Blueprint

Is AI a bubble? Capex, financing and demand

What the hosts and three investor guests say about whether the AI build-out is a bubble, how it is being financed, and why they keep separating demand from profit. Mostly opinion and relayed reporting, with no agreed trigger for a bust.

Type
Blueprint
Confidence
Medium
Sources
24 videos
First published
Updated

Voice and stakes

Voice: mixed (Sam ~30 %, Jess ~25 %, Dave ~15 %, plus guests Josh Wolfe, Scott Stanford and Matt Mazzeo)

Stakes. Sam says he is partly a beneficiary of the AI narrative cycle (Sam · Fire Sam Altman, The End of So (Mar 2026) ▶ 39:24) and that the party ending would make him poorer because the numbers are correlated (Sam · The Internet Is Fighting Back (Oct 2026) ▶ 11:44). Jess says she is long Anthropic and OpenAI while criticising executives (Jess · Matt Mazzeo (Sep 2026) ▶ 30:50), and much of her capex and financing material is reporting from The Information, her own publication (Jess · Why the Met Gala Hates Tech, E (May 2026) ▶ 34:31). Matt Mazzeo says his company is a huge customer of Anthropic and OpenAI (Matt Mazzeo · Matt Mazzeo (Sep 2026) ▶ 21:45). Josh Wolfe says where you stand depends on where you sit on the cap table, and that his firm Lux is a large investor in Hugging Face (Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 22:29).

The argument in brief

  1. Nobody on the show says flatly that there is a bubble. The hosts offer indicators and scenarios, mostly as opinion (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 38:51). Brit is the exception in tone: “the bubble of all bubbles” (Apr 2026) (Brit · Allbirds Pivoted to AI Data Ce (Apr 2026) ▶ 36:17).
  2. Sam’s frame is “three conversations that can all be true”: demand will be enormous, profit is unclear, and the build-out pattern-matches the railroad and power booms and their collapses (Sep 2026) (Sam · Matt Mazzeo (Sep 2026) ▶ 23:07). See Three conversations that can all be true.
  3. Sam and Jess both insist that demand is not profit: huge usefulness does not guarantee a business (Sam · Matt Mazzeo (Sep 2026) ▶ 23:22), and infinite demand does not mean infinite sustainability (Jess · Matt Mazzeo (Sep 2026) ▶ 31:02).
  4. Sam’s “triangle trade” says Washington needs GDP growth to cope with the debt, Silicon Valley offers AI, and New York and Riyadh fund it, so all three need the story to hold (Mar 2026) (Sam · Fire Sam Altman, The End of So (Mar 2026) ▶ 38:59). This is one host’s thesis repeated across episodes, and nobody argues the other side on air.
  5. Financing is where the sharpest outside voices go. Josh Wolfe says chips depreciate over about three years while the debt amortises over ten (Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 31:45); Scott Stanford says the risk becomes systemic only if both price and demand collapse (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 30:33).
  6. Dave gives the only dated forecast: a market retraction in 2027 or 2028, after the IPOs (Aug 2026) (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:37). It is his prediction, not a finding.
  7. No one names a falsifiable trigger. Sam says nobody knows what breaks the narrative, because they would trade it if they did (Sam · Fire Sam Altman, The End of So (Mar 2026) ▶ 40:17).

Capex: announcements versus money in the bank

Jess states the working rule: massive capex press releases are very different from money in the bank, and she relays that heavy Texas rain delayed data-centre projects by almost a quarter (Dec 2025) (Jess · Hype vs Reality: Our 2026 Pred (Dec 2025) ▶ 15:46). Sam turns it into a question about the whole boom: whether the capital being put in now exists (Sam · Matt Mazzeo (Sep 2026) ▶ 24:28). His own answer in the same episode is “who knows” (Sam · Matt Mazzeo (Sep 2026) ▶ 24:32).

(Sam · Matt Mazzeo (Sep 2026) ▶ 24:28)

The commitments the hosts relay are large and partly locked in. Jess reports that The Information’s reporting puts Anthropic’s latest Google cloud commitment at 200 billion (currency unstated), and that OpenAI and Anthropic together account for about half of the hyperscalers’ backlog (May 2026) (Jess · Why the Met Gala Hates Tech, E (May 2026) ▶ 34:31). Someone on the show says a leaked Anthropic S-1 shows over half a trillion dollars in mostly non-cancellable contracts over ten years (Oct 2026) (one of the hosts · Adrian (Oct 2026) ▶ 14:15); a later episode gives the figure as 580 billion (one of the hosts · ⁠The Real Reason Elon Wants Sp (Oct 2026) ▶ 22:06). Neither number is checked. Jess also reports that SoftBank is struggling to raise the money it promised OpenAI (Jun 2026) (Jess · Apple’s New Siri, SpaceX’s $30 (Jun 2026) ▶ 16:31).

Two items are about concentration rather than size. Sam argues that putting $200 billion in one data-centre complex, Meta’s Hyperion, is a national-security problem, “like lining up 10 aircraft carriers” (May 2026) (Sam · Google I/O Just Killed SEO? Ag (May 2026) ▶ 10:19). Brit wonders if everyone has become inoculated to trillions after the Saudi commitment announcement (Nov 2025) (Brit · OpenAI vs Google vs Meta: Busi (Nov 2025) ▶ 13:23).

On the front line, Dave sees the opposite of a glut. He says everyone is compute constrained and that he sees no fall in token demand, only bottlenecks (Jul 2026) (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 34:29). A month later he predicts the pullback anyway, because three to four trillion dollars of capex has gone in without the use cases to absorb it (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:37). That tension is inside Dave’s own statements, not a debate between two hosts. He also says he cannot comment on specific compute contracts (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 34:03).

Financing: SPVs, duration and who pays

Most of the detail comes from one August 2026 investor panel, Peter Deng & Rachel Holt (Aug 2026), hosted by Sam with Josh Wolfe and Scott Stanford. Scott Stanford describes Nvidia convening lenders into special-purpose vehicles that borrow against take-or-pay agreements from the customers of neo-clouds (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 29:10), a facility he puts at about half a trillion dollars (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 28:54). He compares it with Lucent’s vendor financing, when about $8 billion went into financing fibre customers and a 100x technology improvement then collapsed demand and pricing (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 28:11). He says he cannot remember Lucent’s exact spend (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 28:14). Josh Wolfe says the difference is that Nvidia takes no balance-sheet risk (Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 31:17); Stanford adds that the onus falls on pension-fund-backed lenders (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 29:51). See Take-or-pay SPV structure.

Wolfe’s own rule is to match the useful life of an asset to the duration of its financing.

(Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 31:56)

Duration mismatch in compute financing sets chips at about three years against ten-year debt and power contracts of 10 to 20 years (Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 31:45). Stanford’s “double collapse” test says the structure turns systemic only if price and demand both fall (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 30:24), as in the mortgage crisis; otherwise margins compress (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 30:33). Sam concedes he does not know the answer on asset lives and may be “talking in half truths” (Sam · Peter Deng & Rachel Holt (Aug 2026) ▶ 34:04).

The host-only version of the same worry is Dave’s list of questions: is debt-market activity still growing, what is the real depreciation curve for a GPU, and is the business on top growing as fast as the infrastructure (Nov 2025) (Dave · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 40:38). See Dave’s questions on AI infrastructure debt. Jess adds a half-trillion-style caution of her own: she says about half of AI financings are circular supplier-investment deals, an unsourced estimate (Jan 2026) (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 23:40). Asked from the floor who will pay for data-centre debt, a Blackstone executive reportedly answered “we don’t know” (Sep 2026) (Jess · Matt Mazzeo (Sep 2026) ▶ 25:08).

Demand is not profit

Jess hosted a Davos panel on AI infrastructure and reports that her hard questions on circular deals and efficiency were answered with “infinite demand” (Jan 2026) (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 23:10). Her reply is that limitless demand does not give a healthy margin; media showed that on the internet (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 25:31).

(Jess · Matt Mazzeo (Sep 2026) ▶ 31:02)

Sam goes further: two things can be true at once, AI is wonderful to build software with and a bad business (Sam · Elon Musk vs OpenAI Trial, Goo (May 2026) ▶ 27:51). He says all the big labs are bad businesses even if they trade well and are great technology (Jul 2026) (Sam · Stripe Is Trying To Buy PayPal (Jul 2026) ▶ 31:44). Dave’s version is that not every lab will find product-market fit, as with the ZIRP-era SaaS startups (Aug 2026) (Dave · Jeff Dean Leaves Google, Airta (Aug 2026) ▶ 31:02). Sam also offers a motive for the “infinite demand” talk: infrastructure leaders either truly believe, or the cost-benefit favours going all in because being wrong is cheap (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 27:52). That is speculation about their motives, from one episode (Two ways to be red-pilled on AI and the asymmetry card).

Why everyone needs it to work: the triangle trade

Sam says the US needs AI to hit the GDP required to deal with the debt (Jan 2026) (Sam · Founders Beware: Illiquid Valu (Jan 2026) ▶ 21:32), and that “minus AI” the country would be in recession (Dec 2025) (Sam · Paramount vs Netflix: Who Real (Dec 2025) ▶ 55:15). In October 2026 he says DC needs 8 % GDP growth it does not have, so the build-out must work (Sam · Adrian (Oct 2026) ▶ 15:12); the 8 % figure is unsourced (Sam · Adrian (Oct 2026) ▶ 15:12). His shorthand for the stance is “pull out all the stops, IPO everything and pray”.

(Sam · Paramount vs Netflix: Who Real (Dec 2025) ▶ 56:15)

He applies the same political-economy lens to a market wobble: investors had treated the build-out as a free upside option while a strong Trump would “make sure we get paid”, and the midterms weakened that (Nov 2025) (Sam · OpenAI vs Google vs Meta: Busi (Nov 2025) ▶ 33:12). See The triangle trade and Political-economy model of AI valuations.

Half-joking indicators

The hosts’ own bubble gauges are soft. Sam’s rule is that the further the 101 billboards drift from a product your mother would recognise, the likelier a bubble (Dec 2025) (Sam · Paramount vs Netflix: Who Real (Dec 2025) ▶ 52:02), and late-cycle talk is about money raised, not customers (Jun 2026) (Sam · Why SpaceX’s $1.75T IPO Makes (Jun 2026) ▶ 31:58). His more serious one is the second derivative: the rate of new debt and monetisation, not spending already committed (Nov 2025) (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 38:51). Treat the billboard and rent items as jokes (Jess · Paramount vs Netflix: Who Real (Dec 2025) ▶ 54:07).

How the view moved

  • Nov 2025. Sam says the spending ship has sailed and the question is new debt and monetisation (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 38:54); he sees a very good chance of a 30 to 40 percent Nasdaq or Mag 7 drawdown, stated as a hedge (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 21:03). A week later he explains the AI stock fall as politics, not sector news (Sam · OpenAI vs Google vs Meta: Busi (Nov 2025) ▶ 33:16).
  • Dec 2025. Sam says minus AI the US is in recession (Sam · Paramount vs Netflix: Who Real (Dec 2025) ▶ 55:15).
  • Jan 2026. Davos: infrastructure leaders answer with infinite demand (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 23:10); Sam says history’s “Spidey sense” says their prognostications are wrong, though they might be right (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 26:15).
  • Feb 2026. Sam thinks the multi-trillion-dollar deal environment feels unstable but each step is logical enough that the party probably continues (Sam · Elon’s Trillion-Dollar Move, W (Feb 2026) ▶ 13:21).
  • Apr 2026. Brit: “the bubble of all bubbles” (Brit · Allbirds Pivoted to AI Data Ce (Apr 2026) ▶ 36:17).
  • May to Jun 2026. Capex reporting dominates: the Google commitment (Jess · Why the Met Gala Hates Tech, E (May 2026) ▶ 34:31) and SoftBank’s fundraising difficulty (Jess · Apple’s New Siri, SpaceX’s $30 (Jun 2026) ▶ 16:31).
  • Jul to Aug 2026. Dave sees only bottlenecks, then predicts a 2027 or 2028 retraction (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 34:29) (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:37). The investor panel supplies the financing vocabulary.
  • Sep 2026. Matt Mazzeo says there will certainly be a bubble that pops, driven by over-investment in infrastructure (Matt Mazzeo · Matt Mazzeo (Sep 2026) ▶ 22:12); Sam says the real danger is capital loss, not company survival (Sam · Matt Mazzeo (Sep 2026) ▶ 23:53).
  • Oct 2026. Jess reports infrastructure IPOs slowing and private credit pulling back, and predicts Nvidia will be left as investor of last resort (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 09:01) (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 09:28) (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 09:32).

Where the hosts disagree

  • Is tech debt a worry? Dave says that if debt keeps rising, revenue lags and depreciation is faster than it looks, it is a serious problem (Dave · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 41:06). Jess relays editors who covered the financial crisis saying tech debt-to-net-cash ratios are still very conservative (Jess · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 40:05).
  • Will the party continue? Sam: probably, step by step (Sam · Elon’s Trillion-Dollar Move, W (Feb 2026) ▶ 13:21). Dave: a retraction in 2027 or 2028 (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:37).
  • Is it too early to judge lab margins? Sam calls the labs bad businesses (Sam · Stripe Is Trying To Buy PayPal (Jul 2026) ▶ 31:44). Jess says it is too early to call the margin question and that Sam’s competitive dynamic is stronger than industry people admit (Jess · Anthropic’s Bet on Coding Is W (Mar 2026) ▶ 24:20).
  • Is the financing structure dangerous? Wolfe stresses the duration mismatch (Josh Wolfe · Peter Deng & Rachel Holt (Aug 2026) ▶ 31:45); Stanford says it takes a double collapse (Scott Stanford · Peter Deng & Rachel Holt (Aug 2026) ▶ 30:33).

What could make this wrong

  • Predictions here are the hosts’ opinions, and the pullback timing is unknown (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:48). Dave’s debt and depreciation points are open questions, not findings (Dave · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 40:56).
  • The triangle trade rests on one person, with no counter-view on the show and an unsourced 8 % figure (Sam · Adrian (Oct 2026) ▶ 15:12).
  • Dave admits he does not understand what the market is doing (Dave · Meta’s AI Campaign, While Riva (Jul 2026) ▶ 33:42). Sam says the asset-life question may be above his paygrade (Sam · Peter Deng & Rachel Holt (Aug 2026) ▶ 32:32).
  • Figures are rounded, relayed or unsourced: the 3 to 4 trillion capex total (Dave · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 49:14), the half-trillion contracts (one of the hosts · Adrian (Oct 2026) ▶ 14:15), and the 50 % circular-deals share (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 23:40).
  • Jess concedes the CoreWeave chief executive could be right that demand is limitless (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 25:11).

How strong is the evidence

Thirty-eight episodes touch the topic, but the weight sits in a few: Google Just Won AI, TikTok’s “ (Nov 2025) (Nov 2025), Agentic AI Will Break the Inte (Jan 2026) (Jan 2026), Peter Deng & Rachel Holt (Aug 2026) (Aug 2026, the only deep financing discussion) and Matt Mazzeo (Sep 2026) (Sep 2026). Sam supplies the frame, Jess the reporting, Dave the dated forecast and the debt questions. Three outside investors appear in two episodes. Financing detail rests mostly on one panel; the triangle trade on Sam alone. Where the hosts do agree (demand is real, profit is unclear) it is a shared stance, not independent confirmation, and all four hosts are investors or journalists with stakes in the story. Overall: medium.

Go deeper

Sources

  1. (01)

    Why the Met Gala Hates Tech, Elon vs OpenAI Drama, and the Rise of Chinese AI Models

    YouTube

    34:31 into “Why the Met Gala Hates Tech, Elon vs OpenAI Drama, and the Rise of Chinese AI Models”

  2. (02)

    Stripe Is Trying To Buy PayPal For $150b | OpenAI Cyber Incident, Frontier Model Economics, Peptides

    YouTube

    31:44 into “Stripe Is Trying To Buy PayPal For $150b | OpenAI Cyber Incident, Frontier Model Economics, Peptides”

  3. (03)

    OpenAI vs Google vs Meta: Business Model War

    YouTube

    13:23 into “OpenAI vs Google vs Meta: Business Model War”

  4. (04)

    Founders Beware: Illiquid Valuations vs. Real‑World Taxes

    YouTube

    21:32 into “Founders Beware: Illiquid Valuations vs. Real‑World Taxes”

  5. (05)

    Elon Musk vs OpenAI Trial, Google Cloud Surge, Meta’s Blocked Acquisition, Anthropic Winning

    YouTube

    27:51 into “Elon Musk vs OpenAI Trial, Google Cloud Surge, Meta’s Blocked Acquisition, Anthropic Winning”

  6. (06)

    Allbirds Pivoted to AI Data Centers | SpaceX $1.5T IPO, Anthropic $800B, $175M Seed Rounds

    YouTube

    36:17 into “Allbirds Pivoted to AI Data Centers | SpaceX $1.5T IPO, Anthropic $800B, $175M Seed Rounds”

  7. (07)

    AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo

    YouTube

    21:45 into “AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo”

  8. (08)

    Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)

    YouTube

    23:10 into “Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)”

  9. (09)

    Meta's AI Campaign, While Rivals Ask Washington To Slow it Down | Lilian Weng Saga, China Cracks DUV

    YouTube

    33:42 into “Meta's AI Campaign, While Rivals Ask Washington To Slow it Down | Lilian Weng Saga, China Cracks DUV”

  10. (10)

    Paramount vs Netflix: Who Really Wins Warner Bros And What's Next?

    YouTube

    52:02 into “Paramount vs Netflix: Who Really Wins Warner Bros And What's Next?”

  11. (11)

    Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative

    YouTube

    38:59 into “Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative”

  12. (12)

    The Internet Is Fighting Back Against Your AI Agent

    YouTube

    9:01 into “The Internet Is Fighting Back Against Your AI Agent”

  13. (13)

    Stripe Buys OpenRouter, OpenAI vs. Anthropic & the AI Backlash

    YouTube

    49:14 into “Stripe Buys OpenRouter, OpenAI vs. Anthropic & the AI Backlash”

  14. (14)

    The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition

    YouTube

    22:29 into “The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition”

  15. (15)

    Apple’s New Siri, SpaceX’s $300B IPO & the AI Company Killing Its Own Brand

    YouTube

    16:31 into “Apple’s New Siri, SpaceX’s $300B IPO & the AI Company Killing Its Own Brand”

  16. (16)

    ⁠The Real Reason Elon Wants SpaceX and Tesla to Merge Now

    YouTube

    22:06 into “⁠The Real Reason Elon Wants SpaceX and Tesla to Merge Now”

  17. (17)

    Jeff Dean Leaves Google, Airtable Sells, and the Bag-Securing Era

    YouTube

    31:02 into “Jeff Dean Leaves Google, Airtable Sells, and the Bag-Securing Era”

  18. (18)

    Google Just Won AI, TikTok’s “Self-Control Badges”, ChatGPT Enters Group Chats

    YouTube

    21:03 into “Google Just Won AI, TikTok’s “Self-Control Badges”, ChatGPT Enters Group Chats”

  19. (19)

    Google I/O Just Killed SEO? Agentic Search, Universal Cart, and the New Web

    YouTube

    10:19 into “Google I/O Just Killed SEO? Agentic Search, Universal Cart, and the New Web”

  20. (20)

    Why SpaceX's $1.75T IPO Makes No Financial Sense (And Doesn't Have To)

    YouTube

    31:58 into “Why SpaceX's $1.75T IPO Makes No Financial Sense (And Doesn't Have To)”

  21. (21)

    Hype vs Reality: Our 2026 Predictions for Tech, Media, and Culture

    YouTube

    15:46 into “Hype vs Reality: Our 2026 Predictions for Tech, Media, and Culture”

  22. (22)

    Meta Muse vs Instinct vs OpenAI Dots: Do AI Agents Have Any Moat?

    YouTube

    14:15 into “Meta Muse vs Instinct vs OpenAI Dots: Do AI Agents Have Any Moat?”

  23. (23)

    Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)

    YouTube

    24:20 into “Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)”

  24. (24)

    Elon’s Trillion-Dollar Move, WaPo Layoffs & The Super Bowl

    YouTube

    13:21 into “Elon’s Trillion-Dollar Move, WaPo Layoffs & The Super Bowl”

(By email)

Compounding Insights