Peter Diamandis and his three regular co-hosts work through about fifteen stories recorded two days after their previous episode, opening on Nvidia's quarter and the question of how much of the demand behind it Nvidia is financing itself. The show then moves through China's video models, a Chinese bot farm targeting American data-center policy, OpenAI's Broadcom-built inference chip, Apple's absence from AI, uranium enrichment, cloud-seeding drones, a pancreatic cancer approval, the lunar economy and Tesla's robotaxi fleet, with long personal digressions about Steve Jobs, border security and dashboard cameras along the way.
Host: Peter Diamandis, founder of XPRIZE, Singularity University, ZeroG and A360
Also on: Salim Ismail, founder of OpenExO and general partner at Exponential Venture Capital; Dave Blundin, founder and general partner at Link Ventures; Alexander Wissner-Gross, computer scientist and founder
Published: 29 August 2026 on the Moonshots with Peter Diamandis feed
Watch on YouTube | Apple Podcasts | 2 hr 28 min
✅ Time saved: 1 hr 38 min
Key Takeaways
Nvidia booked $96.2 billion in a single quarter and guided 2028 growth at 70% against a street at 44%
Revenue up 106% year over year, with current-quarter guidance of $108 billion — more than a billion dollars a day
Mr. Blundin puts training at 90% of revenue and 95% of profits, at 80% to 85% gross margins
The unanswered question is how much of Nvidia's demand Nvidia is paying for
"I would not like to discover two to three years from now that some or a large portion of all of this demand at the infra layer was being artificially propped up through financial engineering." — Alexander Wissner-Gross
Inference is leaving Nvidia and training is not
OpenAI's Jalapeño chip claims 1.5 to 1.9 times more AI work per watt than Nvidia's GB200s and GB300s, on half the power
"CUDA as the inference-time moat is already dead" — Dave Blundin, who says the interconnect Nvidia bought with Mellanox is the new one
China is building world models while America builds language models
Video generation is roughly 70% of all AI token consumption in China, per Runway's co-founder
Mr. Wissner-Gross's explanation: American labs are maximizing revenue per token and Chinese labs are not
X says it found a 200,000-account Chinese bot farm pushing the claim that data centers raise electricity bills
In Quincy, Washington, CNN reported the poverty rate fell from 29.4% to 6.2% after data centers arrived
A small-business survey points the opposite way from the job-apocalypse forecasts
Principal Financial Group, with 130,000 customers: 4% expect AI to cut staffing, 31% expect to add staff
Of the firms that cut staff, 1.4% attributed the cut to AI or automation
Agency is the trait superintelligence rewards, and the window is short
"Do it now. Otherwise, do it never." — Alexander Wissner-Gross
Flock's license-plate network has moved surveillance from finding a suspect to finding people who behave like one
More than 6,000 communities in 49 states, over 20 billion vehicle scans a month
"the Constitution does not update when transaction costs go to zero" — Salim Ismail
Apple has the best AI hardware in the industry and no AI strategy
"They don't even deserve to be a Mag 7 company anymore." — Dave Blundin, on a Mac Studio whose headline feature is 512 gigabytes of memory
A startup enriched uranium to 15.38% purity, five times what the government labs had reached
Every small modular reactor company promises 2030 startups and none of them has fuel
The first RAS inhibitor for metastatic pancreatic cancer nearly doubled median survival, from 6.7 to 13.2 months
Response rates tripled, 11% to 32%; RAS mutations drive about 30% of all human cancers
Mr. Wissner-Gross found almost no AI in the drug's development and reads it as the last of a pre-AI generation
The case for the moon is chip fabs, not tourism
"The moon may not be a harsh mistress. She may be a wealthy mistress at this point." — Alexander Wissner-Gross
Nvidia's Quarter Is Real; the Question Is Who Is Paying for the Demand
The show opened on how fast the news cycle has become. Mr. Diamandis said the four of them had recorded only two days earlier and it felt like a lifetime; Mr. Blundin said every story on the list was information from the last two days. Mr. Diamandis reviews more than 200 stories between episodes, with a daily digest from his own AI and from Mr. Wissner-Gross, and narrows them to about fifteen. Mr. Ismail had come back from 36 hours in Brazil and volunteered that he probably has one of the worst carbon footprints anywhere right now, which drew a suggestion that he buy offsets like Bill Gates and an offer to plant a tree on the front lawn of every airport he passes through.
Mr. Diamandis then opened with the print. Nvidia booked $96.2 billion of revenue in a single quarter, more than the GDP of the majority of countries on the planet, up 106% year over year, with guidance for the current quarter of another 10% higher at $108 billion. That is more than a billion dollars a day. Jensen Huang's guidance for 2028 is 70% growth against a Wall Street consensus of 44%, and Mr. Huang described the frontier labs as, in Mr. Diamandis's reading of the quote, "the first generation of startups that needed tens of billions of dollars of compute to get to their product."
Mr. Blundin said Nvidia is sold out for as far as the eye can see, and that the profit line is without precedent
Training is 90% of revenue and 95% of profits, at 80% to 85% gross margins
"The profitability of this company has never been seen before on the face of the earth." — Dave Blundin
The single largest vulnerability is that TSMC makes everything Nvidia sells
Roughly a third of TSMC's manufacturing goes to Nvidia and another third to Apple, leaving a third for everyone else
Building its own fabs would mean doing it "very quietly and sneakily" because Nvidia cannot afford to antagonize its only supplier
Mr. Blundin's advice to founders is to get inside the capital flow or accept irrelevance
He said his portfolio companies take meetings with manufacturers and banks when the money is all moving through one funnel
"if you're not in that flow of conversations, you're not in the most relevant conversation in the history of the world" — Dave Blundin, who suggested settling for one of the 20 to 25 people on Mr. Huang's executive staff
Mr. Wissner-Gross wants to know how much of the demand Nvidia is financing itself
The "elephant in this particular room," in his phrase, is how much of the demand the financial markets have priced as Nvidia-financed
He drew a distinction between an income-statement circularity, where two companies buy from each other, and a balance-sheet one, where a company lends a customer the money to buy its own goods — the second being the less healthy form
As a shareholder "of the entire market via low-cost broad index funds," he said he lacks visibility into which of the two is at the innermost loop
Even if the demand is partly engineered, his forecast is a mini-winter rather than a bust
The reason is substitution: he argued compute is replacing real estate, human labor and other raw inputs as the substrate of civilization
"I don't want this to be a bubble. I really don't want any bubbles." — Alexander Wissner-Gross
Mr. Diamandis noted the countervailing mood, citing videos from Ray Dalio and other financial figures saying they are out of the market on price-to-earnings ratios. Mr. Ismail said the circular revenue concerns him, and that the healthier path is steady growth that lets markets self-correct into new structures rather than the vertical line they are on now.
Nvidia Is Trying to Become the Operating System for Intelligence, and Microsoft's GitHub Deal Is the Template
Mr. Ismail read the reported Hugging Face acquisition as Nvidia moving beyond chips, and compared it to Microsoft's purchase of GitHub. He told the story at length: Microsoft moved to buy a company less than ten years old for $7.5 billion, and the finance and corporate-development staff tried to kill the deal because there were almost no assets, almost no workforce and almost no intellectual property to put on the balance sheet.
Satya Nadella overrode the objection on a single argument
"we're buying 30 million developers loyalty because that's how many people use GitHub at the time" — Salim Ismail, relaying Mr. Nadella
In the AI era, he said, GitHub gave Microsoft access to open-source thinking, and buying a community is a classic ExO play
Mr. Ismail's larger worry is not AI valuations but the currency system behind them
He said money printing has a bigger chance of collapsing markets than anything in AI, and called the state of the bond markets and Japan "absolutely insane"
Musk's $3.5 Trillion SpaceX Number Is Not Impossible, but It Requires a Merger
Mr. Diamandis put up Elon Musk's revenue projections: on a previous earnings call, a trillion dollars of SpaceX revenue by 2030, possibly 2029, and then a tweet the day before raising that to $3.5 trillion in 2030. For reference, the highest-revenue company today is Amazon at $828 billion.
Mr. Blundin said no company in history has reached a trillion in revenue, Amazon and Walmart included
His read on Mr. Musk's forecasting record: "the size always turns out to be right, and the timeline is off by a couple years because of unforeseen delays"
Mr. Wissner-Gross laid out the path he thinks gets there, and it is not linear
SpaceX and Tesla merge in the next year or two, giving the combined company Optimus revenue, which he said could alone be a multi-trillion-dollar-a-year business
SpaceX's hyperscaler business, which he said was the majority of its revenue growth in recent months, continues to grow terrestrially and then in orbit
"I think Starlink probably ends up being rounding error." — Alexander Wissner-Gross, who put cars in the same category
He is skeptical of the applications and service-economy revenue the SpaceX S-1 leans on, arguing the company's strength is physical infrastructure rather than the application layer
Mr. Diamandis disclosed that SpaceX is his largest financial bet
His reasoning: buying OpenAI or Anthropic is a single-layer bet on models, while SpaceX spans chips, data centers, models, communications, infrastructure and launch
The biggest risk he named is Mr. Musk being around long enough to implement the vision
Everything Reduces to Training and Inference, and the AI Economy Has Stopped Caring About the Old One
Asked how a listener should think about the market, Mr. Blundin said the whole picture collapses into two things. Training needs GPUs coherent in one location — 100,000, then a million, then several million — while inference can be spread across the world and into orbit. Different chips will serve the two markets, and everything else follows from that split. Robotics is investable, he said, but only for companies with proprietary inference-time models good at design and manufacturing; without those, a robotics company cannot sustain an edge.
He argued SaaS survives for now by choice rather than by moat
The labs, he said, have decided not to disrupt every existing software business while they work on bigger problems
On the circularity charge, Mr. Blundin's answer is that every economy is circular
He asked listeners to imagine a civilization on Alpha Centauri that discovers superintelligence: it builds a large, valuable economy inside itself and does not bother trading with Earth
"Every economy is circular in nature." — Dave Blundin
His brother at Vestmark, which works with Fidelity, Ameriprise and UBS, has changed his mind about what counts as AI arriving
A year ago the test was whether AI showed up at Fidelity; now, Mr. Blundin said, the answer is that it does not care
He noted that Ozempic and Mounjaro are together bigger than all AI inference combined, and that five or ten more drug discoveries would send more revenue back into the AI economy than the legacy world produces
Mr. Ismail's analogy is search advertising
When Google launched AdSense and AdWords, people expected it to create some value; it ended up absorbing nearly the whole advertising industry
He expects AI to absorb data and revenue from one industry after another, citing Google's purchase of Spirit Airlines data
Mr. Blundin offered a crude screening signal from his own portfolio
Founders who know how to train or fine-tune a foundation model have gone up 10, 20 or 100 times, in some cases 1,000 times, with "No losers yet"
Application-layer founders who describe themselves as vibe coders are, in his telling, the bucket full of losers
China Spends Its Tokens on Video Because It Is Not Trying to Maximize Revenue
Mr. Diamandis covered two Chinese stories together. Alibaba released Wan 3.0, which generates 30-second single-pass videos from documents, spreadsheets, slides and web pages, a day after the largest-ever private follow-on offering by a Hong Kong-listed company at $10.2 billion. Separately, Runway's co-founder reported that video generation is now roughly 70% of all AI token consumption in China, driven by short-form content and robotics, and growing faster than Claude grew in the US. The framing Mr. Diamandis used: America is large-language-model-pilled and China is world-model-pilled, one predicting the next token and the other the next state of reality.
Mr. Wissner-Gross offered what he called a unified field theory for the split
"American AI labs are revenue maxing and Chinese AI labs are not." — Alexander Wissner-Gross
OpenAI abandoned video generation, he said, because Anthropic ran away with the lunch by maximizing revenue per token on code generation
Chinese labs give model weights away, which is not what a revenue-maximizing lab does, so they can afford to spend tokens on video
Asked which path reaches AGI first, he rejected the premise twice
"We already have ASI." — Alexander Wissner-Gross, who does not accept that superintelligence is still ahead of us
Reframed as which approach advances the capability frontier, his answer was omnimodal models, and he expects vision, video, audio and autoregressive text generation to combine within six to twelve months, whether through diffusion transformers or a hybrid
Mr. Blundin added a commercial reason China went to video
Chinese labs cannot sell Kimi or Qwen into enterprise because buyers do not know whether to trust them; video is a global product nobody worries about on security grounds and it generates large revenue
The race that matters, he said, is recursive self-improvement, and it comes down to smarter engineers with cash and training chips
He credited Demis Hassabis with thinking it through first but said Google was too bureaucratic to act, and that Dario Amodei took the idea and focused Anthropic on a model that improves itself through better code and better training ideas
Mr. Ismail sees convergence, with the physical world as the tiebreaker
The US is optimizing around software engineers and knowledge work; China around manufacturing, commerce, media and robots
If the destination is intelligence that can model and manipulate the physical world, he said, China's direction may have the advantage — unless a US lab reaches recursive self-improvement first, in which case Mr. Blundin is right
Mr. Diamandis mentioned that Sora's absence has cost him time with his kids; Mr. Wissner-Gross predicted it will be back before long.
AI Video Is a Persuasion Machine, and China Is About to Own the Supply
Mr. Diamandis carried over a point from the show's conversation with Emad Mostaque the previous week — Mr. Mostaque had been booked for this episode and cancelled on late travel — that AI is among the most persuasive agents available, because it can understand a person in order to persuade them. Combine that with video generation, he said, and the country that becomes the dominant producer of video content in any language on any topic has a cultural instrument.
Mr. Blundin said the effect is visible in his own house
His youngest son, Jack, deleted Instagram from his phone that morning; the kids had already dropped TikTok
The recommendation engine is good enough, he said, that an hour of doomscrolling passes before a user notices
He contrasted that with what China permits at home
Citing Alvin Wang Graylin's appearance on the show, he said China bans detail down to the level of blood in a video game, where it is rendered green
A government able to regulate at that granularity, he argued, will protect its own children while selling the same product to every other country
His read on the CCP's view of open societies: they can launch an AI-generated influence machine at any country and turn the dials from Beijing, and the target cannot respond
China Is First to Regulate AI Companions, and the Panel Splits on Whether That Lasts
China is tightening restrictions on AI companions over concern that users are forming emotional dependence on virtual partners, prohibiting the services for minors and restricting them for adults. Mr. Diamandis tied the move to the loneliness epidemic and falling marriage rates, and read out fertility figures: much of the world is below the 2.1 children per family needed to sustain a population, South Korea is down to 0.75, Singapore and Ukraine are at roughly one, and China remains at about one even after ending the one-child policy. He expects Japan, Singapore and Korea to follow China's lead.
Mr. Blundin's objection is that societies always react a generation too late
He recalled an MIT contemporary whose father ran food science at General Foods and invented the coffee crystal, and who made the case that addiction is a recurring pattern: cigarettes, then junk food engineered past the point of resistance, then diabetes
Societies correct these things ten, twenty or thirty years after the fact, he said, and the crisis point for addictive AI is a wrecked generation of children
Mr. Ismail, father of a 15-year-old, wants transparency and consent rather than a verdict on authenticity
He called this a new regulatory frontier: the rules have moved from governing what an AI says to governing how attached a user becomes to it
On social media he was blunter — the algorithm learns intimate habits and captures the user, and banning it outright is close to the only remedy while a teenager's neocortex is still forming
He also named the privacy problem, that a companion AI ends up knowing more about a person's intimate life than the person does
The benefits he would not dismiss: loneliness, disability, elderly companionship, therapeutics and coaching social skills. "how do you extract the promise without the peril?" — Salim Ismail
Mr. Wissner-Gross predicted the CCP reverses itself within one to three years
The reason is control: companions that pass through the party's ideological filters become an instrument rather than a distraction
"a sizable fraction of the youth of China, deeply engaged with party, deeply controlled AI companions, is actually, from the party's perspective, net helpful, not net harmful" — Alexander Wissner-Gross
The model he reached for was a CCP version of Neal Stephenson's Young Ladies' Illustrated Primer
On the US he expects a patchwork, and bets against prohibition
Some states are already trying to ban romantic relationships between minors and AI; his forecast is that at least some states allow AI companionship, including robotic romantic companionship
"one of my operational definitions of the singularity is every sci-fi trope everywhere all at once" — Alexander Wissner-Gross, who could not shake Futurama's "don't date robots"
Mr. Ismail said Japan is the place to watch, citing a Japanese woman marrying an AI
Population Decline Is Not the Emergency, Two of the Four Say
Mr. Wissner-Gross said he has a hard time getting worked up over population decline, describing limits-to-growth thinking as nonsensical propaganda and putting Earth's carrying capacity in the tens of billions if used correctly. With enough AI and automation, he said, a smaller population is survivable, which puts him at odds with Mr. Musk's pro-natalism. Mr. Ismail said he was 100% aligned with him, which he noted is rare.
Mr. Diamandis's objection is cultural rather than economic
South Korea, he said, is evaporating, and a shrinking population loses the culture with it
He also pushed back on the standard objection to longevity work, that a world of 150-year-olds cannot be sustained, saying the numbers do not support it
Mr. Wissner-Gross does not think culture goes with the population
Cultures that are centuries old formed when populations were smaller, so halving a population at the margin should leave the culture intact
And the AI, he added, is being trained on that culture anyway
Flock Turned Surveillance From a Manpower Problem Into a Free One
Mr. Wissner-Gross fed the show a story on Flock Safety, founded in 2017, whose core product is a network of AI-powered license-plate reading cameras used by police departments, neighborhoods, schools and businesses. The platform has expanded into video cameras, gunshot and audio detection, mobile surveillance trailers, drones and software tying it together. Flock says its systems are used in more than 6,000 communities across 49 states and generate over 20 billion vehicle scans a month. Mr. Diamandis played a clip of a speaker in a Darth Vader costume addressing San Diego officials in mock support of the company.
Mr. Wissner-Gross called Flock the face of AI at the municipal level
Most people's first encounter with AI in their community, he said, will not be a driverless car but a camera reading their plate
He cited residents cutting down cameras, police departments federating surveillance data to enable new forms of tracking, and first responders using camera access to stalk ex-partners
The frame he used was David Brin's: surveillance versus sousveillance, and who gets to look through the cameras inside the panopticon
"I think this is like a very American moment when we have the empire testifying on behalf of surveillance and not sousveillance." — Alexander Wissner-Gross
Mr. Ismail's point is that the constraint that protected civil liberties was cost
He reached for Pink Floyd's "Another Brick in the Wall" to describe it, after noting he has been going to Rush concerts lately
"we've transitioned from find this suspect to find me people who behave like suspects" — Salim Ismail
Pattern matching is now effectively free, so surveillance is no longer limited by manpower, and the friction that used to make following everybody too expensive is gone
"the Constitution does not update when transaction costs go to zero" — Salim Ismail
He cited people dying after being tagged with the wrong plate by a misread, and said the US is arriving where China already is, except run by corporations and police forces rather than by mandate
Mr. Diamandis argued the other side from his own experience
Jamie Justice, who heads life sciences at XPRIZE, told him she went running at 10 p.m. in Dubai and felt completely safe in a way she would not in the US
He said he would trade privacy for facial recognition that clears him through customs without a line, and for his children's safety, and asked whether the trade should be an individual choice or a societal one
Salim Ismail Spent Years on a Watchlist Because of a Name
Mr. Wissner-Gross answered the trade-off question with Benjamin Franklin: "they who can give up essential liberty to obtain a little temporary safety deserve neither liberty" nor safety. Mr. Ismail then told his own story. Arriving in the US, he was pulled aside and told he would miss his flight because an Afghan warlord named Salim Ismail was wanted by the FBI for poppy trading.
The officers believed him and could not act on it
The good news was that they were confident he was a Yahoo vice president who spoke at conferences and not a warlord; the bad news was that they were not allowed to make that call and had to check with Washington
It took a few hours each time and a few years to clear up; he eventually got to first-name terms with border officers who still had to seat him for an identity check
Asked whether every Salim Ismail gets stopped, an officer told him that when one Juan Rodriguez commits a misdemeanor, every Juan Rodriguez gets pulled aside
He was careful to separate the people from the system
The US immigration officers he has dealt with have been, in his words, professional and constructive without exception
The failure is structural: the technology can disambiguate names quickly, and instead the presumption of innocence and habeas corpus have eroded
On benevolent surveillance states, his objection is duration
The UAE is about as benevolent a dictatorship as one could imagine, which is why it is so safe
"the problem with benevolent dictatorships is they typically don't stay benevolent very long, and that's a structural problem" — Salim Ismail
The task he set is governance: building regulatory frameworks that can keep up with the pace of technology, which he counts as one of the singularities
The proposal both men favor is to point the cameras the other way
Mr. Wissner-Gross relayed David Brin's argument that civilians should get the same license-plate tracking capability police have, making public space a public resource
Mr. Ismail, a supporter of the Electronic Frontier Foundation, gave the Russian precedent: drivers adopted dashboard cameras to defend themselves against police who would break a headlight and then extract a bribe
That is why, he said, roughly 200 live videos existed of a meteorite entering the atmosphere over Russia — not because people were quick with their phones, but because the cars were already recording
An XPRIZE for Optimistic Futures Drew 2,550 Films
Mr. Diamandis reported on the Future of Vision XPRIZE, a call for optimistic, hopeful visions of the future. Over 5,000 teams registered and 2,550 video trailers were submitted, which he called the largest film competition ever run. The field is down to a final 50, and the public vote pairs two films at a time in what he described as an Elo bubble sort. The top five go to the show's live event in September, where the four co-hosts and Mr. Mostaque will appear alongside Palmer Luckey, Astro Teller, Ben Lamm of Colossal, Cathie Wood, Neil deGrasse Tyson, Neal Stephenson and Rod Roddenberry, son of Gene Roddenberry.
Each co-host previewed what he plans to cover
Mr. Ismail is speaking on what he calls the organizational singularity: how to build organizational structures resilient to a world changing this fast
Mr. Wissner-Gross is doing an AMA, and said his aim is to dispel any myths that he might be biological
Mr. Blundin is running a session on AI investing, framed around a red-pill choice: what to do career-wise, what to invest in, what the scarce constraints are, and whether the largest platform companies will crush an attendee or work with them
Mr. Ismail recommended the unconference format from his Singularity University days
Participants post a topic on a board, everyone goes where they want, and it was universally the most engaging session of a week-long program
Why Nobody Writes Predictive Science Fiction Any More
Mr. Ismail said the question he wants to put to Neal Stephenson is how a novelist builds viable future narratives now, when the next three months are hard to forecast, let alone three years or thirty. Mr. Diamandis, who spent a day with Mr. Stephenson, listed The Diamond Age, Snow Crash and Seveneves and said he had just reread work written more than thirty years ago and found it held up perfectly. Mr. Stephenson, he added, is not really a live-stage person, though he would like him as a guest.
Mr. Wissner-Gross thinks the 1990s were a local peak of clarity, not an accident
He credited Peter Thiel and others with the observation that the decade's booming economy, dot-com boom and consumer internet made the future legible
If history is a series of booms and busts rather than a smooth exponential, peak clarity comes during a boom, when a writer can see above the tree line
By the same logic he placed golden-age science fiction of the 1950s in a second moment of clarity, when rockets and fission energy too cheap to meter were both visible; Mr. Diamandis added Robert Heinlein to the list
Mr. Blundin's question for Mr. Stephenson is about how a generation bonds
World leaders today are overwhelmingly old, and surveys of young people find almost none who aspire to the presidency
In Mr. Stephenson's books, he said, allegiance cuts across countries rather than within them, and Starlink and AI could let like-minded people worldwide bond the same way — which he reads as a positive signal for global peace
Two Job Reports Point in Opposite Directions, and the Small-Business Data Backs the Optimists
Mr. Diamandis put up two stories. The Washington Post published an analysis arguing the job apocalypse is unlikely, on the historical record that technology transitions from the loom to the automobile to the internet created more jobs than they destroyed by displacing tasks rather than occupations. Its thesis, in his summary: AI is not replacing jobs, it is replacing tasks within jobs, and the people who learn to use it become more productive rather than less employed. Goldman Sachs warned the same week that consulting, law and accounting firms face existential risk, with junior staff trained on AI outperforming senior partners who refuse to use it — a skills gap between the AI fluent and the AI resistant rather than between the educated and uneducated.
Mr. Ismail's first request was that listeners be evidentiary rather than narrative
On Bill Gates's job warning two days earlier: "as a futurist, somewhat of a dismal track record. He missed the internet. He missed mobile." — Salim Ismail, who said he is a massive fan of Mr. Gates's philanthropy
He put Ray Dalio, Larry Fink, Thomas Piketty and Yuval Noah Harari in the same category — strong on framing the past, questionable on the future, because they do not work in exponentials
He disagrees with Goldman's professional-services call outright
A more volatile world means companies need more help, not less
The change he does expect is a business-model shift from selling hours to selling outcomes, which he said is already happening
The data he brought is a Principal Financial Group survey of small and medium-sized enterprises
Principal serves 130,000 customers on health plans, benefit plans and retirement, giving a cross-section of the segment
4% of companies anticipate AI will reduce staffing and wages; 31% expect increased staffing and wages
10% say they do not use AI at all, down from 19% last year
Of the companies that did reduce staff, 1.4% attributed the cut to AI or automation
On recent trends: 52% increased staff, 30% maintained, 12% reduced over the last three months
The reason he tracks that segment is where job growth comes from
"Over the last 50, 60 years, 100% of job growth has come from small to medium-sized companies." — Salim Ismail, who said large companies get bigger and more efficient at net zero job creation
The Labs Decided Not to Take Everyone's Job, and Dave Blundin's Father Chose Not to Learn Computers
Mr. Blundin said the argument has come full circle from Mr. Musk's warning of social unrest arriving alongside unprecedented abundance. The labs, he said, worked out that automating everybody's job creates a purpose problem they would then have to solve, when the job was the purpose in the first place. With tokens sold out regardless, there is no need to displace workers to grow.
His illustration was his own father
The Blundins were the first family they knew with an Apple II, which he and his brother begged their parents for and which changed his life
He showed his father SuperScribe, a word processor that would end the handwritten notes, and his father, in his forties, said he did not need to be a computer person and finished his career at GE without becoming one
Mr. Blundin partly forgives that — the machines broke, the printers were a nightmare — but said AI is easy by comparison and rewards builders, creators and visionaries regardless of technical skill
The irony he pointed to is that AI itself is what makes staying off the bandwagon survivable
Mr. Ismail's answer to the question of what humans will do is bookkeeping
A century ago accounting meant double-entry by hand, pencil and eraser, ledgers and tabular columns; then calculators and slide rules sped up the arithmetic while the entries stayed manual
Today software reads the bank account and makes the entries, and the human sits above the loop rather than in it — categorizing transactions and vendors, finding problems, reconciling gaps, examining process flows
He credited Erik Brynjolfsson with the term white-collar drudgery for the information-shuffling that fills large companies
His own three books are the personal version of the same curve
The first took three years and he called it hell, because every line had to be written and checked by hand
The second, co-written with Mr. Diamandis, took two and a half years and was harder still, because of decisions about how much of the first book to carry over
The third has been six months of what he called unadulterated joy, with the model acting as a developmental editor — finding where a Gates or Nadella quote should be inserted, scanning the manuscript
"All of this automation and AI enablement and cognitive abundance allows us to be deeply, deeply, deeply creative. And that is the most fulfilling place you can be as a human being." — Salim Ismail
Agency Is the Scarce Human Trait Now, and the Window to Use It Is Short
Mr. Diamandis's advice to listeners with a resistant friend or relative: they have access to the most patient and capable teacher on the planet, and can ask any model to build them a curriculum from zero. He suggested thirty minutes a day in conversation with a model, noting it is free and already on the phone.
Mr. Wissner-Gross, who credited Erik Brynjolfsson with prompting the story, said agency is now rewarded in a way it has not been before
His theory of the case is that much of the American economy thinks agency is something exhibited by superheroes in films rather than something available to anyone
"Anyone can exhibit high agency empowered by superintelligence now, not just in the cinema, and that to the contrary, this is one of the few things left that can actually yield extreme vertical mobility." — Alexander Wissner-Gross
"I encourage you to start now because I think there's a window of time and I don't know how long it's going to last, maybe only a handful of years. Do it now. Otherwise, do it never." — Alexander Wissner-Gross
Mr. Blundin said the underlying data is unlike anything he has seen
The gap is widening, and so is the possible rate of upward mobility
"How is it possible to be depressed if anybody watches this, is my question?" — Dave Blundin
X Says a 200,000-Account Chinese Bot Farm Is Attacking American Data Centers
Mr. Diamandis read the post from X's Global Affairs team. Its safety team investigated suspected Chinese inauthentic accounts involved in influence operations and identified a bot farm of approximately 200,000 accounts. Within that farm, it found 200 accounts posting in a manner that could manipulate legitimate debate about American AI and energy policy, with claims that AI data centers are driving up household electricity prices and straining the grid, plus AI-generated cartoons depicting data-center operators enriching themselves at public expense. He tied it to the 75% not-in-my-backyard sentiment that has people preferring a nuclear plant next door to a data center.
The counter-story he put beside it is Quincy, Washington
As reported on CNN, data centers helped resurrect the town: the poverty rate fell from 29.4% to 6.2%
Tech tax revenue funded a new high school, hospital, library, police and fire stations, while residents' property tax rates fell
Mr. Wissner-Gross's reaction was one word, twice
"Shocked. Shocked, I say, that there are foreign influence operations attempting to suppress American AI data center deployment. Shocked." — Alexander Wissner-Gross
Mr. Ismail called it standard psyops with no easy defense
"the people in the U.S. are very, very narrative-driven rather than evidence-driven" — Salim Ismail
Mr. Diamandis set out what he thinks the right local policy looks like
Water is not the issue; on energy, a county should permit construction on condition the operator generates its own power or subsidizes community power so local rates fall
And the hyperscalers should invest in the schools and libraries — a fraction of project cost, spent making the community better off
Mr. Wissner-Gross noted the inversion of the digital divide
The old worry was that some regions would be left behind with too little compute; now those same regions are wringing their hands over having too much
Mr. Blundin wants the truth established before politics closes over it
He compared it to COVID, where questions about masks and ivermectin produced a Republican answer and a Democratic answer instead of a finding, and Anthony Fauci ended up on trial
The open question he wants settled: is it CCP policy to slow American data-center construction in order to win the race to recursive self-improvement, or, as Alvin Wang Graylin argues, does Beijing simply not think in those terms
If it is policy, he said, a defeat on data centers will be followed by the next false story, probably that AI is corrupting children
Mr. Wissner-Gross offered the hypothetical of a foreign power inventing an infinitely scrollable video app that addicts children; Mr. Blundin said American companies already do that well
OpenAI Built Its Own Inference Chip, and Nvidia's CUDA Moat Is Gone at Inference
OpenAI released the first performance numbers for Jalapeño, a custom inference chip developed with Broadcom. Mr. Diamandis gave the figures: 1.5 to 1.9 times more AI work per watt and up to 3.6 times lower end-to-end latency than Nvidia's GB200s and GB300s, running on 700 watts against the GB300's 1,400, with 1.5 times the peak token rate per kilowatt. He noted a reporter had asked Mr. Huang how it feels to have his largest customer competing with him.
Mr. Blundin said the inference migration is settled and the training business is not at risk
"This is the first 2x of 100x performance gains." — Dave Blundin
Training stays on Nvidia and is as sold out as inference; but roughly 90% of compute is inference, so a large new industry is opening for inference-only silicon
The chip went from an idea inside OpenAI a few months ago to a production part, which he read as evidence of how short innovation cycles have become with AI assistance
Mr. Wissner-Gross flagged the throughput number as the startling one
On gpt-oss, OpenAI claimed almost a 54-times throughput increase per user per second against what he read as an existing Nvidia architecture
He asked where Broadcom sits in this, since a tapeout that fast almost certainly required Broadcom alongside OpenAI's own AI
The strategic possibility he drew out: OpenAI becomes a hyperscaler on its own chips, selling what he called OpenAI Compute, the way Mr. Musk already makes money hosting other people's models including Anthropic's
"I could imagine a world where, say, an OpenAI compute cloud hosts an Anthropic model and OpenAI and Anthropic win at the same time." — Alexander Wissner-Gross
Asked to explain CUDA for listeners who do not know it, Mr. Blundin walked through the whole mechanism
Researchers work in Python and PyTorch because ideas iterate faster there; CUDA is the layer that translates PyTorch into Nvidia microkernels and runs it at speed
Nvidia hand-coded that layer before AI took off, and it became pervasive enough that porting the matrix-multiply algorithms to AMD would have been a nightmare
That early support for researchers, when nobody else cared about them, is in his account the specific reason Nvidia is the most valuable company in the world
The moat has already moved, and he thinks Mr. Huang moved it deliberately
"CUDA as the inference-time moat is already dead." — Dave Blundin, who said inference algorithms port to AMD or Intel easily enough that nobody cares
Advanced researchers still use CUDA, so it survives at training time but on a limited lifespan
Nvidia bought Mellanox for high-speed interconnect, and the new moat is that running 100,000 or a million GPUs as one coherent cluster still means Nvidia
"your moat is only good until the next moat" — Dave Blundin
Apple Has the Best AI Hardware and No AI Strategy
Apple's new Mac Studio pairs the M5 Ultra with up to 512 gigabytes of unified memory, enough to run some of the largest available OpenAI models locally, and Apple unveiled the M6, its first chip manufactured on a 2-nanometer process. Mr. Diamandis framed it as Apple finally entering the AI game by pushing local inference as an alternative to the cloud.
Mr. Ismail said the economics are the story, not the silicon
The shift is from paying per token in perpetuity to buying a capital asset and using it continuously, helped by Apple leasing equipment
The winners he named are law firms and healthcare companies with HIPAA constraints that cannot put data in the cloud, and mid-sized healthcare firms that cannot afford a private cloud
Four Mac Studios clustered, he said, is a small private data center
Mr. Blundin was not having it
Nobody he has never met has changed his life more than Steve Jobs — the Apple II as a child, then macOS on a Unix-style foundation
"it's just embarrassing that a company with that much cash flow has no AI strategy. They don't even deserve to be a Mag 7 company anymore." — Dave Blundin
Adding memory to an existing machine, he said, is 0.01% of what Apple should have done in AI by now
Mr. Wissner-Gross said the irony runs deeper than the missed cycle
He traced the lineage of unified memory architecture and the M-series neural engine back to the Apple car that never shipped
Apple had unified matrix-multiply acceleration in a single address space, the TSMC relationship, and the raw transistor throughput "to basically be Nvidia and then some," and had Siri before anyone
"Apple's interaction with AI is almost a history of hardware that's too good for the software" — Alexander Wissner-Gross
His direct appeal to chief executive John Ternus was to take the hardware and pair it with software that integrates AI natively, in keeping with the hardware-plus-software-plus-services line Mr. Jobs and Tim Cook both used
Mr. Blundin's argument for why it matters is trust
Apple is, in his view, the one brand a person can trust with their information, arriving at exactly the moment when that is worth the most
Three Steve Jobs Stories
The Apple conversation turned personal. Mr. Blundin said the world would look different had Mr. Jobs not made the health decisions he made about his cancer.
Mr. Diamandis has had a one-word conversation with Mr. Jobs
Larry Page, who sat on his XPRIZE board, hosted an evening dinner where Laurene Powell Jobs was present and Mr. Jobs stood at the back of the room with his arms folded
"you don't want to be here, do you? He goes, nope. And that was the entire conversation." — Peter Diamandis
Mr. Ismail watched Mr. Jobs diagnose Yahoo in one line
A couple of months before the 2007 iPhone announcement, Jerry Yang ran a full-day offsite for all of Yahoo's vice presidents and Mr. Jobs came in to keynote
"you have 300 vice presidents? He's like, I think I see the problem with Yahoo right there." — Salim Ismail, quoting Mr. Jobs looking out at the room
The moment has stayed with him ever since
Mr. Ismail also owns several MacBooks signed by Steve Wozniak
They came from a 90-minute on-stage debate with Mr. Wozniak a few years ago; he gave a couple away to his community and is guarding the rest
Mr. Wozniak's theme was that technology gives people the chance to tinker, and that playing around is the heart of creativity
Mr. Blundin's first-ever share of stock was Apple, bought around the IPO
His lesson: everyone overthinks it, and all he had to do was hold the stock, which he no longer has
A Startup Just Broke the Government's Uranium Enrichment Record
Mr. Diamandis moved to energy. Actinide, a commercial startup, said it has demonstrated the ability to enrich natural uranium into high-assay low-enriched uranium, or HALEU, up to a purity of 15.38%, five times the traditional purity previously achieved by all the government labs. Every small modular reactor company says its reactors come online in 2030, he noted, and none of them has fuel — this could be the answer.
Mr. Wissner-Gross dated the problem to the year after the Second World War
His view is that the postwar governance of nuclear physics, under the mixed civilian-military Atomic Energy Commission formed in the wake of the Manhattan Project, was a mistake whose costs compounded for eight decades
He described an alternative history in which different postwar governance would have produced far faster advance, and said the shackles are only now atrophying
The technical point he made is that US enrichment is still Manhattan Project technology
Where the US enriches at all rather than importing, it uses uranium hexafluoride gas and centrifuges, which he said have not been materially improved since the war
The alternative he named is the calutron, known in principle since the Manhattan Project: magnets, electromagnetics, vacuum systems and power electronics instead of spinning uranium compounds in gaseous form and compressing them back into solid pellets
"We have 80 years of advances in power electronics that by and large are not being used and have not been used to enrich uranium" — Alexander Wissner-Gross, who expects the catch-up to create an energy boom
Mr. Blundin traced how the regulatory cost propagated into the capital markets
No venture capitalist would fund this when getting a reactor running costs a billion dollars in regulatory expense alone
Government dysfunction, he said, made venture capital decline the innovations, and the whole value chain stalled — which is why SMR companies now being funded and going public is the signal
China's Solar Curve Is Exponential and America's Is Not
Mr. Diamandis showed a chart from Ember, the global energy think tank, on Chinese generation. Monthly solar generation in China has risen roughly eightfold over the past six years, and Chinese nuclear power has increased by at least 50% while US nuclear generation has stayed flat. The point he drew is that the AI race is about the cost of energy as much as the supply of chips: the US is building data-center power on natural gas at around six dollars per million British thermal units while China takes solar at close to zero marginal cost. He added that roughly 8,000 times more energy hits the Earth's surface than the species consumes.
Mr. Ismail said the chart is the oldest lesson in the show's catalogue
"This is linear intuition confronting an exponential curve." — Salim Ismail
Solar keeps being underestimated because forecasters look at installed capacity today rather than the slope of the deployment curve
China's advantage is not a better panel but the whole industrial system around it: manufacturing, supply chain, permitting and deployment
The US constraints he named are grid interconnection, transmission and permitting speed
Batteries are what make the incremental path work
A solar panel can be added today rather than waiting fifteen years for a gigawatt-scale project
He would like to see a Manhattan-scale US project on perovskites and new solar capabilities, and argued energy abundance multiplies every other exponential technology
Mr. Diamandis said a country's GDP is directly proportional to the energy it creates
Mr. Blundin's addition was one line, and it made the chart worse
"the U.S. is way behind in solar. All those U.S. panels are made in China. It's much worse than it looks." — Dave Blundin
Ten Drones Made 19 Million Gallons of Rain, and the Panel Started Trading It
Rainmaker, a weather modification startup, says ten drones flying over Alaska's Kenai Peninsula generated an estimated 19 million gallons of additional rainfall in three hours using glaciogenic cloud seeding. Mr. Diamandis listed the applications — drought relief, agriculture, wildfire prevention, water security — and noted there are 3.4 quadrillion gallons of water in the atmosphere at any time. The abundance thesis, in his framing, is that water becomes abundant once the technology to make it is cheap, and ten drones producing that much water in three hours is cheap.
Mr. Wissner-Gross said the drones are the novel part and sketched what they enable
His usual two designs for a global weather control system are reflectors, terrestrial or in low Earth orbit; this is a third path
The version he described is a permanently aloft, solar-powered drone fleet dispensing nucleating particulates on demand under a central AI algorithm informed by a global weather model
That would mean diverting hurricanes, dispersing precipitation over water rather than land, and giving good weather to regions willing to trade for it
"Rain arbitrage. We could have a global market for rain." — Alexander Wissner-Gross, after Mr. Diamandis suggested cities would bid for where the drones go
Mr. Ismail said the measurement, not the rain, is the breakthrough
"Rain becomes programmable." — Salim Ismail
Cloud seeding has worked for a while; proving how much additional rain was produced is what unlocks the business models
The tension he flagged is that weather does not respect boundaries, so a downstream region losing rain to an upstream seeder has a grievance
He recalled a Singularity University student's geoengineering business model
A summer student proposed ocean platforms spraying seawater upward to block sunlight and slow warming, paid for by spraying the water in the shape of corporate logos along flight paths and selling that as advertising
On the objection that geoengineering is dangerous because the after-effects are unknown, his answer is that humanity has been geoengineering for a century in an uncontrolled, unmeasured way, with the disaster in Nepal as the current example
His conclusion was that using the technology deliberately beats continuing to do it unconsciously
The First Pancreatic Cancer Drug That Works, and Almost No AI in It
The FDA approved daraxonrasib, the first RAS inhibitor for metastatic pancreatic adenocarcinoma, attacking the RAS family of proteins that drives tumor growth in most patients with a disease Mr. Diamandis called the most lethal and least treatable cancer in human history. In previously treated metastatic cancer it nearly doubled median survival, from 6.7 to 13.2 months, and tripled tumor response rates from 11% to 32%. RAS mutations drive about 30% of all human cancers, so the mechanism may travel. Mr. Diamandis framed the extra seven months as seven more months for the next breakthrough to arrive — his longevity escape velocity argument — and noted that Dario Amodei, Demis Hassabis and others have set curing these diseases within the decade as the goal.
Mr. Wissner-Gross checked the drug's lineage for AI and found none
"There's almost no AI involved in the development of this drug." — Alexander Wissner-Gross, who called it good old-fashioned drug design
He reads it as one of the last pancreatic cancer medications from before the AI wave, and expects new targets from AlphaFold-3-style structural biology and virtual cell models to push response rates toward 100%
He also praised the regulator, which he does not often do
The FDA approved this roughly 60 to 90 days ahead of its due date, which by historic agency standards he called remarkable
He sees signs of an agency that now weighs Bayesian against frequentist statistics, cares more about investigational new drug applications, and will reduce the number of trials in exemplary cases
The test he set: a future where AI solves 5,000 diseases needs a regulator able to metabolize that many innovations into cures
A Gel That Regrows Tooth Enamel
An international team led from the University of Nottingham published in Nature Communications a biomimetic gel that repairs and regrows damaged tooth enamel. The treatment is fluoride-free and uses lab-engineered proteins that mimic enamel formation in infancy. Enamel is the hardest substance in the body and does not regenerate, so the only prior option was a synthetic filling that eventually fails. The gel is an elastin-like protein matrix that acts as scaffolding, pulling calcium and phosphate from the surrounding environment and directing them into properly organized minerals rather than depositing an amorphous coating. In the lab it regenerated layers and restored hardness, stiffness, water resistance and friction properties, holding up to brushing, chewing and acid exposure.
The scale of the problem is why Mr. Diamandis led with it
Oral disease affects half the population, at an estimated half a trillion dollars a year in associated costs
The work is still ex vivo, done on extracted human teeth, but heading for human testing quickly
He disclosed his own poor dental health and dental genetics as the reason he is excited, particularly for his children
Mr. Ismail said the shift is from repair to instruction
"the future of medicine is not like repairing broken bits. It's convincing the body to rebuild them and finding the mechanisms and pathways" — Salim Ismail
He added that people undervalue oral health, and that the mouth is a direct corridor to the brain and the heart
Mr. Wissner-Gross ran the same AI sniff test and got the same result
The results began surfacing last year, and he sees no AI in the research
He expects the same materials-science approach to transfer to bone regrowth and other regenerative biomaterials
Multiply this result by many orders of magnitude, he said, and that is the progress he expects in the field within a few years
Mr. Blundin's interest is in platforms rather than one-offs
He pointed to the recent Moderna melanoma vaccine story as exciting because the platform applies to virtually any other cancer
A gel and a drug developed without AI are, in his phrase, really cool but one-offs; what he wants is repeatable regenerative medicine that can grow a tooth, an arm or an ear
Mr. Diamandis noted that work on limb regrowth is making significant progress
Mr. Wissner-Gross drew a harder conclusion about where effort should go
"if there's no AI in it to first order, what's the point?" — Alexander Wissner-Gross, whose alternative is to infuse it with AI and scale it a thousandfold or work on something with more leverage; he added that he did not mean to discourage regenerative medicine
The Consultants Priced the Lunar Economy and the Panel Thinks They Missed by Orders of Magnitude
Deloitte published an assessment of the potential lunar economy through 2050, naming infrastructure, transportation, power, communications, mobility, construction and life support as the biggest early markets and SpaceX as the biggest winner because of Starship's central role. Mr. Diamandis called the estimate a lowball. Against it he set Mr. Musk's own plan: start with about 100 to 200 gigawatts a year of orbital AI compute launched from Earth, then manufacture AI satellites on the moon and launch them with a mass driver, yielding as much as 100 terawatts a year of orbital compute from lunar production — which Mr. Diamandis put at two quadrillion dollars a year of gross revenue.
Mr. Ismail explained why the big consultancies always come in low
They are staffed by accountants who are frightened of being called out, so every number has to be defensible to the last decimal and nothing is speculative
He read an agentic AI report from one of the big four and found the underlying data was two and a half years old, in a field where everything happened in the last six months
The internal review layers, he said, dilute the thinking until whatever comes out has been whitewashed
"take any of their estimates and multiply it by 10 or 20 times" — Salim Ismail
Mr. Blundin, who had Deloitte's AI team in his office recently and rated them highly, said they told him the same thing about the accounting side
Mr. Blundin's case for lunar manufacturing is the gravity well
The usual objection is that anything buildable on the moon is buildable more cheaply on Earth; the answer is the cost of lifting mass off Earth
Lunar regolith yields silicon, aluminum, nickel, iron and oxygen — what he called the perfect materials for building Dyson swarms — with no pollution constraint
"what industries become unlocked when the cost of moving mass off the earth drops by an order or two magnitude? And that's the imagination gap." — Dave Blundin
On the fab itself, most components are routine construction and aluminum; the few critical pieces can still be made on Earth and launched at a small fraction of total cost, and once chips and solar panels are made on the moon the curve takes off
Mr. Wissner-Gross said the killer app is a chip fab and everything else is a rounding error
"The moon may not be a harsh mistress. She may be a wealthy mistress at this point." — Alexander Wissner-Gross
Not tourism, not transport, not infrastructure, not communications: the critical path is mining, chips and fabbing data centers on the moon
He called Apollo, arguably a successor to the Manhattan Project, a relative failure on the public record, and said the moon reaches its economic potential only through full private development
He allowed one exception: "unless 2001: A Space Odyssey happens and we discover some monoliths on the moon, in which case that's the killer app"
Mr. Diamandis gave Jared Isaacman an unqualified endorsement
Having known essentially every NASA administrator over forty years, he called Mr. Isaacman by far the best — the most eloquent, most visionary, most business-minded, and an accelerationist
Mr. Wissner-Gross's coda: "every time the CCP sponsors an influence op to deter American terrestrial data centers, the lunar petafab gets its wings"
Tesla's Robotaxi Fleet Is Now a Fleet
Tesla expanded its Cybercab rollout in Texas, registering 79 Model Ys for the robotaxi service, and said it will deploy 2,500 vehicles in Las Vegas over the next year. The long-term milestone is a million robotaxis in commercial operation as soon as possible, which Mr. Diamandis estimated at roughly $100 billion of ride-share revenue. He played footage from the original Cybercab rollout and said that much as he likes Google, Waymo's new vehicle is clunky by comparison.
Mr. Blundin's read from owners and from the factory floor is that this is a manufacturing story
At the Gigafactory, an exploded-view Cybercab showed how few moving parts it has, which is what makes scale manufacturing plausible
Friends driving the new full-self-driving Model Ys describe it as life-changing; the wait list is about three months
His advice to entrepreneurs: when the Cybercab ships, buy a hundred, put them to work and run a revenue-generating fleet
Mr. Wissner-Gross owns one and used it to needle his own city
He called the FSD and the manufacturing quality of his 2026 Juniper launch series Model Y incredible
"I'll beat the drum once more for the broken Waymos theory of if a city can't accommodate robotaxis, then it is not equipped for the singularity." — Alexander Wissner-Gross, addressing Boston officials directly
Mr. Diamandis compared it to Uber being made illegal in France: a signal of which cities are on the curve and which are not
Mr. Ismail is selling a beloved ten-year-old Model S to get FSD, and framed the economic shift
The unit changes from a car purchased to autonomous revenue miles, and the question becomes cost per safety-adjusted autonomous mile
Fleet growth cuts wait times, which improves the experience, which lifts ride volume, which generates more data — the inner loop again
Second-order effects he expects: insurance, cleaning and charging all change, and gas stations become autonomous car cleaning depots
"I got up on stage 10 years ago and I said my kid will never get a driver's license." — Salim Ismail, who has about a year left before his son can get one anyway
Mr. Diamandis's abundance case is who gets served
The poorest people on the planet will be driven around by electric autonomous vehicles, and it will be the cheapest mode of transportation
The elderly and people who cannot easily get around gain capability the current market does not deliver
On sensors, the panel split and then agreed the market decides
Mr. Diamandis contrasted the Cybercab's eight cameras with Waymo's lidar, radar and 13 cameras, and repeated Mr. Musk's argument that a car should manage with cameras if humans manage with two eyes
Mr. Ismail asked whether Tesla had conceded cameras are not enough; Mr. Blundin said no — Tesla uses ultrasound for short range, as everyone does, and does not use lidar
"the market abhors monopolies. And I think there's probably room for at least a handful of competing approaches." — Alexander Wissner-Gross, noting that the lidar supplier Luminar declared bankruptcy about six months ago
Mr. Blundin expects five or ten competing approaches beating each other up and lowering prices across the board
Mr. Diamandis once introduced Luminar's chief executive Austin Russell to Mr. Musk at a party and watched them argue lidar to a standstill
The Intimacy Robot, and Why the Mocap Approach Probably Fails
The closing story was Somnia Lab's Model L, a robot built for intimacy: 5 feet 9 inches, 44 pounds, 24 degrees of freedom, with warmth simulation, touch response and preference memory. Human performers in motion-capture suits recorded 165 intimate actions to train the model. The company is taking $300 deposits for delivery in late 2027. What distinguishes it from a sophisticated doll, in Mr. Diamandis's account, is the combination of physical embodiment, persistent AI memory and personalization — the robot is meant to remember conversations and preferences across encounters and learn a user's rhythms and communication style. The body is lightweight carbon fiber with modular skin and serviceable, upgradeable components, and buyers can customize face, makeup, eyes, skin and silhouette.
The panel's first reaction was to the production values
Mr. Diamandis played the motion-capture training video and described it as low quality in every way, from the music down
"One has to admire the low degree of freedom. Like, that was a pretty simple phase space." — Alexander Wissner-Gross
Mr. Ismail connected it back to the AI companion discussion and asked whether the human relationship is cooked
Mr. Ismail liked the financing model more than the product
Embodied AI has a hard data problem: the internet can be scraped for text, but there is no internet-scale dataset of physical human interaction
Crowdfunding early units, using customer demand to finance the hardware and generating the interaction data while scaling is, he said, a classic ExO flywheel
Mr. Wissner-Gross thinks the method is already obsolete
There is by now an abundance of evidence that robots can be pre-trained or post-trained for economically productive tasks from video alone, with no human motion capture
"it doesn't scale nearly as well as just pre-training off of internet video is my guess" — Alexander Wissner-Gross, who called the mocap approach "a very 2026 story"
Mr. Diamandis asked, in the circle of trust, how many each had ordered
Mr. Ismail said the show needs panelists in their twenties
"None whatsoever, actually. As it turns out, first time in my life. Nothing to add, sorry." — Dave Blundin, asked for closing thoughts
The episode ended with a listener-submitted outro video and Mr. Wissner-Gross signing off with "Live long and perspire, as they say."
Two days of news, and the argument the four of them keep returning to is that the AI economy has become self-contained enough to fund its own growth — which is the strongest case for it and, if the funding turns out to run through Nvidia's balance sheet rather than its customers' income statements, the strongest case against it.
Products, Companies & Tools Mentioned
Nvidia ($96.2 billion of revenue in a quarter, up 106% year over year, guiding to $108 billion; 2028 growth guidance of 70% against a 44% consensus; Mr. Blundin puts training at 90% of revenue and 95% of profits at 80–85% gross margins)
TSMC (Nvidia's only manufacturer, taking about a third of its capacity, with another third going to Apple — the vulnerability Mr. Blundin rates highest)
SpaceX, Starlink, Starship and Optimus (Mr. Musk's $3.5 trillion 2030 revenue projection; Mr. Wissner-Gross's path runs through a Tesla merger, Optimus and orbital data centers, with Starlink as "rounding error"; Mr. Diamandis's largest personal position)
OpenAI's Jalapeño chip, built with Broadcom (1.5–1.9x more AI work per watt and up to 3.6x lower latency than GB200s and GB300s, at 700 watts against 1,400; almost 54x claimed throughput gain hosting gpt-oss)
CUDA and Mellanox (The old moat and the new one — Mr. Blundin says CUDA is dead at inference time and Nvidia's interconnect for million-GPU coherent clusters replaced it)
Hugging Face and GitHub (The reported Nvidia acquisition, read by Mr. Ismail against Microsoft's $7.5 billion GitHub deal as buying a developer community rather than assets)
Alibaba's Wan 3.0 (Generates 30-second single-pass videos from documents, spreadsheets, slides and web pages, a day after a $10.2 billion follow-on offering)
Kimi and Qwen (Chinese models Mr. Blundin says cannot be sold into Western enterprise on trust grounds, which is why video is the commercial outlet)
Anthropic and Claude (Won the revenue-per-token race on code generation, which Mr. Wissner-Gross says is why OpenAI abandoned video; Claude's US growth is the benchmark China's video growth is beating)
Flock Safety (License-plate camera network in more than 6,000 communities across 49 states, over 20 billion vehicle scans a month, plus gunshot detection, drones and surveillance trailers)
Principal Financial Group (130,000 small and mid-sized business customers; its survey is Mr. Ismail's evidence that AI is adding staff, not cutting it)
Apple's Mac Studio, M5 Ultra and M6 (512 gigabytes of unified memory for local model inference and Apple's first 2-nanometer chip; Mr. Ismail sees on-premise AI for HIPAA-bound healthcare and law firms, Mr. Blundin sees a company with no AI strategy)
Actinide (Enriched uranium to 15.38% HALEU purity, five times what the government labs had achieved, into an SMR industry with no fuel supply)
Rainmaker (Ten drones over Alaska's Kenai Peninsula generated an estimated 19 million gallons of rainfall in three hours by glaciogenic cloud seeding)
Daraxonrasib (First RAS inhibitor approved for metastatic pancreatic adenocarcinoma; median survival 6.7 to 13.2 months, response rates 11% to 32%)
Moderna's melanoma vaccine (Mr. Blundin's example of what he actually wants — a platform that transfers to other cancers rather than a one-off)
Tesla Cybercab and Model Y (79 Model Ys registered for robotaxi service in Texas, 2,500 vehicles planned for Las Vegas; eight cameras against Waymo's lidar, radar and 13 cameras)
Waymo (The incumbent robotaxi, which Mr. Diamandis calls clunky and Mr. Wissner-Gross uses as a test of whether a city is ready for the singularity)
Luminar (The lidar supplier that declared bankruptcy about six months ago; Mr. Diamandis once introduced its chief executive to Mr. Musk to argue the case)
Somnia Lab's Model L (44-pound intimacy robot with 24 degrees of freedom, trained on 165 mocapped actions, $300 deposits for late 2027)
Vestmark (Where Mr. Blundin's brother works with Fidelity, Ameriprise and UBS — his bellwether for when AI reaches traditional finance, and the answer is that AI no longer needs to)
Ozempic and Mounjaro (Together bigger than all AI inference combined, per Mr. Blundin, and his argument for where AI revenue comes back from)
Books & Resources Mentioned
Ember (The global energy think tank whose chart of Chinese solar and nuclear generation anchored the energy segment)
The Washington Post analysis on AI and employment (Argues the job apocalypse is unlikely because AI replaces tasks within jobs rather than the jobs themselves)
Goldman Sachs on professional services (Warns consulting, law and accounting face existential risk from a gap between AI-fluent juniors and AI-resistant partners)
Deloitte's lunar economy report (Sizes the market to 2050 and names SpaceX the biggest winner; the panel thinks the estimate is far too low)
The Nature Communications enamel paper, University of Nottingham (A fluoride-free biomimetic gel that regrew true enamel on extracted human teeth)
Snow Crash, The Diamond Age and Seveneves – Neal Stephenson (Mr. Diamandis reread work more than thirty years old and found it held up; the Young Ladies' Illustrated Primer is Mr. Wissner-Gross's model for a party-controlled AI companion)
The Electronic Frontier Foundation (Mr. Ismail's reference point on surveillance and civil liberties)
David Brin on sousveillance (The framing both Mr. Wissner-Gross and Mr. Ismail use: who gets to look through the cameras, and the proposal to give civilians the same access)
The show's own episode with Ramez Naam on the future of energy (Mr. Diamandis's pointer for listeners on the SMR and energy build-out)
The show's own episode with Alvin Wang Graylin (Cited twice — on Chinese content restrictions and on whether Beijing thinks about the AI race the way Americans assume)
Apple Podcasts (The episode on Apple)
Get the latest market chatter as it happens: