where AI Actually Moves the Needle in Industrial Automation | Bloomberg Leaders
Breaking Up Honeywell Wasn’t About the Activist — It Was Already the Plan
“I always felt that we need to simplify... each company can stand on its own feet with these growth vectors.” — Vimal Kapur
Kapur, who rose through Honeywell’s India joint venture before eventually running the whole company, said the decision to split Honeywell into three standalone businesses — aerospace, automation (which he now leads as Honeywell Technologies), and specialty materials — predated activist investor Elliott Management’s public $5 billion stake and breakup demand in late 2024. He said Elliott’s letter arrived the same day as everyone else learned about it, with no advance call, but that Honeywell was already far enough along in its own strategic thinking that there was little disagreement over the destination, just some overlap on timing. His advice to other CEOs facing an activist: have a clear, fact-based conviction on strategy before the letter ever arrives, because “if you are all over the place, the opposition can win.”
Two Forces Converged: Aerospace Growth and the AI Moment
Kapur said two things happened simultaneously when he became CEO in 2023: the aerospace industry entered a strong growth cycle, and ChatGPT put AI at the center of every corporate conversation. Both convinced him that automation and aerospace deserved to be run as independent, focused companies rather than compartments of a sprawling industrial conglomerate — a structure he said made sense for the prior 20 years of shareholder value creation but was increasingly an obstacle to further top-line growth.
Where AI Actually Helps: Productivity Now, Transformation Later
“Transformational will occur if we redraw our work right, in which we can use agents as part of our workflow, and that requires us to reimagine the work by itself.” — Vimal Kapur
Kapur said Honeywell’s AI impact so far has been concentrated in productivity gains, particularly in software development and testing, rather than genuine transformation of how work gets done — that shift, he argued, requires redesigning workflows around AI agents rather than simply layering AI onto existing processes. He sees the bigger opportunity in serving customers: Honeywell’s automation business has collected operational data from buildings, airports, refineries, and warehouses for years, primarily for control purposes, and the emerging opportunity is turning that data into tools that let a five-year employee perform at a fifteen-year veteran’s level — critical given customers’ worsening shortage of skilled operators as their workforces retire.
Boeing CEO Kelly Ortberg on the 737 Ramp-Up, “Turning the Corner” After Years of Crisis, and Why the Next-Generation Narrowbody Is Still a Decade Away | Bloomberg at Farnborough
“We Haven’t Quite Finished Turning the Corner”
Nearly two years into the job, Ortberg described Boeing as still mid-recovery but making real progress: the company is ramping 737 production toward a 47-per-month rate with aspirations to reach 63, and 787 production toward 10 per month this year and higher next. He credited the Trump administration with helping Boeing’s sales campaigns and trade positioning but was clear that rebuilding trust with regulators and stakeholders has been primarily Boeing’s own execution work, not government support.
44,000 New Aircraft Needed Over 20 Years, Despite Oil at $90
Ortberg called the aviation market “super resilient” despite $90 oil and Middle East conflict, citing Boeing’s own newly published 20-year forecast for 44,000 new aircraft and noting that aircraft ordering cycles are long enough that near-term macro volatility doesn’t meaningfully move demand. He said AI is increasingly embedded across Boeing’s design processes, back-office operations, and eventually its products and services, and expects the more meaningful automation “step function” to arrive not through retrofitting existing aircraft but through the next-generation narrowbody, whose design can be built around automation from scratch.
Next-Gen 737 Replacement: Not Ready, Market Not Ready Either
“The market is not quite ready for the new airplane... the customers are telling us, let’s focus on the existing product.” — Kelly Ortberg
Ortberg pushed back gently on suggestions the 737 replacement could launch early next decade, saying Boeing itself needs a couple more years to improve cash flow and pay down debt before it’s financially ready, and that customers are prioritizing durability improvements to the current fleet over a clean-sheet replacement. He placed the likely timeline toward the back end of next decade, and said Boeing hasn’t yet decided between GE’s open-fan engine concept and more conventional technology for that future aircraft.
Europe’s Rearmament Will Need American Partners
On European defense, Ortberg said Europe cannot fully rearm without partnering with prime contractors like Boeing, citing existing joint ventures such as the MQ-28 with Germany, and expects the relationship to remain partnership-based rather than consolidating into M&A in the near term. He noted defense and civil aviation increasingly share the same tier-two and tier-three supply chain, meaning both markets ramping simultaneously will strain the same suppliers.
IMAX CEO Rich Gelfond on Why “The Odyssey” Broke IMAX’s Box Office Record, and a Blimped Camera Built Just for Christopher Nolan | Bloomberg Talks
A Record Opening, Nearly 50% Above Oppenheimer
“We did $52 million on a like-for-like basis, the highest we’ve ever had... almost 50% above Oppenheimer.” — Rich Gelfond
Gelfond said Christopher Nolan’s The Odyssey delivered IMAX’s best opening ever, driven partly by scarcity: only a subset of IMAX’s roughly 2,000 global screens are true film-based theaters capable of showing the movie as Nolan shot it, and film prints cost $50,000 each per theater — economics that make it impossible to put IMAX film in every market despite viral demand for even 2 a.m. and 3 a.m. screenings. IMAX added shows in 42 theaters between midnight and 3 a.m. to meet demand, and shows at London’s BFI are already sold out roughly two months out.
A New Camera Built Specifically for This Film
Gelfond said IMAX engineered an entirely new camera system for The Odyssey, including a “blimp” housing that eliminates the traditionally loud film-spool noise of IMAX cameras — opening up close-up shots that weren’t previously possible with the format. He expects the success of The Odyssey to encourage more directors to shoot primarily or entirely on IMAX film cameras, projecting five or more film-format releases in 2027, up from limited prior usage.
A $1.4 Billion Box Office Target Built on a Full 2026 Slate
IMAX is targeting a record $1.4 billion global box office in 2026, backstopped by The Odyssey, Dune 3 (already sold out around its Christmas release), Spider-Man, Avengers, and a new Tom Cruise film in the back half of the year. Asked directly about May reports that IMAX was exploring a sale to entertainment companies, Gelfond declined to comment beyond IMAX’s standing policy of confirming deals only once signed.
Airbus CEO Guillaume Faury on a 9,200-Aircraft Backlog, Engine Supply Struggles, and Racing Boeing to the Next-Generation Narrowbody | Bloomberg at Farnborough
A Backlog Above 9,200 Aircraft, With Oil Prices Actually Helping
Faury described demand as very strong despite a worsening Gulf war and oil near $90 a barrel, arguing higher fuel prices are actually boosting demand for Airbus’s fuel-efficient aircraft even as broader macro conditions stay unpredictable. Airbus delivered 15% more aircraft in the first half of 2026 than the same period in 2025 and is targeting around 870 deliveries for the full year, consistent with reaching a production rate of 75 narrowbodies per month.
Pratt & Whitney Issues Easing, But Not Uniform Across the Supply Chain
“Let’s not think that the Pratt & Whitney situation means that we have supply chain issues everywhere. That’s not the case.” — Guillaume Faury
Faury said Airbus has adjusted its production planning around Pratt & Whitney’s geared turbofan engine delivery delays but that the situation is stabilizing, while CFM and Rolls-Royce engine programs remain on track. He confirmed Airbus’s target of launching its next-generation single-aisle aircraft around 2030, for entry into service in the second half of the 2030s, delivering roughly 25% better fuel burn — with the choice between a traditional geared turbofan and an open-rotor engine architecture still undecided pending further risk assessment.
On European Defense: Still Committed to FCAS, Rearmament Needs America
Faury said Airbus remains part of Europe’s Future Combat Air System (FCAS) program alongside France and Germany, even as the originally structured sixth-generation fighter component of that program has been restructured, and downplayed any near-term move to join the rival GCAP program with the UK, Italy, and Japan. He was direct that Europe cannot fully rearm without the United States as an ally, even as it pushes to become more self-sufficient within NATO.
Airbus CEO Guillaume Faury on Engine Durability Improvements, China Order Flow, and Why the Next Airbus Might Look Similar From a Distance | CNBC at Farnborough
Engine Durability, Not Just Production Speed, Was the Real Problem
Faury said the aviation industry’s two central engine challenges — ramp-up speed and time-on-wing durability — have both improved meaningfully, with Pratt & Whitney’s advanced geared turbofan showing better durability and a declining number of aircraft grounded (AOGs) due to engine issues. He said Airbus has adjusted its 2026 production planning to account for the disruption but now has “what we need” to hit full-year delivery guidance.
Demand Is Broadly Distributed Globally, Including Fresh China Orders
Asked whether China remains the biggest long-term growth region, Faury said demand for new aircraft and fleet renewal is fairly evenly distributed across geographies rather than concentrated in any single region, pointing to recently announced new orders from Chinese airlines as one data point among many. He noted total industry demand across the next 20 years remains very strong for both single-aisle and widebody aircraft, consistent with Boeing’s own updated forecast.
The Next Airbus Will Look Similar From a Distance, Very Different Up Close
On whether Airbus’s next-generation single-aisle could take a radically different shape — such as a blended-wing-body design — Faury said the general architecture will likely look broadly familiar from a distance, but major differences will emerge on closer inspection, with key design trade-offs still to be finalized in the coming years ahead of the targeted 2030 program launch.
AI Founders & Executives Deep Dive
Anthropic’s Boris Cherny on How Claude Code Went From Side Project to 90% of Anthropic’s Own Codebase, and Why “Moats Matter Less” in the Age of AI Coding | Odd Lots
Claude Code Wasn’t a Product Bet — It Was a Safety Research Tool
Cherny, who leads Claude Code at Anthropic, said the product’s origins trace back to Anthropic’s core safety mission rather than a deliberate business strategy: understanding whether a model is genuinely safe requires more than lab-based interpretability work, it requires putting the model into real-world use and watching how people actually use it. Because models interact with the world primarily through code, coding became the natural proving ground — and Claude Code emerged from that logic, not from an initial plan to build a coding business. It has since become, in his words, “a big contributor” to Anthropic’s overall revenue.
90% of Anthropic’s Own Code Is Now Written by Claude Code
“100% of my code has been written by Claude Code since November of last year... across Anthropic, I think the average is something like 90%.” — Boris Cherny
Cherny said nearly all of Anthropic’s internal software — products, infrastructure, and increasingly research code — is now generated by Claude Code itself, a practice the team calls “dogfooding” that runs on the same public API and models available to any customer. He said growth in Claude Code usage has directly tracked major model releases — Opus 4, Opus 4.5, and Opus 4.6 each produced a visible inflection point in adoption — because the harness benefits automatically whenever the underlying model improves.
Guardrails Against Prompt Injection: Alignment, Neural Probes, and Sandboxing
Cherny described a layered defense against prompt injection — where malicious instructions embedded in a webpage or document attempt to hijack the model’s actions — combining model-level alignment training, “neural probes” built through mechanistic interpretability that detect injection attempts inside the model’s internal representations, and a new low-friction permission mode called Auto Mode. He said Anthropic ran an external red-team competition offering $20,000 for researchers who could successfully prompt-inject Claude Code, and no one succeeded, unlike with competing models tested in the same exercise.
A Migration That Took 11 Days Instead of a Year
Cherny cited a real internal example — engineer Jared migrating the Bun JavaScript engine that powers Claude Code from the Zig language to Rust in 11 days, at a cost of roughly $50,000 in compute credits, a project that previously would have required a team of engineers a full year and simply wouldn’t have been greenlit given the cost. He said this kind of large-scale code migration, including legacy COBOL modernization at major banks, is now one of Claude Code’s core strengths.
Why Competitive “Moats” Matter Less, But Not Zero
Asked whether AI coding tools erode traditional business moats like switching costs, Cherny invoked the “Seven Powers” framework to argue that switching-cost-based moats are weakening — since Claude Code can now port a codebase between vendors on request — but that businesses built on multiple combined moats (scale economics, network effects, proprietary resources) remain largely intact. On rival messaging suggesting companies should avoid deep dependence on Anthropic’s models in favor of self-hosted open-source alternatives — an argument he attributed to Microsoft’s CEO and a viral Alex Karp interview — Cherny said Anthropic cannot see individual customer conversations even when debugging bugs, and argued that businesses betting on self-hosted infrastructure will fall behind the continuous intelligence gains only available by staying on the frontier.
Engineering Roles Are Splitting, Not Disappearing
Cherny said the traditional distinctions between engineering, design, and product roles are dissolving as coding becomes accessible to everyone on a team — Claude Code’s own designers and product managers now write code directly — and predicted roles will instead segment into “prototypers,” “builders,” “maintainers,” “scalers,” and “sweepers” who polish finished products. He compared the current moment to a well-known 1996 Harvard Business Review study on why personal computers didn’t immediately boost office productivity: gains only appeared at companies that put the computer at the center of workflows and eliminated the paper process entirely, rather than treating it as a side tool — the same principle he says now applies to Claude adoption inside large companies.
OpenAI Chairman Bret Taylor on Why “Token Efficiency” Matters More Than Open-Weight Pricing, and How Sierra Prices AI on Outcomes, Not Tokens | CNBC
Open-Weight Models Aren’t Necessarily Cheaper to Run
“There’s this thing called token efficiency, and it turns out the frontier models are much, much more token efficient.” — Bret Taylor
Taylor, OpenAI’s chairman and co-founder of AI agent company Sierra, pushed back on the narrative that Chinese open-weight models like Kimi K3 are a clear cost advantage over U.S. frontier models, arguing that even if a model is cheaper to license, it may require more total tokens to complete the same task — meaning true cost-per-outcome often favors frontier labs. He said the two things every CEO he talks to (roughly 100 a month, by his account) actually care about are “tokenomics” — confidence they’re getting value from AI spend — and what he calls “sovereignty,” or maintaining a durable competitive moat in the AI era.
Why U.S. Labs Need to Win on Every Price-Performance Tier
Taylor argued frontier labs need to offer the best available model at every combination of speed, cost, and intelligence level a business might need — from cheap, fast models for routine tasks like transaction screening to maximum-intelligence models for high-stakes work like insurance actuarial analysis — and that compute efficiency, not simply open weights, is what ultimately drives usable cost. He acknowledged genuine uncertainty about whether Chinese models like Kimi K3 were trained through legitimate development or model distillation from Western labs, but said the more important question for buyers is simply which model is cheaper to run in practice, not how it was built.
Sierra’s Model: Pay Per Outcome, Not Per Token
Taylor described Sierra’s approach — AI agents that power customer service for companies like DirecTV and SiriusXM, plus a new “Horizon” product for long-horizon tasks like loan origination and medical preauthorization — as pricing customers per successful outcome rather than per token consumed, which he compared to paying for Gmail by the CPU cycle. He said this approach directly addresses both tokenomics anxiety (customers know exactly what they’re paying for) and sovereignty (Sierra helps customers build a compounding, defensible data asset about their own customer relationships, independent of which underlying model is used). Asked directly whether Sierra’s margins hold up given the token cost of running long AI-driven phone conversations, Taylor confirmed customers pay more than Sierra’s underlying compute cost, aided by Sierra’s ability to amortize R&D on specialized low-latency voice infrastructure across many customers rather than each company building it individually.
Venture Capital Deep Dive
Andreessen Horowitz’s Connor Love on Why American Dynamism Is Becoming a Proxy for the Most Valuable Companies in the World | Bloomberg Tech
From Army Captain in Iraq to a16z’s American Dynamism Team
Love, a former Army captain who served in Iraq before moving into venture capital at Lightspeed and now joining Andreessen Horowitz’s American Dynamism practice, described the connection as deeply personal rather than purely strategic. He credited a chance conversation roughly seven years ago with Catherine Boyle — then at a different firm, now an American Dynamism co-founder — for introducing him to Silicon Valley and Anduril specifically, at a time when he was still deployed and thinking about his next career move.
“You Cannot Have a Defense Industrial Base Without an Industrial Base”
“Technology is saying you can be both important to America and our allies and be some of the most valuable companies in the world.” — Connor Love
Love argued American Dynamism extends well beyond defense technology narrowly defined, encompassing energy, core manufacturing, and space — framing the category as being about the subcomponents and industrial base that underpin next-generation autonomous systems and platforms, not just the headline defense primes like Anduril and Anduril rival Castelion. He contrasted this moment with historically important American industrial companies like Boeing, arguing that technology now allows a company to be simultaneously critical to national interests and among the most valuable businesses in the world — a combination he says wasn’t previously possible at this scale.
The Low-Cost Autonomy Thesis Playing Out in Real Conflicts
Love pointed to Ukraine and the conflict involving Iran as live proof points for a “low-cost autonomy” thesis reshaping defense economics: technology platforms that are roughly ten times cheaper than legacy systems while matching performance and mass fundamentally change deterrence calculus for the U.S. and its allies. He sees this dynamic — cheap, mass-producible autonomous systems outperforming expensive legacy defense platforms — as the underlying substrate driving venture investment across the sector, from missile systems to ground autonomy.
Anduril “Hasn’t Shown Its Last Product”
Asked about a16z’s continued conviction in Anduril, in which Love had previously invested while at Lightspeed, he said the company’s complexity of mission — serving the Department of War and the American warfighter — means there will always be another product in the pipeline, pointing to Anduril’s newly announced partnership with Archer on an autonomous VTOL platform as evidence the opportunity set keeps expanding rather than narrowing.
If you found these summaries helpful, please leave a comment or send us a message. We appreciate your feedback!
