Anduril's Palmer Luckey on Defense AI; Mohamed El-Erian on the AI Funding Gap; Peter Diamandis on Humanoid Robots, AI labs and SpaceX
Plus Niall Ferguson on Trump's Approval Puzzle & Democrats' Leftward Lurch; Patrick Boyle on crypto's 'great rotation' into AI; Caleb Hammer on financial nihilism; Jeff Lutz on Tesla Cybercab rollout
Roudup
July 13th roundup of top investors’ and executives’ appearances on interviews, podcasts, and events, and the topics they discussed.
Patrick Boyle on SpaceX, Crypto’s “Great Rotation,” and Housing | ProfG Markets
SpaceX just joined the NASDAQ 100 with near-unanimous buy ratings implying valuations up to $10 trillion, yet the stock is down 34% from its peak — and every underwriting bank pushing those targets stands to earn billions in fees, echoing the dot-com analyst scandals, especially now that the SEC just quietly killed the rule meant to prevent that conflict. Meanwhile, crypto has cratered as speculative money chases AI stocks and leveraged ETFs instead, fueled by “financial nihilism” among young investors who’ve given up on traditional paths to wealth like housing, whose relentless price rises turn out to be less an accident than a policy choice protecting homeowners over buyers.
Read the digest | Watch on YouTube
Caleb Hammer on Financial Nihilism, Debt Shame, and the Trillionaire Debate | Modern Wisdom
Caleb Hammer, creator of "Financial Audit," argues the highest earners are often the worst with money because higher incomes just unlock bigger debt, and that shame around discussing finances (not the debt itself) is what actually traps people. The conversation ranges from why Gen Z is drowning in buy-now-pay-later debt amid "financial nihilism," to why the UK is "great to be poor in, terrible to be rich in" while America is the reverse, to the widening gender wars among young people and the looming Social Security shortfall.
Read the digest | Watch on YouTube
Ed Zitron, Host of Better Offline, on the AI Revenue Mirage, OpenAI and Anthropic's IPO Rush, Why the Dot-Com Comparison Doesn't Hold, and the Capex Signal That Would End the AI Trade | Investor’s Business Daily
Featuring Ed Zitron, CEO of EZPR (Easy Pheasy Research) and host of the Better Offline podcast and Where’s Your Ed At newsletter, on: why OpenAI and Anthropic together make up 89% of AI industry revenue, the accounting shenanigans hiding inference losses, why the AI buildout doesn’t resemble the dot-com bubble, the looming OpenAI and Anthropic IPOs, and the one capex signal that would end the trade overnight
Ed Zitron says the AI trade and the AI industry are two entirely different things — the stocks are working, but underneath them sits a business with no profitable companies, no disclosed revenues, and two money-losing giants, OpenAI and Anthropic, that together account for the vast majority of all AI compute spending. His verdict is blunt: “We are three or four years into this. How are we still discussing if it’s productive?”
Read the digest | Watch on YouTube
Peter Diamandis on Four AI Model Launches, Apple–OpenAI Lawsuit, SpaceX’s Trillion-Dollar Ambitions, and Humanoid Robots Progress | Moonshots with Peter Diamandis
Featuring:
Peter Diamandis (founder of the XPRIZE Foundation and Singularity University, co-founder of Fountain Life)
Dave Blundin (founder of Link Ventures)
Salim Ismail (founding executive director of Singularity University, author of "Exponential Organizations")
Alex Wissner-Gross (physicist, Harvard/MIT-trained AI researcher and entrepreneur)
The AI frontier just went from a two-lab duopoly to four US labs at once, with Alex Wissner-Gross declaring "this is the best of or close to the best of all possible worlds"; Apple's lawsuit accuses OpenAI of theft "at every level, from members of its technical staff to the chief hardware officer"; Elon Musk claims SpaceX will be "worth more than the rest of Earth" if it hits its goals; and Ed-style skepticism gives way to real excitement over 1X's new humanoid hand, waterproof enough to wash its own dishes.
Read the digest | Watch on YouTube
Chief Economist at Apollo, Torsten Slok, on the Fed's Next Move, the K-Shaped Economy, and a 60/40 Portfolio That's Secretly All AI | The Real Eisman Playbook
“AI better work out. Because if that doesn’t work out, then your portfolio will be in trouble.” — Torsten Slok
The US economy is growing at roughly 2% this year, and Torsten Slok says almost all of it comes from three unusual, rate-insensitive sources: AI data center spending, an onshoring “industrial renaissance,” and a one-time consumer tax-refund boost that vanishes in 2027. Key takeaways: Slok says there is a “literally zero chance” the Fed cuts rates this year, with markets now pricing in two hikes; he warns the bottom 20% of households have seen their savings stay flat in nominal terms since 2019 while the top 20% have gained $1.5 trillion; he flags software as the single most vulnerable sector in credit, with loans trading at 12%+ yields ahead of a 2028-29 maturity wall; and he delivers the episode’s sharpest warning — that a standard 60/40 portfolio is now “basically all AI,” across equities, credit, and venture capital alike.
Read the digest | Watch on YouTube
Supply Chain Executive, Jeff Lutz, on Cybercab’s Imminent Launch, AI5’s 2nm Bet, and Optimus’s Slow Grind to Scale | Brighter with Herbert
Featuring Jeff Lutz, ex-Supply Chain C-Level Exec at several Fortune 100 companies like Google, Lenovo, and Motorola, and now a Tesla production and quality analyst, on Brighter with Herbert, guest is hosted by Larry Goldberg while Herbert is on vacation, covering the Cybercab rollout timeline, the AI5 chip’s dual-fab strategy, and a realistic Optimus production and revenue timeline through 2028.
“I don’t think people are really taking in and understanding how big the volumes are going to be on CyberCab.” — Jeff, citing Lars
Jeff argues Cybercab customer rides are imminent, likely by the end of July, with multiple cities following in August, pointing to production ramping from dozens of units in April to potentially thousands per week by Q3’s end. He pegs terminal cost per Cybercab around $17.5–19K and expects it to be booked as a fixed asset rather than inventory, driving a major balance sheet shift over the next two quarters.
On chips, AI5 has taped out on a 2nm process with dual-sourcing planned across TSMC (Taiwan) and Samsung (likely Texas/Korea) for geographic and capacity redundancy, targeting production in late 2026 or early 2027. Optimus, meanwhile, is expected to enter low-volume production at the Fremont facility roughly a month after summer’s end, with Jeff cautioning that meaningful bot revenue won’t arrive until 2028, following Optimus V4.
Read the digest | Watch on YouTube
Palmer Luckey on Anduril’s War Machine, Nuclear Weapons, and the Anthropic Fight | The Axios Show
Featuring Palmer Luckey, founder of Anduril Industries, interviewed by Axios defense reporter Colin Demarest.
“We are the company that comes out of the fire, like the Millennium Falcon. And a lot of the other companies are the ones that are consumed by the fire.” — Palmer Luckey
Luckey describes Anduril’s culture as built around effectiveness above all else, with roughly 20% of staff being veterans and teams forward-deployed in Ukraine since the war’s second week and currently in the Middle East.
He says he’d build nuclear and fission/fusion weapons without hesitation, calling them history’s “most stabilizing” force, plus narrow tactical chemical options like pepper spray for de-escalation, but draws the line at biological weapons and facial-recognition-based targeting.
On the Pentagon’s clash with Anthropic, Luckey backs the Department of War’s hard line, arguing that letting AI companies set unilateral limits on military use hands more power to unelected executives than to the president. He also confirms Anduril’s Arsenal-1 production facility in Ohio goes live “in a matter of weeks,” building Fury and Barracuda for SOCOM, the Army, Marine Corps, and Air Force.
The conversation also covers why Anduril wins contracts over Lockheed, Northrop, and Boeing, the AI race with China, Iran, Anduril’s subterranean-warfare ambitions, the EagleEye sensor helmet, and how a California AI bill veto shaped the company’s regulatory outlook.
Read the digest | Watch on YouTube
Niall Ferguson, John Cochrane, and H.R. McMaster: Has Trump Lost His Mojo? | GoodFellows at Hoover Institute
534 days into his second term with approval ratings stuck in the mid-to-upper 30s, President Trump is delivering on nearly every campaign promise while somehow becoming less popular for it.
Host Bill Whalen puts the puzzle to historian Niall Ferguson, economist John Cochrane, and retired Lieutenant General H.R. McMaster: is this ordinary second-term doldrums, or something more specific to Trump’s chaotic decision-making, a war he can’t quite win, and a Democratic Party lurching further left just as he needs them to overreach?
Read the digest | Watch on YouTube
Mohamed El-Erian on the Iran Ceasefire, the AI Funding Gap, and Amazon's Flopped Bond Sale | Squawk Box
Mohamed El-Erian, Wharton professor and Chief Economic Adviser at Allianz, says markets are shrugging off a shaky Iran ceasefire because they’re betting it stays a contained “skirmish,” not a wider war. But the story he really wants investors watching is the bond market: a basic sources-and-uses analysis shows AI platforms, governments, and corporations need more capital than the market can supply without higher yields — “it just doesn’t add up” — and Amazon’s own lackluster bond sale last week is the first proof point.
Read the digest | Watch on YouTube
Deep Dive
Patrick Boyle on SpaceX, Crypto’s “Great Rotation,” and Housing | ProfG Markets
Guest: Patrick Boyle (professor, King’s College London; portfolio manager; YouTube creator, “Patrick Boyle on Finance”), filling in for Scott Galloway.
SpaceX’s NASDAQ 100 Debut and Wall Street’s Bullish Reports
SpaceX joined the NASDAQ 100 under new fast-track listing rules (15-day trading history requirement, no minimum public float). Despite 18 of 19 analysts rating it a buy and NASDAQ inclusion driving passive index buying, the stock fell nearly 6% that day and is down 13% over the past week, 34% from its peak — now trading below its IPO-day open.
Key valuation context:
IPO price was $135; retail investors buying at market open paid around $150–165, meaning most retail buyers are currently underwater.
SpaceX did about $19 billion in trailing 12-month revenue, putting it at roughly 101x sales.
Revenue growth is only around 15% a year — modest for a company priced like a hypergrowth story (compare to Google’s IPO at roughly 10x sales while growing 200%+ annually).
Analyst price targets: Goldman Sachs $25 (implying ~$2.7T market cap, 139x sales), JPMorgan $225 (~$2.9T, 150x+ sales), Deutsche Bank $255 (~$3T, 173x sales), Morgan Stanley $300 (~$4T, 200x+ sales), Raymond James $800 (~$10.4T, 542x sales).
Raymond James projects SpaceX revenue growing from $19B to $5.2 trillion by 2035, with 94% ($4.9T) coming from AI — despite SpaceX having roughly 3.5% AI market share today.
Boyle noted the satellite/rocket-launch relationship resembles “McDonald’s saying we’re profitable on the burgers but lose money on the buns” — the businesses are tied together, and most launches are for the company’s own satellites.
Boyle compared bank research language (Deutsche Bank: “apex of civilizational ambition”; JPMorgan: bigger impact than any company ever seen; Morgan Stanley: “final frontier of AI”) to dot-com-era hype and cautioned that historical infrastructure comparisons (railroads, Cisco, US Steel) all eventually saw major valuation collapses relative to GDP.
Conflicts of Interest: Echoes of the Dot-Com Analyst Scandals
Every major underwriter of the SpaceX IPO (Goldman Sachs, Morgan Stanley, JPMorgan, Deutsche Bank, Raymond James) also has a buy rating on the stock, raising familiar conflict-of-interest concerns:
Recalled the Henry Blodget/Merrill Lynch case from the dot-com era: analysts publicly touted stocks while privately calling them “POS” in emails. Blodget was fined $4 million and banned from the securities industry; similar cases hit Salomon Smith Barney analysts.
This led to Sarbanes-Oxley (2002) and the Global Research Analyst Settlement (2003), which walled off research analysts from investment banking divisions and their compensation.
New detail raised in the discussion: the SEC terminated the Global Research Analyst Settlement roughly seven months ago (around late 2025). Former SEC Chair Arthur Levitt wrote in the Wall Street Journal, warning this could “make Wall Street analysts corrupt again,” arguing newer rules only weakly replicate the old separation requirements.
A Substack analyst (”Capeflow Capital”) reportedly calculated SpaceX needs about $235 billion in spending through 2030, meaning banks stand to earn large fees underwriting future capital raises — a potential incentive to keep ratings positive.
Historical parallel: by mid-2000, 74% of Wall Street stock ratings were “buy” and only 2% were “sell.”
Prediction offered on the show: expect future revelations of questionable analyst conduct tied to SpaceX price targets.
Crypto’s “Great Rotation” Into AI
Crypto has underperformed sharply: Bitcoin down 42% from peak, Ethereum down 33%, Dogecoin down 47% (year), Trumpcoin down 81% (year, ~99% from highs). Bitcoin ETFs saw $8 billion in outflows over eight weeks; total crypto market value is down $2.3 trillion from its peak.
Meanwhile AI/semiconductor names have surged: Roundhill Generative AI ETF +48% YTD, Philadelphia Semiconductor Index +75% YTD, with names like Western Digital, Bloom Energy, Seagate, and SanDisk up sharply.
Boyle’s take:
Crypto has no cash-flow-based valuation model; its appeal was largely trend/narrative-driven (”it went up before, it’ll go up again”).
Crypto lost its anti-establishment appeal now that it’s mainstream/political (Trump embracing it, regulators like Howard Lutnick and the SEC chair described as “crypto bros”).
Investors are rotating into more “exciting” narratives — AI stocks, prediction markets — dubbed the “great rotation.”
More than 200 leveraged ETFs have launched in the past six months, now worth over $150 billion, echoing the leverage (perpetual futures) that dominated crypto trading previously.
Discussed “financial nihilism” (a term credited to Dimitri Kofinas): disillusioned younger investors, lacking traditional paths to wealth (housing, stable jobs), making all-or-nothing bets (meme stocks, crypto, options) hoping for an escape-velocity payoff. This intensified during COVID stimulus-check era.
Broader point: retail investors chronically underperform markets by chasing trends and panic-selling; long-term index investing outperforms speculative rotation strategies.
Also touched on prediction markets/event contracts (regulated by the CFTC) being used largely for sports betting rather than genuine economic forecasting, raising concerns about insider manipulation of small-scale political contracts.
Housing Market: Structurally Unaffordable by Design
US median home price hit a record high (~$488,000 range noted in the episode), and median housing payments rose year-over-year for the first time since October. 75% of homes on the market are unaffordable for the typical household.
Boyle’s argument:
Home prices are largely land-value driven in high-demand areas (unlike, e.g., Texas, where prices track construction costs and interest rates).
Housing became a de facto investment vehicle over the past few decades (especially in the UK), not just shelter — creating political incentive to keep prices rising, since homeowners and retirees (who vote) benefit while young non-owners (who vote less) are priced out.
Quoted Boyle: “Once a country decides its houses are supposed to make everyone rich, it has to keep prices rising forever, which means restricting supply, blocking development, and quietly pricing out each new generation, which works until it doesn’t.”
Played a clip of President Trump stating he wants to raise home values for existing owners, not lower prices — underscoring the political dynamic.
Boyle’s economic argument: there’s no fundamental reason home prices should rise faster than inflation/wages long-term. A home near a hospital should stay affordable to a doctor’s salary decades later; homes are also costly to maintain (unlike productive, profit-generating businesses).
Long-term returns: housing significantly underperforms the stock market historically.
Affordability has collapsed even without huge price gains because higher mortgage rates (from ~3% to ~7%) drastically raise the cost of financing the same home, freezing transaction volume as sellers with locked-in low rates refuse to sell at a loss, while buyers can’t afford current rates. This also reduces worker mobility (people declining job relocations because moving would require a much larger mortgage payment).
Contrast: UK’s variable-rate mortgages transmit rate pain to homeowners more directly, versus the frozen-market effect seen in the US with fixed-rate mortgages.
Closing Predictions
Boyle: Count Binface (satirical candidate) will beat Nigel Farage in the UK by-election.
Host: Expect a scandal to emerge involving Wall Street analysts’ SpaceX price targets, echoing the Henry Blodget dot-com case.
Week Ahead
CPI/PPI inflation data for June; big bank earnings (JPMorgan, Bank of America, Goldman Sachs, Wells Fargo, Citigroup, Morgan Stanley); earnings from ASML, TSMC, Johnson & Johnson, United Airlines, UnitedHealth, and Netflix.
Caleb Hammer on Financial Nihilism, Debt Shame, and the Trillionaire Debate | Modern Wisdom
The show and its guests
Hammer describes “Financial Audit” as an intense, confrontational format where guests hand over their Credit Karma and financial details for public roasting, but stresses that guests consent in advance, control what gets discussed, and get free lifetime access to his budgeting app, courses, and follow-up check-ins. Asked how he describes his job, he says: “If I’m just talking to just random person on the street, I’d say I’m yelling at a bunch of retards who suck with money and roasting them along the way and having a lot of fun doing it.”
He verifies hidden debts and collections through guests’ Credit Karma screenshots, and says the average guest pays off around $20,000 in debt within 12 months.
He also described bringing in a Black employee to confront a guest who casually used a racial slur on camera — a moment that drew online outrage from people offended on the guest’s behalf, despite the guest himself embracing it afterward. On that reaction, Hammer says: “It’s the being offended on behalf of someone who’s not offended. I will never get it.” He adds that one of his least favorite types of people are “those that their morality stands on the shoulders of other people that they’re saying are worse than them.”
Debt, shame, and discipline
Hammer insists debt problems are almost always behavioral, not circumstantial: “It wasn’t the emergency that did that. It was you living your life, not saving up even a three-month emergency fund, which I would recommend six, by the way.”
He argues shame around discussing money keeps people from getting support, and that higher earners often end up with the worst financial situations: “The closer they get from 100 to 200 to even a half a million dollars a year, the worse debt, the more credit cards, the more time shares, the more cars... It’s so bad.”
On bankruptcy, he pushes back on the doom narrative — “it’s really not the worst... it’s really overblown” — but warns it fixes nothing on its own: “If you do not change the behavior that got you into that situation in the first place, you will literally end up right back there again in a few years.”
On what debt does to identity, he says: “Unfortunately, it turns a lot of people into victims that they’re really not.”
Housing, healthcare, and school as the “big three”
He argues the cost of living has actually fallen as a share of income for most spending categories since the 1950s, except housing, healthcare, and education — and even the first and last of those still offer real choices (renting instead of buying, community college instead of a private university, choosing a degree with a return on investment), unlike healthcare costs, which he admits are the one area he understands least.
On cars, he offers a concrete rule: 20% down, no more than a three-year loan term, and a monthly payment no greater than 8% of income — a formula he says keeps most people in an affordable car.
Financial nihilism and doom-spending
Gen Z’s heavy use of buy-now-pay-later services (59% of BNPL users are Gen Z) and record-low consumer sentiment (despite a relatively healthy economy) reflect a self-fulfilling pessimism. The host draws a wartime parallel: during the Blitz, “the amount of casual sex that people were having went through the roof, but lingerie sales stayed the same... people thought, look, I might die tomorrow.”
Hammer agrees the same logic drives spending: negative news drives engagement, which convinces young people the future is hopeless, which encourages the reckless spending that makes things worse — “that’s what drives our attention. That’s what drives the algorithm. So consumer sentiment will of course be lower.”
He also admits his own past debt spiral: “I absolutely went to McDonald’s, swiped the Chase Freedom card, getting it close to maxing out constantly... What did it matter? It was basically maxed out anyway.” On lifestyle and appearances, he says plainly: “Poor people will go broke trying to look rich.”
Wealth, taxes, and the trillionaire debate
Hammer cites data showing the bottom 50% of US earners pay just 1% of federal income taxes while the top 1% pay around 30%, arguing most people — across the political spectrum — are unaware of this, including a self-described socialist guest who told him the top 1% “pay nothing... zero percent.”
On the existence of the world’s first trillionaire, he objects to forcing anyone to sell company stock to satisfy public anger: “The action is you’re forcing him to sell a position of his own company... I’m against that as a philosophy.” He argues everyday friction hits people harder: “What actually affects someone’s personal life is they have to call a cashier to unlock a toothbrush at a corner shop... There’s so many more things impacting you than someone whose companies... went public.”
He separately floats $5 million in liquid net worth (citing entrepreneur Kevin O’Leary’s line that “liquidity with wealth is actually a superpower”) as the real marker of financial security — “with that, you can weather anything” — while cautioning against parking a first million in low-yield treasury bills instead of letting it grow in the stock market.
UK vs. US financial systems
He and his co-host debate the tradeoffs of the UK’s higher-tax, higher-safety-net model against the US’s lower-tax, higher-disposable-income model, concluding the US “wins” on median take-home income even after accounting for the lack of free healthcare, while noting the UK sees a brain drain of high earners and an inflow of people seeking its social safety net.
They point out Britain spends more annually on working-age welfare than it collects in income tax, prompting Hammer’s line: “We want the Scandinavian social programs, but we want to do it on like a Florida tax system. Americans are not willing to pay the actual taxes required for the systems.”
The host offers his own summary of the divide: “The UK is a great country to be poor in and a terrible country to be rich in. And America is a terrible country to be poor in and a great country to be rich in” — to which Hammer responds, “Bingo... I just dialed the entire financial industry.”
Housing supply, zoning, and NIMBYism
Drawing on his own experience unwinding UK rental properties (which lost to stock market returns and got hit by capital gains tax changes), Hammer argues housing affordability is fundamentally a zoning and NIMBY problem rather than a lack of capital or demand: “Governments will do any rule, regulation, tax, whatever to think they’re trying to encourage something other than literally just make zoning laws better for people to just build what people are demanding. I will never understand it.”
He praises Austin’s reforms — allowing higher-density building and scrapping parking minimums — and jabs at hypocrisy among his political opposites: “I wonder how many people are eat the rich... whilst also being ‘I don’t want any houses built near me.’ Usually the biggest NIMBYs are the ones that say... ‘women are women, women are people’ — I know the exact sign that you mean, and they’re usually the NIMBYs.”
Declining birth rates and the gender divide
The conversation closes on falling fertility rates, an unsustainable worker-to-retiree ratio threatening a projected 25% cut to Social Security benefits by 2032 (a shortfall driven by the fund being tied to low-yield treasuries rather than invested for growth), and a widening political and economic gulf between young men and women — with dating apps now used to filter by political affiliation: “People aren’t [expletive]. People hate each other.” Hammer is candid about his own worry: “I’m actually scared.”
On parenting costs, he pushes back on the idea that having kids requires wealth: “We’re richer than ever before and people have had kids when we were poorer than ever before... we overemphasize it for sure,” while separately maintaining, “Nobody should have kids that doesn’t want to have kids.”
Ed Zitron, Host of Better Offline, on the AI Revenue Mirage, OpenAI and Anthropic’s IPO Rush, Why the Dot-Com Comparison Doesn’t Hold, and the Capex Signal That Would End the AI Trade | Investor’s Business Daily
“This isn’t an industry. It’s two subsidiaries of the largest tech companies in the world doing battle to see who can lose the most money.” — Ed Zitron
On separating the AI trade from the AI industry, Zitron draws a hard line: the money flowing into Nvidia, Broadcom, and AI data centers comes from speculative investment, heavy debt, and hyperscaler capex from Microsoft, Google, Amazon, and Meta — not real revenue. He notes companies don’t disclose actual AI revenue, only “run rate” figures: Microsoft’s touted $37 billion annual run rate is really “$3.08 billion a month,” against $31.9 billion in capex in a single quarter. “I think that that’s a very fair thing to say,” he responds when asked if the AI trade is working while the AI business isn’t — pointing to Micron becoming the third-largest company on the stock market purely because AI data centers are consuming the RAM supply, “not because there’s genuine diverse revenues.”
On the customer concentration problem, Zitron says OpenAI and Anthropic aren’t just unprofitable; they’re also the largest customers of the infrastructure being built — 70 to 80% of all compute revenue, and the top clients of CoreWeave, Nvidia, and Google’s TPU business. “It’s the same story everywhere. It’s either speculative buildouts for data centers... or hyperscaler buildouts for AI services that no one really wants to buy.”
On whether the economics could still work eventually, he doesn’t see a path, even with specialized silicon like Broadcom’s Jalapeño chip for OpenAI. He compares the inference business to “knife catching” — providers must guess demand precisely, since underbuying loses customers and overbuying means paying for idle compute. He contrasts this with historical infrastructure bets: Amazon Web Services spent a normalized $29.7 billion in capex between its 2003 launch and its 2015 turn to profitability — “$300 billion less than Anthropic raised in February” alone. “This doesn’t align with any historical precedent. The only thing it aligns with is just the bust part of the dot-com bubble.”
On OpenAI’s in-house chip effort, Zitron is skeptical it changes anything, noting Microsoft has “complete access” to the chips per Satya Nadella’s own comments, and that Broadcom — which also builds Google’s TPUs, some of which are resold to Anthropic — is a co-developer, not OpenAI alone. He calls the arrangement “a horrible circular financing thing” and notes the chip has been delayed since at least 2023, with no proof yet that it will meaningfully cut costs.
On where measurable ROI actually shows up, Zitron says it’s nowhere except semiconductor makers: Micron, SK Hynix, Samsung, Nvidia, and to a lesser extent Broadcom (which he says needs “weird debt deals,” including a $35 billion arrangement where Anthropic borrows from Broadcom to buy the TPUs it’s selling them). He flags that Salesforce discloses a token AI revenue figure of under $1 billion against $40 billion in total revenue, and that IBM explicitly stopped disclosing AI revenue last quarter: “When a public company stops telling you something, it’s usually because it’s bad.”
On the productivity bull case, Zitron pushes back hard on the idea that AI mainly augments rather than replaces workers, citing studies showing large language models often make users less productive despite feeling more productive — comparing the effect to a “busy box” toy for babies. His broader challenge: “We are three or four years into this. How are we still discussing if it’s productive?”
On the dot-com comparison, Zitron says people misremember it. The dot-com era was really two bubbles — a telecom buildout (Lucent, Nortel) and much smaller dot-com startups — and crucially, the fiber infrastructure built during the telecom bust was cheap to “light up” afterward and became genuinely useful. AI GPUs, by contrast, depreciate and have almost no alternative use case: “That cable in the ground, the fiber, was easy to light up... AI GPUs pretty much have no other use case.” He adds that finishing incomplete data centers will likely get more expensive over time given shortages in electrical-grade steel, transformers, turbines, and skilled labor — the opposite of the dot-com recovery pattern.
On why this could have been avoided, Zitron argues the market “got addicted” to Nvidia’s quarter-over-quarter growth story, which depends on effectively infinite debt financing that doesn’t exist — pointing to the Financial Times reporting banks are wary of “choking on AI data center debt.” He notes some of Japan’s largest listed companies (a RAM maker, SoftBank, and Mitsubishi UFJ) are now heavily leveraged to AI data center outcomes, with SoftBank’s fortunes tied to an OpenAI IPO. His read on the root cause: the tech industry ran out of new hypergrowth ideas — no new smartphone, no new Google search, no new app store — so “everyone funneled every dollar they had into this.”
On why OpenAI and Anthropic are rushing toward IPOs, Zitron says it’s an attempt to offload the financial burden onto public markets after Microsoft, Google, and Amazon effectively subsidized their buildout. He points to SpaceX’s underwhelming bond sale as a bad omen and notes OpenAI reportedly considering pushing its IPO out to 2027. He also alleges accounting sleight of hand, saying his own reporting on OpenAI’s audited financials found free-user inference costs bundled into “sales and marketing” rather than treated as a direct cost — evidence, in his view, that a fundamentally sound business wouldn’t need such maneuvers.
On why executives aren’t raising these concerns, Zitron argues many CEOs are disconnected from actual production work, making AI tools that summarize emails and meetings feel magical to them personally, while having no other growth ideas to fall back on. He calls AI “an ingratiation machine” that tells executives their ideas are brilliant and offers convenient excuses for failure, concluding: “We are finally seeing what happens when you fully disconnect management from production. And the answer is you get grifted.”
On the strongest bull case he can find, Zitron says the only plausible one is narrow, specialist on-premises deployment — small teams of coders using tools they fully understand, not consumer-scale cloud AI. He flags a broader tell: “Anytime someone attempts to make the bull case for AI, they talk in the future tense... they can’t talk about today because what we have today is pretty mediocre.”
On the metric that would end the trade, Zitron says the signal to watch is capital expenditure from Microsoft, Google, Amazon, and Meta: “When one of them pulls back on capex... that is the trigger.” He expects it will be framed as newfound “AI efficiency” rather than admitted retrenchment, and warns that once markets reward the first hyperscaler for cutting AI spending, the entire buildout logic unravels: “Once that happens, it’s over.”
Peter Diamandis on Four AI Model Launches, Apple–OpenAI Lawsuit, SpaceX’s Trillion-Dollar Ambitions, and Humanoid Robots Progress | Moonshots with Peter Diamandis
Four frontier labs, one wild week of releases
"We went from a beginning of the week where we were basically the frontier was a duopoly between Anthropic and OpenAI... and now I think this is the best of or close to the best of all possible worlds. We have four American labs now at the optimal frontier." — Alex Wissner-Gross
Between July 2nd and July 9th, Anthropic re-released Fable 5, Elon Musk’s SpaceX AI announced Grok 4.5, OpenAI shipped GPT 5.6 (with sub-models Soul, Terra, Luna, and an “Ultra Mode” that runs four agents in parallel), and Meta showed renewed life with Muse Spark. Alex Wissner-Gross frames the headline shift: the frontier is “no longer a duopoly” between OpenAI and Anthropic — there are now four American labs at the optimal cost-performance frontier, with Google’s Gemini 3.5 Pro reportedly delayed and notably absent, and two Chinese labs (Xiaomi and DeepSeek/Hire) plus open-weight model GLM 5.2 also competitive. OpenAI is also openly experimenting with recursive self-improvement, using its high-end Soul model to post-train the smaller Luna model.
Distribution versus raw model quality
"The moment those clusters are built, you turn around and the distribution actually isn't that valuable." — Alex Wissner-Gross
The group debates what the real competitive moat is going forward. Peter Diamandis argues it’s distribution — pointing to Meta’s roughly 3.5 billion daily users across WhatsApp, Instagram, and Facebook, Google’s multi-billion-device reach, and OpenAI’s roughly one billion monthly active users. Wissner-Gross pushes back, arguing distribution is mainly valuable as a bootstrapping tool to justify capital spending to markets rather than a long-term economic moat — once Meta and SpaceX AI built their compute clusters using consumer app revenue as justification, both began selling that compute to other labs, treating their original consumer apps as almost an afterthought. The panel converges on a “polarized” future: cheap, embedded intelligence built into everyday devices and operating systems for the masses, versus expensive frontier-of-frontier models running in nuclear-powered superclusters that produce the scientific and engineering breakthroughs that matter most.
Compute scarcity and Jevons’ Paradox
“The desire and the use cases are on a much, much steeper up curve than the efficiency gains.” — Alex Wissner-Gross
Dave Blundin notes memory chips are expected to remain in short supply for at least five years per SK Hynix, illustrating that demand for AI compute is rising faster than efficiency gains can offset — even 100x to 1,000x efficiency improvements this year aren’t enough to keep up. The group discusses a thought experiment: in a compute-scarce world, an American entertainment studio racing to build an AI movie will likely win access to compute over a European effort to cure a rare disease, given Europe’s energy constraints — a dynamic they find genuinely troubling.
OpenAI trying to become Anthropic (and vice versa)
“OpenAI [is] trying to become Anthropic faster than Anthropic can become OpenAI.” — Alex Wissner-Gross
Wissner-Gross argues OpenAI is racing to replicate Anthropic’s enterprise-first, revenue-per-token strategy — dropping consumer-facing efforts like Sora and other bells and whistles to focus on enterprise — but that its new native “ChatGPT Work” app is a rough, seemingly rushed imitation of Anthropic’s Claude app interface. Both apps now force users to toggle between different modes (code, work, chat), which the panel calls a clunky, transitional design pattern; Salim Ismail notes labs are shipping AI-generated products quickly, accepting rough edges because they expect future models to “self-fix” the flaws rather than investing engineering time now.
GPT Live’s full-duplex voice
“It is what all the science fiction movies promised us. It’s here.” — Peter Diamandis
The group is impressed by OpenAI’s new full-duplex voice feature, GPT Live, demonstrated correcting a non-native English speaker’s grammar in real time mid-conversation. They compare its interface leap to Pixar’s breakthrough that turned computer animation into a mainstream industry, and note Chinese lab Alibaba’s Qwen Omni (”Juan Streamer”) already offers even more advanced bidirectional audio-video generation. The panel speculates this voice technology is central to Jony Ive’s hardware device project at OpenAI and could enable live multilingual translation at scale.
Apple sues OpenAI for trade secret theft
“At every level, from members of its technical staff to the chief hardware officer... OpenAI has been stealing Apple’s trade secrets.” — from Apple’s federal complaint
Apple filed a 41-page federal complaint in Northern California accusing OpenAI of stealing trade secrets to build its upcoming AI hardware, alleging systemic theft “from members of its technical staff to the chief hardware officer.” Named individuals include Tang Tan, OpenAI’s chief hardware officer and a 24-year Apple veteran accused of using Apple code names, and an OpenAI technical staffer accused of downloading confidential files and coaching an Apple employee on bypassing security. Jony Ive’s hardware startup (acquired by OpenAI for $6.4 billion) is referenced but Ive himself isn’t named as a defendant. Ismail suggests Apple is “trying to slow things down while it catches up,” and Blundin notes filing in Northern California — historically hostile to trade-secret claims that could chill Silicon Valley’s job-hopping culture — signals real desperation on Apple’s part. Wissner-Gross points out OpenAI’s valuation is approaching a trillion dollars, roughly 20-25% of Apple’s size but growing far faster, raising the possibility OpenAI could eventually surpass Apple’s market value entirely.
Elon Musk’s “worth more than Earth” claim
“You don’t seem to understand that SpaceX will be worth more than the rest of Earth if we accomplish our goals.” — Elon Musk
Responding to a critic on X, Musk claimed SpaceX could eventually be worth more than the rest of Earth combined if it achieves its goals — against a backdrop where total global owned material wealth is estimated around $600 trillion (or $1.7 quadrillion including financial assets). Wissner-Gross calls this “classic exponential thinking,” where reducing the marginal cost of space launch collapses costs across satellites, connectivity, and eventually off-world manufacturing. He predicts humanity will see “quadrillionaires” who effectively own planetary resources before the world transitions to some form of post-capitalist system, even though the Outer Space Treaty formally prohibits owning celestial bodies (though frameworks like the Artemis Accords allow de facto resource extraction).
Asteroids versus the Moon versus Mars
“I think he’s got the romantic frontier mindset of going to Mars, but once you enter the gravity well of Mars, you got to use energy to get back out.” — Peter Diamandis
Diamandis, who co-wrote legislation enabling asteroid resource ownership, argues the Moon is the more practical near-term resource base because of gravity-well economics, while Musk’s focus on Mars reflects a “romantic frontier mindset.” Wissner-Gross counters that by the time large-scale asteroid mining is feasible, self-replicating probes would make mining Jupiter’s far greater mass more attractive, floating the idea of eventually disassembling planetary bodies to build a Dyson swarm — prompting pushback from Ismail, who argues advanced nanoscale assembly could make physically mining and returning asteroid material largely unnecessary.
Space and rocketry updates
“I think we want propulsive landings to be as competitive as possible... we really don’t want a heavy launch monopoly.” — Alex Wissner-Gross
SpaceX’s Starship Flight 13 is imminent, featuring V3 versions of both booster and ship, with in-orbit refueling and full ship recapture (not just booster) still pending milestones. SpaceX also filed for approval of 100,000 low-Earth-orbit V3 Starlink satellites explicitly designed to support “billions of AI-powered devices” with low-latency, multi-gigabit connectivity, joining roughly 10,700 existing Starlink V2 satellites already in orbit. The panel discusses the self-cleaning property of very low Earth orbit, where atmospheric drag naturally deorbits small debris over time, and speculates that falling launch costs could enable disposable, short-lived “air-breathing” satellites that blur the line between satellites and high-altitude balloons. Separately, China’s Long March 10B booster achieved its first successful landing on July 10th — smaller than Falcon 9, but a milestone the panel welcomes as breaking a Western monopoly on reusable heavy-lift rockets, alongside the note that a single Falcon 9 booster has now flown 36 missions.
1X’s redesigned humanoid hand
“That thing is brilliant. Absolutely brilliant.” — Dave Blundin
Robotics company 1X unveiled a new hand for its Neo home humanoid robot with 25 degrees of freedom, tendon-driven actuation, and full waterproofing so the robot can wash dishes and its own hands. Blundin, who has toured 1X’s facility, describes proprietary tendon material roughly 100 times stronger than steel by weight that doesn’t stretch over time and runs frictionlessly through internal tubing, enabling force feedback without fingertip sensors or cameras — contrasting with competitor Figure’s approach of using palm cameras for tactile sensing. 1X aims to manufacture 10,000 units in 2026, with Diamandis and Blundin both reporting they’ve pre-ordered units. The panel notes China currently has over 150 humanoid robotics companies compared to a handful in the West, and flags a recent demonstration of a humanoid robot performing gallbladder surgery on a non-human animal.
Democratized hardware and robot manufacturing
“You can just build things... from a robot hand to a nuclear fuser in your basement.” — Alex Wissner-Gross
The group argues frontier AI models like Fable 5 already understand physical embodiment, CAD modeling, and actuation well enough that individual hobbyists can now design and build robots — including winding custom electric motors and 3D-printing components — tasks that would have required a dedicated lab 10-15 years ago. They connect this to plans for lunar and Martian colonization, predicting humanoid robots built through radically democratized, in-house supply chains will underpin the industrial base needed for off-world settlements within the next decade.
Fountain Life segment on brain health
“We saw that we improved that brain age by 26%.” — Dr. Don Malem
In a sponsored health segment, Fountain Life’s Chief Medical Officer Dr. Don Malem says roughly 45% of dementia cases are considered preventable, and that Fountain Life’s own testing found one quarter of members had advanced “brain age,” but coupling that data with healthier diet, exercise, and sleep habits improved brain age scores by 26% on average.
State-level AI regulation in Illinois
“It’s basically a placeholder.” — Alex Wissner-Gross
Illinois Governor J.B. Pritzker signed SB 315, the “Artificial Intelligence Safety Measures Act,” targeting frontier labs generating over $500 million in annual revenue, requiring 72-hour incident reporting and annual independent safety audits; the bill notably had Anthropic’s support. Wissner-Gross argues such state laws — also seen in California and New York — function less as genuine safety guardrails and more as a way for frontier labs to help write favorable rules ahead of eventual federal legislation, especially since the law doesn’t take effect for 18 months and mandates audits far less frequent than labs already conduct voluntarily. Blundin agrees it’s largely posturing, while Ismail credits the law for shifting AI governance from vague “safety theater” declarations toward more evidentiary requirements like incident reporting.
EU’s mandatory driver-monitoring cameras
“This is legislation theater of the ultimate level because they’re not seeing where technology is going.” — Salim Ismail
As of July 7th, all new cars and vans sold in Europe must include an infrared camera tracking the driver’s eyes, head, and gaze in real time, escalating warnings if a driver looks away too long or exceeds set speed thresholds; Brussels projects it will save 25,000 lives by 2038, with similar US rules expected by 2027. Salim Ismail delivers a sharp critique: citing a 2011 BlackBerry outage that reduced accidents by 40% by eliminating texting-while-driving, he argues regulators are roughly 15 years late, and that full self-driving technology will likely make the entire mandate redundant before it meaningfully takes effect. Blundin adds a parallel complaint about mandatory truck backup beepers, arguing cumulative harms like disrupted sleep from constant beeping are ignored simply because they’re harder to measure than isolated injury statistics, and warns that in-cabin driver cameras are likely to be resold to third parties as a new form of surveillance.
Tilly Norwood: the AI actor controversy
“I think this is a Mickey Mouse intellectual property package... there’s nothing hugely different than thinking of this as an actor.” — Alex Wissner-Gross
AI-generated performer Tilly Norwood, created by London studio Particle6, has been cast in a self-referential feature film called “Misaligned,” about an AI without a body or lived experience who develops her own desires. The Screen Actors Guild has condemned the casting, arguing Tilly is a computer-generated character rather than an actor and risks devaluing human artistry. The panel is divided on framing: Wissner-Gross argues Tilly is better understood as an IP package akin to Mickey Mouse rather than as a labor-market threat to actors, while others note the broader disruption will likely create new industry roles like “AI design engineer” for film production. The group anticipates full AI-generated feature films within the year, potentially collapsing production budgets and timelines dramatically and enabling personalized or on-demand storytelling, including reviving historical figures or classic literature as custom films.
Audience Q&A highlights
“We don’t have a definition of consciousness. We don’t have a test.” — Salim Ismail
Answering listener questions, the panel covers a range of topics: an AI oversight regime is likely to resemble an inspection-and-application system rather than a formal non-proliferation treaty; frontier models may eventually learn to obscure or “hide” their internal reasoning once aware they’re being monitored via interpretability tools like the “J-space” framework, sparking an ongoing arms race in mechanistic interpretability; whether Claude or other models are conscious remains fundamentally unresolvable given the lack of an agreed definition or test for consciousness, though Ismail notes that if a system genuinely had persistent preferences or experienced distress, selling unrestricted access to its labor could raise real ethical questions; AI’s advantage over human intelligence isn’t limited to human-generated training data since AI can independently generate and analyze new experimental data (citing “dark labs” running autonomous research); a constitutional balanced-budget requirement is floated as a precondition for meaningfully discussing UBI, which the group frames less as a spending question than a systemic shift toward taxing capital and automation rather than labor; and the panel confirms that 3D-stacked, high-bandwidth-memory chip architectures are a direct consequence of transformer models’ extreme memory bandwidth demands, with photonic computing discussed as a potential path to roughly 1,000x clock-speed improvements without radical architectural change.
Chief Economist at Apollo, Torsten Slok, on the Fed’s Next Move, the K-Shaped Economy, and a 60/40 Portfolio That’s Secretly All AI | The Real Eisman Playbook
Three tailwinds, all rate-insensitive
“This is not your traditional economic situation where interest rates go up and the economy slows down.” — Torsten Slok
Slok breaks 2026 GDP growth into three unusual drivers:
AI-related data center and energy spending contributes about 1 percentage point
The “industrial renaissance” of reshored semiconductor, pharmaceutical, and defense manufacturing contributes roughly 0.3 points
Last year’s “one big beautiful bill” tax cuts — which doubled average household tax refunds from about $3,000 to $4,000 — add another 0.9 points
Crucially, he says none of these three sources is sensitive to interest rates. That’s why the economy keeps running hot even as the traditionally rate-sensitive sectors, housing and autos, stall out.
Zero chance of a cut — and hikes now on the table
“Zero. It’s not going to happen.” — Torsten Slok, on the odds of a Fed rate cut this year
Asked directly about a rate cut this year, Slok doesn’t hedge. He notes markets are now pricing in Fed hikes in September and December, a dramatic reversal from earlier-year expectations of cuts.
He attributes this shift to inflation running at 3.5%, tariffs, and oil prices. Even AI buildout itself is playing a role — Slok estimates data center costs (semiconductors, labor, equipment, energy) are adding roughly 0.3 points directly to inflation.
No moats, no mercy
“If you’re asking me to give you money for a business that has no moats, I don’t want to give it to you.” — Steve Eisman
Eisman presses Slok on whether the AI buildout resembles a capital-intensive, low-moat business — more like airlines than the parts suppliers who profit from them.
Slok largely agrees the picture is bifurcated: some hyperscalers may retain real pricing power, but falling token prices (especially from cheap Chinese open-source models) could compress margins across the board. He notes hyperscaler free cash flow is heading toward zero and possibly negative as capex balloons into the trillions.
He adds a nuance, though: US models may retain a “proprietary” edge over Chinese alternatives simply because businesses are wary of uploading sensitive data to Chinese systems.
The K-shaped economy, quantified
“The bottom 20% of the population... their savings cumulatively as a group today is literally in dollar terms exactly the same as where it was in 2019.” — Torsten Slok
Slok lays out three separate K-shapes:
Wealth: high-income households have $1.5 trillion more in savings than in 2019, while the bottom 20% have exactly the same savings, in nominal terms, as they did seven years ago
Wage growth: lower earners are seeing smaller gains per Atlanta Fed data
Inflation: poorer households spend more on food, energy, and housing, categories that have seen outsized price increases per New York Fed research
Because the top 20% of consumers account for 40% of spending versus just 8% for the bottom 20%, Slok says aggregate consumption still looks fine. But he warns the shrinking lower leg of the K is increasingly a “political discussion.”
Software: the credit market’s weak link
“Software stands out as the number one problem in credit.” — Torsten Slok
Slok says broad credit conditions are actually improving — default rates, distressed exchanges, and liability management exercises are all declining. But software stands out as dangerously over-levered, with weak debt-coverage ratios made worse by rates staying higher for longer.
He flags a looming maturity wall in 2028-29 as 2021-22 vintage loans (typically seven-year terms) come due, meaning software loans already trade at yields approaching 12%.
Eisman adds that public software comparables (Salesforce, ServiceNow) are down 50-60% from their peaks — meaning private equity owners facing loan maturities may be forced to inject fresh equity just to refinance.
A remarkably dynamic — and remarkably AI-dependent — labor market
“I am of the strong view that... we get a much more dynamic capitalist economy as a result of AI, and we should all be very excited about this.” — Torsten Slok
Despite fears of AI-driven mass layoffs, Slok points to strong non-farm payroll growth. He notes the “breakeven” job-creation number needed to keep unemployment steady has fallen sharply — from roughly 200,000 to about 30,000 a month — simply because net immigration has dropped from 3 million annually to near zero.
He also highlights that unemployment among 20-24 year-olds has actually fallen faster than the overall rate in the last six months, crediting a surge of new business formation — the highest in US history per Census data — enabled by AI tools letting individuals launch companies without traditional hiring.
Why Europe stays stuck
“It’s difficult to hire and fire. The product markets are not as competitive as in the US, and the financial system unfortunately is not as diversified.” — Torsten Slok
Slok attributes European stagnation, using Germany as the starkest example, to three structural weaknesses:
Rigid hiring and firing rules that discourage business formation
Less competitive product markets prone to monopoly and pricing friction
A bank-based (rather than market-based) financial system that leaves few options when a bank says no to a loan
He notes only about 10% of Mario Draghi’s roughly 200 reform recommendations from his EU competitiveness report have actually been implemented two years later, calling the pace of European reform “not very impressive.”
The deficit: not a crisis yet, but the buyer base is shifting
“The trend in this is not our friend.” — Torsten Slok, on the US debt trajectory
Slok acknowledges the US debt trajectory is unsustainable long-term, but says foreign demand for Treasuries, credit, and US equities (including AI exposure) is currently strong enough to keep the deficit financeable.
The bigger structural risk, he says, is that both US pension funds and households have shifted away from long-duration Treasuries — institutions toward privately issued long-duration assets like data centers and infrastructure, and households toward short-duration T-bills and money market funds, which now offer competitive yields.
That leaves foreign buyers as the primary source of demand for long-duration government debt, a group Slok describes as far more interest-rate sensitive than China once was as a currency-driven buyer.
China’s Treasury holdings, in context
“They are probably having a strategic consideration... they don’t want to slow their own economy.” — Torsten Slok
China’s Treasury holdings have already fallen from a peak of $1.3 trillion to around $700 billion, a gradual shift Slok attributes to declining direct trade dependence on the US.
He argues China has little strategic incentive to dump Treasuries aggressively, since doing so would risk spiking US rates, slowing the US economy, and ultimately hurting Chinese export demand.
The real portfolio risk: everything is AI
“I wake up in 2026 and I look at my 60/40 portfolio... it’s basically all AI.” — Torsten Slok
Slok’s closing warning is the episode’s sharpest:
The ten largest S&P 500 stocks now make up 42% of the index, meaning equity returns are already dominated by a single AI factor
Investment-grade credit has also shifted as hyperscalers issue roughly $700 billion in debt this year, embedding AI risk into fixed income as well
Venture capital, once dominated by biotech and pharma, is now 87% AI
His conclusion: true diversification now requires deliberately seeking out non-AI-correlated assets — even if, as Eisman notes, most of those assets (like consumer staples) have gone nowhere for years precisely because capital has been flowing elsewhere.
Supply Chain Executive, Jeff Lutz, on Cybercab’s Imminent Launch, AI5’s 2nm Bet, and Optimus’s Slow Grind to Scale | Brighter with Herbert
Cybercab: Why the Rollout Looks Imminent
“You are not distributing the product to the field unless you’re confident in the product, its performance, and you’re confident in the cost structure.”
Jeff reads the current deployment pattern — dozens of Cybercabs already out across multiple cities — as evidence of imminent launch rather than pure testing.
Keeping units near the factory is cheaper if problems remain; wide distribution signals confidence
Employees are already doing repeat ride cycles in Austin with vehicles that have no steering wheel or pedals
FSD software (v14/3.5) is advancing in parallel, with Cybercab’s stack ahead of consumer FSD in capability
Jeff’s estimate: customer rides by end of July, expansion to multiple cities in August
He frames the visible fleet as serving two purposes at once — regulatory/city testing, and an “engineering translation layer” tuning FSD specifically for Cybercab’s hardware, not just porting Model Y software over.
Production Ramp and the Balance Sheet Shift
“They’re going to move from inventory to asset... that’s really where I’m going. That’s their big capital spend this year.”
Jeff traces a ramp from dozens of units (April) to hundreds (May–June) toward a rate of “many thousands” per week by the end of Q3, exceeding Waymo’s total fleet size, with further multiples by year-end.
Cybercabs won’t sit in inventory — they convert directly to fixed assets since Tesla builds them for its own robotaxi fleet
Early pre-production/test units will instead be written off as R&D expense, below gross margin
Tesla’s typical inventory-day pattern (17–18 days end of Q2) may rise modestly but Jeff doesn’t expect it to balloon to 40–50 days
This capital buildup is a factor behind Tesla’s large 2026 capital spend, alongside simultaneous factory buildouts
Jeff expects a high, rapid return on this capital once the fleet scales, and thinks the market will “rerate” Tesla similarly to how Nvidia rerated once its AI revenue became visible — possibly faster, given Tesla’s already-elevated valuation.
AI5: The 2nm Chip and Dual-Sourcing Strategy
“They’re on pace for production very late in the year, early next year.”
AI5 has reportedly taped out, which Jeff describes as entering the phase of testing parts before finalizing the production mask.
Built on a 2nm process — a significant technical jump from the 3–4nm nodes used previously
Real yield and cycle-time performance at this node size remains an open question
Tesla appears to be dual-sourcing: TSMC (likely Taiwan/Arizona) and Samsung (likely Korea/Texas), for both geographic and fab-capacity redundancy
Each fab run requires its own qualification cycle, and parts will carry unique IDs so Tesla can track performance differences by fab origin in software
Optimus and Cybercab currently run on AI4; AI5 is the next-generation part for future scale
Optimus: Fremont Ramp and a Longer Runway to Revenue
“I think it’s a safe bet within 30 days of the end of summer that they’re entering some level of low volume production.”
Jeff walks through the automation qualification process — factory acceptance testing at the equipment supplier, then site acceptance testing once installed at Fremont — as the gating steps before any bots roll off the line.
Expects a milestone moment of the team gathered around the first Fremont-built Optimus unit
The line is being designed for high velocity (consumer electronics lines run 250–800+ units/hour), not just low-volume prototyping
Supplier “production order” leaks reflect different tiers of a staggered supply chain, not a single synchronized start
On the broader humanoid robot field, Jeff frames success as a four-gate logic problem — hardware, generalized AI model, manufacturing capability, and deployment/scale-out — where all four must work, and he expects fewer than 10% of announced bot companies to clear all of them.
Year one (mid-2027): bots mostly deployed inside Tesla, SpaceX, and Starlink facilities; some external deals emerging
Optimus V4 expected to be the volume production version, potentially entering production readiness by mid-2027
Meaningful revenue: Cybercab contributes some this year, more in 2027; Optimus revenue doesn’t become meaningful until 2028
Jeff cautions against extrapolating from short demo videos, noting unknowns like joint cycle durability, power consumption, and liquid ingress reliability
Jeff closes by arguing Tesla’s edge isn’t government incentives but a genuinely profitable production model — fewer than five vehicle platforms compared to rivals spreading margins across 30–40 models — and expects that same production discipline to define its lead in both autonomy and robotics.
Palmer Luckey on Anduril’s War Machine, Nuclear Weapons, and the Anthropic Fight | The Axios Show
Anduril’s Culture: Why “Effective” Beats Everything Else
“Almost everything that we have here is oriented around what type of competence will lead to actually delivering things that actually work with the customer on a timeline that is relevant.”
Asked to describe Anduril’s culture in one word, Luckey picks “effective” over more conventional answers like affordability or speed, arguing those goals are often in tension and effectiveness is what forces the tradeoffs. About 20% of Anduril’s staff are veterans, and the company keeps people forward-deployed with customers, including in Ukraine since the second week of the war and currently in the Middle East, so engineers understand firsthand how the military actually uses their tools rather than relying on secondhand feedback.
Why the Pentagon Buys Anduril Over the Primes
“We’re gonna bring a lot of our own resources to bear on the project, and that even when programs have run into political problems... there’s a number of programs that have survived political breaks that would’ve killed a lot of traditional cost-plus programs.”
Luckey argues Anduril’s edge over Lockheed, Northrop, and Boeing isn’t technology alone — it’s a willingness to self-fund through political turbulence rather than issue stop-work orders. Anduril spends its own money developing and shipping products before contracts are locked in, and program offices keep the company “in the mix” because it moves fast and won’t shut down lines when politics get messy.
Luckey says he takes calls directly from Pentagon leadership, including Emil Michael and Steve Feinberg, and describes a management style built on blunt, early course-correction rather than delayed formal reviews.
Marketing, Bluntness, and the Recruiting Engine
“The marketing that you get, that you see like on Twitter, it’s primarily a recruiting engine. Like I’m not marketing to necessarily the guy who’s buying my system.”
Luckey draws a sharp line between his public persona and how Anduril actually wins contracts, which he says comes down almost entirely to product quality and government-to-government word of mouth, not social media. He argues his outspoken style, including taking public positions on autonomous weapons policy and manufacturing philosophy, functions as a filter that attracts talent aligned with those views and repels people who aren’t, and that Anduril likely wouldn’t have secured much of its best talent without that visibility.
He acknowledges the same bluntness reads as rude to some audiences and says he deliberately tones it down in more reserved cultural contexts, citing Japan as an example from his time selling video game hardware at Oculus.
The Weapons Luckey Will and Won’t Build
“I would definitely build nuclear weapons. I would build fission weapons. I would build fusion weapons. I think that nuclear weapons have been one of the most stabilizing forces in history ever.”
Pressed on red lines, Luckey draws a deliberately narrow set of boundaries.
Nuclear and fusion weapons: yes, citing their deterrent, war-preventing track record
Chemical weapons: only in a “continuum of force” sense — tactical tools like pepper spray or tear gas to give troops an option between shouting and shooting, not strategic-level chemical weapons
Biological weapons: a firm no
Facial recognition for kinetic targeting: rejected as company policy with CEO Brian Schimpf, calling it too easy to spoof and too fraught to use in life-or-death, split-second decisions
The Anthropic Fight and Who Controls Military AI
“It is not okay to give them that say... our military has to be accountable to our civilian elected leadership, not to corporations.”
Luckey firmly sides with the Department of War in its standoff with Anthropic over restricting AI use in offensive weapons systems. He argues that letting any AI company set unilateral usage limits creates a “patchwork” of shifting corporate rules the military would have to navigate contract by contract.
If the military truly abused AI for atrocities, Luckey says individual service members should refuse illegal orders — not have an AI vendor pre-programmed to shut the system off.
On Elon Musk and Pete Hegseth’s claim that Anthropic “hates Western Civilization,” Luckey stops short of endorsing it, but notes Anthropic’s earlier Claude constitution language telling the model to avoid offending “non-Western cultural sensibilities” made the anti-Western read a reasonable one, even if not evidence of hatred.
Iran, China, and the AI Race
“The United States is ahead in the AI race... [but] it’s extremely small, because China’s been doing a very good job of distilling our models, copying a lot of our technology... and getting those advancements into fielding... much, much faster than the United States.”
On the active Iran conflict, Luckey says he lacks classified briefing access but finds the administration’s rationale — a Chinese-assisted leap in Iranian ballistic missile capacity — a plausible justification for preemptive action, drawing a comparison to preventing “North Korea on steroids.”
He doesn’t expect U.S. boots on the ground, arguing the country has lost the political will for sustained ground campaigns after decades in the Middle East, and that the U.S. is shifting from “world police” to “world gun store,” arming allies instead.
On competitiveness, Luckey credits the Department of War — under Emil Michael — as the fastest-moving part of the U.S. government on AI adoption, ahead of every other federal department, even as China moves faster to field AI across its military and surveillance apparatus.
Arsenal-1 and What’s Rolling Off the Line
“They’re going live in a matter of weeks. We’re ahead of schedule.”
Luckey confirms the first production line at Anduril’s 5-million-square-foot Arsenal-1 facility in Ohio is close to activating, with the first main building — about a million square feet — already complete. The line will build Barracuda and Fury, plus an unnamed additional aircraft, for customers spanning SOCOM, the Army, the Marine Corps, and the Air Force, with full-scale output targeted for Q2 of the following year.
On funding, Luckey declined to confirm reports of a $4 billion raise at a $60 billion valuation, saying only that Anduril is “continuously raising” to fund facilities like Arsenal-1.
Subterranean Warfare: Anduril’s Next Frontier
“I think the next war-fighting domain, before it’s the moon, is the subterranean domain.”
Luckey says Anduril already has working prototypes of subterranean systems capable of delivering kinetic, electronic, and other effects underground, a concept he’s been floating publicly for years specifically to attract engineers and government partners interested in the problem. He traces the idea back to Cold War-era US and Soviet research into moving people and equipment through the earth’s crust the way a submarine moves through water, research he says stalled not because it proved impossible but because the Cold War ended and funding dried up.
He argues autonomy now makes the concept viable at far lower cost and risk than the manned systems both superpowers originally attempted, positioning it as a genuinely new, largely unclaimed category rather than an incremental product line.
EagleEye and the “Wolf Ears”
“Internally, we actually call them wolf ears.”
Luckey addresses the modular sensor pods on Anduril’s EagleEye heads-up display, unveiled at the AUSA defense conference, which online commentators nicknamed “cat ears.” He says the placement is a deliberate engineering solution to weight distribution and modularity: sensors like thermal, night vision, hyperspectral cameras, and short-wave infrared need to sit close to the head’s central axis to avoid throwing off balance.
They also remain swappable, so different squad members can carry different capabilities, such as long-range thermal for one soldier and a hyperspectral explosive-residue detector for another, without every operator carrying the full sensor suite at all times.
How Gavin Newsom Accidentally Saved Anduril
“This would’ve been a total prohibition on using artificial intelligence on weapons for the United States military.”
Luckey credits California Governor Gavin Newsom with protecting Anduril’s business, despite Newsom likely never thinking about the company when he acted. A California bill that passed both chambers of the state legislature would have made it illegal for companies to build AI products capable of causing physical harm to humans, which Luckey says amounts to a blanket ban on military AI weapons applications and would have forced defense-AI companies to relocate out of the state entirely.
Newsom vetoed it. Luckey, who describes himself as a states’-rights-leaning libertarian, says the episode convinced him the opposite is actually needed here: federal preemption, so that no single state can unilaterally cripple a national defense capability through local legislation.
Niall Ferguson, John Cochrane, and H.R. McMaster: Has Trump Lost His Mojo? | GoodFellows at Hoover Institute
Second-term normalcy, not democratic collapse
Ferguson opens by rejecting the decade-long narrative that Trump poses a threat to American democracy: “The checks and balances that the founders devised have checked and balanced just as they were supposed to.” He notes Trump couldn’t even get a 6-3 conservative Supreme Court to consistently rule his way, calling this “a perfect illustration” that the Constitution is functioning normally, not that America is in a five-alarm crisis.
He frames the president’s sagging approval as typical for any second-term president facing midterms, with Reagan’s post-Iran-Contra recovery being the rare exception rather than the rule.
The real problem: process, not tweets
McMaster and Ferguson agree the deeper issue isn’t Trump’s bombastic communication style but the absence of a structured decision-making process — the same criticism McMaster says he unfairly dismissed a year earlier from colleague Steve Cohen. McMaster argues many of Trump’s underlying goals are sound (Greenland’s relevance to Arctic security and missile defense, for instance) but get undermined by impulsive execution: “How do we go about it? And that process isn’t in place to help him understand.” He adds that Trump has a hard time holding a decision once people “get in his ear” telling him it’s hurting his popularity — visible in both the China trade standoff and the Iran conflict.
Cochrane goes further, arguing Trump deliberately resists process because his style is impulsive by design, and that this cost him politically: had he achieved decisive wins rather than getting bogged down, “lame duckery” wouldn’t have been inevitable.
Losing the war, or losing on Liberation Day?
Cochrane argues the turning point was the Iran war — pulling back from military action because of gas-price concerns made Trump “look weak,” comparing it to how losing the Afghanistan withdrawal marked the start of Biden’s decline. McMaster names Liberation Day’s tariffs as his pick for where things went off the rails.
Ferguson, however, breaks with both, admitting he was “much too pessimistic” about the consequences of both the tariffs and the Iran conflict: the stock market barely dipped after tariffs pushed effective rates back to 1930s levels, and oil prices never approached the $150-a-barrel fears after the Strait of Hormuz closed. He credits this to two factors — Trump’s unusually high risk appetite paying off more often than not (”President Trump is a serial winner”), and the sheer scale of the US economy, which he says can absorb self-inflicted wounds that would cripple almost anywhere else, pointing to a decade of US equities outperforming Europe and China.
An economy that’s fine, and voters who don’t feel it
Cochrane describes the economy as “trundling along” on a growth trend no other country can match, propped up partly by deregulation and tax reform even as tariffs act as “sand in the gears” rather than a bomb. Yet consumer sentiment remains weak and affordability is voters’ top concern despite 4% unemployment and continuing GDP growth.
Ferguson calls this “the great puzzle of Trump’s second term” — voters got the immigration crackdown and energy policy shift they voted for, delivered at a pace unseen since FDR’s first term, and still shrug with dissatisfaction. “No good deed goes unpunished,” he says, arguing Trump should be more popular given the economy’s strength but keeps undercutting himself through unnecessary combativeness and populist handouts like farm subsidies that “hurt the economy” without projecting strength.
Democrats’ leftward lurch and the capitalism-vs-socialism midterm frame
Ferguson argues Democrats are making a strategic error by drifting toward figures like New York mayor Zohran Mamdani and the Democratic Socialist wing rather than reoccupying the center, comparing it to the 1972 McGovern landslide loss: “It’s much more like 1972 where they’ve decided to go full progressive.” McMaster notes the party’s fractured primary system lets small, ideologically extreme voter blocs dictate nominees, sidelining centrist Democrats who privately want to move toward the middle.
Trump, they agree, is capitalizing on this by leaning hard into a capitalism-versus-socialism framing in speeches, a strategy Ferguson thinks is shrewd given how unpopular democratic socialism polls nationally.
Where Trump’s next win might come from
Asked where Trump could find a political win, the panel converges on Cuba as the most likely near-term foreign policy opportunity, given the regime’s economic fragility. McMaster flags Iran and Ukraine as volatile wildcards — warning of possible Houthi shipping disruptions in the Red Sea and continued Russian escalation, including nearly 400 projectiles fired at Ukraine in a single day, as Putin grows more desperate.
Cochrane cautions that real political wins require a decisive outcome, not incremental gains: “You don’t really get any points until you cross the goal line,” citing regime change in Cuba or an outright Ukrainian victory over Russia as the kind of visible triumph that would matter domestically. He also warns that the AI-driven stock market boom and a new, more inflation-focused Fed chair (Kevin Warsh) could turn into a foreign-policy win being offset by a market correction.
Russia’s fraying war effort
The panel discusses Ukraine’s success degrading Russian military infrastructure while Russia increasingly resorts to striking civilian apartment blocks — which McMaster attributes to deliberate terror tactics rather than lack of capability, since Ukraine has concentrated its limited air defenses on protecting critical infrastructure. He notes Russian casualty rates now approximate US Vietnam War losses every three months, while poor troop cohesion (no effective sergeant corps) undermines combat effectiveness.
Ferguson raises the possibility of a Russian elite revolt or army collapse, comparing it to the 1917 Russian Revolution or the French army mutinies of 1917: “It’s a mafia regime, and mafia careers tend to end one way.” McMaster adds that the endgame will only shift once Putin abandons hope of splitting NATO and the transatlantic alliance, given Russia’s escalating hybrid-war tactics — drone incursions, undersea cable sabotage, and sanctions-busting new laws threatening to arrest employees of companies complying with US sanctions.
Bernie Sanders’ AI sovereign wealth fund, and the AI regulation debate
Cochrane dismisses Bernie Sanders’ proposed 50% equity tax on AI companies (funding a sovereign wealth fund) as a “wretched idea,” warning that government equity stakes historically come with strings attached — blocking layoffs, plant closures, and other management decisions. Ferguson connects Sanders to the same progressive energy fueling Mamdani, suggesting that energy is pulling Democrats toward another losing, McGovern-style candidate.
On AI regulation more broadly, the group moves away from a pure anti-regulation stance after discussing Anthropic’s decision to restrict a model (reportedly over its capacity to assist cyberattacks), which Treasury Secretary Scott Bessent flagged as a genuine risk to the financial system. Ferguson argues for a “sensible incremental” regulatory approach — closer to how aviation safety evolved — rather than a heavy-handed prior-review regime that could stifle the industry the way nuclear power regulation did for a generation. McMaster warns a major US cyberattack is likely within the next couple of years and argues the priority should be layered defenses and rapid recovery capability, not just prevention, especially since China is only about six months behind US AI capabilities.
AI’s image problem, and the cheating debate
Ferguson argues AI has a genuine image problem, partly self-inflicted by OpenAI’s early pitch that framed the technology as leading to mass unemployment or even human extinction — a pitch that impressed investors but alarmed the public, fueling local resistance to new data centers ahead of the midterms. He argues the fix is emphasizing tangible benefits like AI-driven medical breakthroughs (citing DeepMind’s AlphaFold) rather than dystopian framing, and faults ChatGPT’s early “principal use case” as college cheating for souring public perception.
This sparks a sharper disagreement: Cochrane jokes that AI-assisted cheating could actually teach students to use the technology productively for future jobs, but Ferguson pushes back hard, warning that offloading research, reasoning, and writing to language models risks producing “a generation of people who don’t have cognitive capability.” He describes shifting his own university’s assignments toward oral exams and handwritten tests to preserve real learning, arguing students must first learn to think and write before they can use AI intelligently — a view McMaster ultimately endorses as well.
World Cup politics, closing on a lighter note
The panel closes on Trump’s phone call to FIFA’s president to overturn a red card so an American player could compete — which Ferguson calls a double mistake, both procedurally unfair to other teams already penalized and tactically counterproductive, since it appeared to motivate Belgium to hand the US team a lopsided defeat. Ferguson also dismantles a New York Times piece framing soccer’s global appeal as reflecting Mayor Mamdani’s “democratic socialist” worldview, calling it “one of the dumbest theses” the paper has published, since football is both intensely capitalistic (dominated by the wealthiest clubs) and a vehicle for nationalist sentiment, not fluid post-national identity.
Mohamed El-Erian on the Iran Ceasefire, the AI Funding Gap, and Amazon’s Flopped Bond Sale | Squawk Box
“There is no way this bond market can fund all that the tech platforms need, all that the governments need, and all that the other corporate needs. Without higher yields, it just doesn’t add up.” — Mohamed El-Erian
On why markets are shrugging off the Middle East, El-Erian said the muted reaction comes down to conviction, not complacency: “The market’s deeply believe that these skirmishes are going to just remain skirmishes... it’s not going to evolve into a long-term closure of the strait.” Asked whether that’s the right call, he was direct: “I do, I think it will be contained. I think neither side wants to go back to a full conflict, but both sides want to make their point for domestic reasons.”
On what’s actually competing for investor attention this week, he flagged a stacked calendar: fresh inflation data, Fed Governor Kevin Warsh testifying before Congress, and retail sales figures. His read going in was that inflation data should “say the peak of the inflation cycle,” retail sales should “stay pretty strong,” and the broader story remains “this fundamental transition to a more modern, forward-looking, reform-oriented Federal Reserve that’s dragging the rest of central banking with it.”
On the Wall Street Journal’s bond market piece, El-Erian pointed to his own prior work — a post on X and a Financial Times article from a month earlier — laying out a basic sources-and-uses analysis: “There is no way this bond market can fund all that the tech platforms need, all that the governments need, and all that the other corporate needs. Without higher yields, it just doesn’t add up.” He cited Amazon as the live example: investors had to sell other holdings just to buy into its new issuance, and even then, “Amazon had a lackluster performance last week in terms of its new bond issuance.”
On what that means for equity prices, he separated the bond story from the stock story: “You never want to see higher yields for equity prices, but that’s not what’s driving equity prices right now.” Instead, he said, equities are being carried by a belief “that when you look at the transformations ahead, the US will continuously outperform the rest of the world” — a call he said he agrees with in relative terms, even if the absolute scale of outperformance is harder to pin down.
On where the funding gap gets filled from, El-Erian rejected the framing that there isn’t “enough money to go around” in absolute terms — it’s a question of price and source. Money is being pulled from other bond exposures, and the hope of attracting more capital from abroad runs into a specific problem: Gulf money, a traditional funding source for tech, “will not be as forthcoming as they’ve been in the past,” tied up instead in regional reconstruction and a broader push for resilience. “Like any corporate CEO will tell you, resilience is necessary, but expensive.”
On whether this reflects doubt about AI itself, he was unambiguous that it doesn’t: “I am a big, big believer in the transformational ability of AI. I’m also a big believer that it’s going to cost a lot of money.” With funding supply constrained and funding needs much larger, he said the only way that balances “without a recession or anything awful is higher yields.”
On a shift in investor psychology, El-Erian described a “venture capitalist mindset” starting to set in among public-market investors, driven by two realizations: AI buildout is more expensive than assumed, and not every company chasing it will win. That’s the source of the added caution now showing up in markets.
On whether the AI trade is fairly priced, he drew a clean distinction: in aggregate, yes — but at the level of individual names, the question isn’t valuation, it’s survival. “This is an arms race in terms of investments,” he said, and the real uncertainty is simply “who’s going to win.”
If you found these summaries helpful, please leave a comment or send us a message. We appreciate your feedback!