Ed Zitron on Anthropic's IPO (S-1), AI Debt, and Counterparty Risk

Anthropic

Across the shows: 8 speakers in other episodes, 9 bullish and 19 bearish calls, past 30 days

• Ed Zitron · EZPRBearish

  Anthropic was in worse financial shape than OpenAI in 2025, spending $2.75 per dollar of revenue.

  Zitron ran the numbers from Anthropic's leaked S-1 for 2025: Anthropic lost about $8B on about $4.6B of revenue, spending $2.75 to make each dollar versus OpenAI's roughly $2.60 on his own reporting. He says the 'more profitable, more sustainable, more stable' narrative about Anthropic was the opposite of the 2025 reality. He also disputes the 2026 profitability claims: they are on an adjusted basis, so the company is not actually profitable. He speculates the Q2 turn was likely driven by Elon Musk discounting compute, a cost reduction he calls one of the most aggressive in history. He doubts a profitable AI lab is possible at all, and says 2026 numbers must be dramatically better or the company is in distress.

  “Anthropic, at least in 2025, was a worst business in OpenAI... Spent $2.75 to make a dollar, and OpenAI did $2.60”at 0:48

• Ed Zitron · EZPRBearish

  Nearly half of Anthropic's revenue flows through Google and Amazon resale, a dangerous concentration risk.

  Per the S-1 disclosure reported by Reuters, about 47% of Anthropic's revenue came from Google and Amazon reselling its models. Zitron calls this a worrying concentration risk because Amazon also funds and sells models to OpenAI, and Google sells its own Gemini and provides compute to OpenAI. The result is that Anthropic's largest customers by proxy are also its competitors. Its success therefore depends on those partners' willingness to sell its tokens, and they can discount or resell them however they choose. He says the narrative that Anthropic was less concentrated than OpenAI was the reverse of the reality, and that the 2026 numbers will show whether the concentration has worsened.

  “about 47% of Anthropic's revenue came through sales from Google and Amazon reselling their models, which is a worrying concentration risk”at 1:07

• Ed Zitron · EZPRBearish

  Anthropic's IPO is questionable; the numbers are poor and retail investors may lose hundreds of millions.

  Zitron argues the S-1 numbers are bad: 2025 losses, adjusted-basis profitability claims, and very large non-cancelable obligations. He thinks the IPO should not be happening unless 2026 numbers are dramatically better. He calls it peculiar that the most important IPO of the year has been delayed repeatedly, with target dates slipping from August through September to October and November. He suspects the numbers did not look good to investors. He says the only reason it is allowed to proceed is free-market rationale that helps retail investors lose money.

  “Like, this is not a good business. And unless their 2026 numbers are dramatically different, I gotta ask, like, why are they even being allowed to go public?”at 8:34

• Ed Zitron · EZPRBearish

  Anthropic's $400B+ non-cancelable compute commitments are a terrifying counterparty risk for it and its hyperscaler partners.

  Zitron cites the S-1 with about $518B of total compute commitments, of which $413B is non-cancelable. That includes $252B across Microsoft, Google and Amazon and $161.2B in Broadcom TPU lease obligations. He compares this with quarterly revenue of about $11.5B and says the distance between these numbers is severe. He says Google, Amazon and Microsoft have these contracts in their analyst expectations, so unpaid obligations would force sharp cuts to their future revenue guidance. Even if Anthropic were profitable, he argues it cannot pay these costs from revenue or cash flow, so it would have to borrow, perhaps $50–100B a year in high yield, and one bad quarter could collapse the business. He also cites Dario Amodei's comment that misjudging compute purchases would make him bankrupt.

  “Anthropic has over $400,000,000,000 worth of noncancelable compute agreements... That alone is absolutely terrifying.”at 9:58

• Ed Zitron · EZPRBearish

  Anthropic's annualized revenue growth is leveling off as enterprise token spending gets pulled back.

  Zitron agrees with a ranking-site chart showing Anthropic's annualized run rate levelling off while OpenAI keeps growing, though still behind. He explains it with a 'token-maxing' phase in the first half of the year. Anthropic switched customers to token-based billing in Q1, and CEOs told staff to spend as much as they liked. Then they saw the bills and pulled back, bit by bit. Large contracts with minimum spends have delayed the visible drop, but he expects the slowdown to show through.

  “They've shown Anthropix ARR, annualized run rate, Jesus Christ, kind of leveling out and OpenAI is growing, but still behind them.”at 23:27

• Ed Zitron · EZPRBearish

  Anthropic's reported run-rate figures are deceptive compared with its actual quarterly revenue.

  Zitron compares Anthropic's Q2 annualized run rate of about $48B with the $11.5B of revenue it actually booked in the quarter, calling them 'almost completely different numbers.' Q1 revenue was $4.8B, so revenue did roughly double, but annualized revenue rose from about $8–9B at the start of the year to $48B. He says run rate is undefined, so companies can annualize a single high month or a large contract. He suggests Anthropic may be counting a $10M minimum-spend contract as $10M times 12 or 13 instead of spreading it over the contract term. He argues the numbers are deceptive and calls for SEC regulation of private-company revenue disclosure.

  “Q2 Anthropic was at 48,000,000,000 annualized run rate, They made 11,500,000,000 in revenue. Those are two very, very different numbers”at 24:39

• Ed Zitron · EZPRBearish

  Anthropic's doom-style safety warnings in its S-1 are a marketing liability that is hurting its IPO story.

  Zitron says the AI-safety and doom narrative was a useful marketing mechanism for years, pushed by Sam Altman and Dario Amodei, but it got out of control and became a liability. Now the S-1 must include risk factors saying the product 'might kill people.' On its own this would not matter, but he argues it adds to the other problems in an already unprofitable business. He says the company has 'blinked' and cannot easily shake these stories. He suggests the CFO should have asked whether the warnings belonged in the filing.

  “that narrative got out of control that is only making things worse”at 29:38

• Ed Zitron · EZPRBearish

  Anthropic likely won't pay its data center bills, and it needs cash it cannot easily raise.

  Ed says he doubts Anthropic pays its data center bills on time, which slows CoreWeave's collections and adds interest. OpenAI and Anthropic make up about 80% of data center demand, and their ability to pay depends on venture-backed customers that lose money and need ongoing venture funding. He also expects Anthropic, once public, to need debt that is priced like junk, with huge issuance every year, and he thinks one of the AI labs could collapse.

  “So it won't be our customers don't pay, though I think OpenAI doesn't pay, and I don't think Anthropic pays.”at 59:16

• Ed Zitron · EZPRBearish

  Anthropic will likely be pushed through an IPO near $2 trillion, but its stock will be volatile and worry investors.

  Zitron expects Anthropic to be crammed through an IPO despite a chance it doesn't list or gets reworked. He expects a volatile stock that pops and drops, and says investors will get worried once the first quarterly results are shared. He is waiting on the public S-1 and its 2026 numbers, the biggest unknown.

  “I think Anthropic will be pressed through. I think they get 2,000,000,000,000. I think they have a volatile stock. I think maybe they pop and they drop and they pop and they drop.”at 1:19:06

OpenAI

Across the shows: 8 speakers in other episodes, 4 bullish and 12 bearish calls, past 30 days

• Ed Zitron · EZPRBearish

  OpenAI's $70B run-rate claim is inflated; he expects about $36B of revenue this year.

  Zitron says the reported jump in OpenAI's run rate to $70B, up from $40B a month earlier, is suspicious because it is undefined. He thinks the company is annualizing a non-recurring item, most likely API usage, which is not recurring like subscriptions. People are meant to read the number as expected revenue, but he forecasts about $36B of revenue for OpenAI this year. He also argues OpenAI's accounting is opaque: it does not count Microsoft-driven revenue as its own, and the Microsoft revenue-share payments are small. Zitron concludes that these run-rate figures are useless and that only audited numbers should count.

  “a 30,000,000,000 jump in run rate in a single month makes me very suspicious... It won't. It's gonna be 36,000,000,000 for this year for OpenAI, I think.”at 19:11

• Ed Zitron · EZPRBearish

  OpenAI likely won't pay its CoreWeave bills, and its compute spending depends on endless venture funding.

  Ed says the CoreWeave debt problem is not just whether customers pay. He thinks OpenAI doesn't pay and expects delays to build interest before payment. OpenAI and Anthropic account for about 80% of data center demand, and CoreWeave has 360-day payment terms with OpenAI. He says OpenAI is projected to spend $750B on compute through 2030, which must come from venture capital that has put $800B into AI since 2023, or from Nvidia or SoftBank, which cannot keep funding rounds. Its revenue growth also raises costs, so it is a cash flow problem.

  “So it won't be our customers don't pay, though I think OpenAI doesn't pay, and I don't think Anthropic pays.”at 59:16

• Ed Zitron · EZPRBearish

  OpenAI is likely to have a nightmare IPO and may struggle to go public.

  Zitron says OpenAI will have a nightmare IPO regardless of other conditions and may have real trouble going public. This follows his broader argument that the many conditions for the AI-linked financing chain to work, including an OpenAI listing at a high enough valuation, are unlikely to all hold.

  “I also think OpenAI is gonna have a nightmare IPO ing regardless. I think they may have real trouble going public.”at 1:19:57

Meta Platforms

Across the shows: 5 speakers in other episodes, 10 bullish and 0 bearish calls, past 30 days

• Ed Zitron · EZPRBearish

  Meta's Muse agent is costly to run and has no clear revenue model.

  Zitron says Muse delivers the agent promises made three years ago, such as booking appointments and running websites, but at a very high cost. Each user gets a virtual machine with 8GB of RAM on top of large language models, which is a massive infrastructure burden. He grants that it has some utility, such as finding a delivery store, but asks where the revenue model is. Proposed transaction fees would be shreds of revenue on a very high-volume, opex-heavy business, and subscriptions compete with OpenAI's $20–$500 plans. He also flags privacy harms from the agent, such as personal data being exposed.

  “at what is very clearly an obscene cost. You've got a little virtual machine with eight gigabytes of RAM for each person using this.”at 33:10

• Ed Zitron · EZPRBearish

  Meta's stock rally on its Muse AI product is insane since the product generates negative revenue.

  Ed says Meta's stock skyrocketing on Muse is insane. He argues that whatever one thinks of the product it is generating negative revenue and has become a massive infrastructure burden, and he asks where the money will come from. In his view, markets are rewarding a product narrative that does not yet support the valuation.

  “off of Muse is insane. However you may feel about the product, it's generating negative revenue.”at 38:49

• Ed Zitron · EZPRBearish

  Meta's off-balance-sheet Hyperion data center financing trades at 89 cents and is a warning for the sector.

  Ed describes Project Beignet Investor, the Hyperion data center for Meta, as a variable interest entity. Meta owns about 20% of it, so the debt stays off Meta's balance sheet even though Meta builds, funds, equips and exclusively uses the site. The deal is one of the better ones yet trades at 89 cents on the dollar. Ed concludes that every data center project raising money from here will need 8%, 9%, 12% or 13% money, which becomes untenable at some point.

  “That's one of the better deals and that's selling at 89¢ on the dollar. So basically every data center project that is raising from here... will be raising minimum at 8, maybe 9, maybe 12, maybe 13.”at 56:09

NVIDIA

Across the shows: 7 speakers in other episodes, 12 bullish and 1 bearish calls, past 30 days

• Ed Zitron · EZPRNeutral

  Nvidia's equity-style investments in AI buildout are less risky than Broadcom's debt-financed approach.

  Zitron says Nvidia is 'a little smarter' than Broadcom in financing the AI buildout. Nvidia classically makes equity investments in customers such as xAI (SpaceX), Lancium, Stargate Abilene and CoreWeave, which then raise money to buy its GPUs. Nvidia puts real money in, but it is not a liability on its balance sheet. He acknowledges there may be some future risk to net income, but says it is nowhere near as bad as taking on $35B of debt.

  “NVIDIA is by comparison actually being a little smarter because NVIDIA classically does equity investments”at 16:41

• Ed Zitron · EZPRBearish

  Nvidia's revenue likely drops in fiscal 2028, and its year-forward guidance looks unrealistic.

  Ed thinks Nvidia's revenue will drop in fiscal 2028, which begins February 2027. He says Nvidia has given year-forward guidance that requires it to make around $67B (transcribed as $6.70B) just to make its own customers' purchases work, and he does not see that happening. He cites a Jefferies study that about 50% of chips sold in 2026 and 2027 will have nowhere to go, and says roughly $200B of GPUs are not installed anywhere. Nvidia's sales are tied to customers' capital, which is drying up.

  “NVIDIA will have to make $6.70 or more billion dollars just to make their own. They did a year's forward guidance, which is bonkers. I don't see that happening.”at 1:08:11

SoftBank Group

• Ed Zitron · EZPRBearish

  SoftBank literally can't keep funding AI rounds, and its near-10% financing looks unsustainable.

  Ed says the AI build-out depends on venture capital, Nvidia and SoftBank for funding. Nvidia cannot afford too many rounds, and SoftBank literally can't. He notes SoftBank just raised about $11B at a rate barely below 10%, and asks how any of this works. He also refers to Masayoshi Son's routine of borrowing more to pay off debt, which he sees as a failing model.

  “Or from companies that like NVIDIA that really can't afford to do too many rounds of SoftBank who literally can't.”at 1:06:30

• Ed Zitron · EZPRBearish

  SoftBank faces a heavy debt refinancing wall and margin-loan pressure tied to Arm and OpenAI.

  Zitron says SoftBank needs many things to go right. Arm must grow indefinitely and OpenAI must list at a valuation that lets SoftBank use its already-collateralized stock to raise more money. He cites roughly $40 billion of debt to refinance in the next year, including chunks from 2020 and 2021 issued when rates were lower.

  “Because SoftBank, I think, has $40,000,000,000, maybe more of debt needs to be refinanced in the next year. Chunks of it from 2020 and 2021 where interest rates were lower.”at 1:16:54

Broadcom

• Ed Zitron · EZPRBearish

  Broadcom is a risky AI chip play, reliant on three customers and increasingly financed with debt.

  Zitron says a large chunk of Broadcom's revenue is now AI chips, mostly sold to three companies: Google for TPUs, which are mostly rented to Anthropic, Anthropic itself, and OpenAI with its custom chip, plus Meta's MTIA chip. Unlike Nvidia's equity investments, Broadcom is taking on debt, including a $35B private credit package in which Apollo and Blackstone are maybe involved. He says Broadcom also financed some chips for OpenAI. He points to debt from its 2023 VMware acquisition, a story that it would raise $60–100B more, and VMware growth that is petering out. He concludes Broadcom is playing a risky game and is a 'real dog of a company'.

  “So Broadcom is a real dog of a company too... Most of those are being sold to three companies.”at 15:55

CoreWeave

• Ed Zitron · EZPRBearish

  CoreWeave's roughly $7B of bonds would reprice near 13%, and its debt-funded model looks unsustainable.

  Ed estimates CoreWeave's roughly $7B of straight bonds issued in 2025-26, carrying about 9%, would cost about 13% if repriced at current spreads and the selloff, possibly higher. He calls CoreWeave's claim that its finance costs are falling a lie. He says the company has raised through leveraged loans, bonds, convertibles, at-the-market equity and a sale to Nvidia, which looks like a company raising to stay ahead of its bills. He cites an analyst estimate that CoreWeave must raise about $100B over four years. Delays in bringing data centers online (he says the London project is cancelled) let interest build before customers pay. Its customers include Microsoft, Meta, OpenAI, Anthropic and Jane Street, and he doubts OpenAI and Anthropic will pay on time.

  “If you reprice them, all of that debt, which is still around nine points at the time, you're looking at 13%.”at 53:29

Oracle

Across the shows: 2 speakers in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Ed Zitron · EZPRBearish

  Oracle's bonds already price like junk and a downgrade would force forced selling by funds.

  Ed says Oracle is dangerously close to a downgrade, sitting one notch above junk. A downgrade would force pension and investment funds that cannot hold junk to sell, pushing its borrowing costs up. He says Oracle already trades like junk: short-dated bonds would yield about 7.1-7.5% and long-dated ones about 8.1-8.3%, and a New Mexico data center debt sits at 89-91 cents on the dollar after a force majeure notice. He adds that Oracle still needs to borrow, must refinance cheap 2020 debt at 2.8% in April at far higher rates, and that its stock rallied on an OpenAI run-rate headline despite power problems at its Port Washington data center.

  “And let's be clear about something. Oracle is already pricing like junk.”at 55:17

Blue Owl Capital

• Ed Zitron · EZPRBearish

  Blue Owl's private credit lending to data centers is poorly collateralized and at risk of not being repaid.

  Ed says Blue Owl is in pretty much every Oracle Stargate data center and in multiple CoreWeave projects. He expects most private credit debt tied to these projects will not be repaid if the data centers are delayed or fail. He argues private credit is not collateralized well enough to withstand that, and singles out Blue Owl. He also cites a report that Blue Owl took fifteen minutes to decide to invest in Stargate Abilene, which he takes as evidence that lenders have not thought hard about the risk.

  “In fact, I think most of the private credit debt doesn't get paid here. And I don't think private credit is collateralized well enough to weather that, especially not BlueOwl.”at 1:03:21

Arm Holdings

• Ed Zitron · EZPRBearish

  Arm's share price is exposed to forced selling if SoftBank faces margin calls.

  Zitron notes SoftBank owns about 80% of Arm and has margin loans against it. If Arm falls, those loans could be called or SoftBank may have to sell Arm shares. That selling would drive the stock regardless of Arm's fundamentals, because investors would trade on SoftBank's selling.

  “So if ARM crashes, those margin loans that SoftBank has are gonna get called or SoftBank will have to start selling off ARM, which will create a catastrophic effect”at 1:16:08

Banks

• Ed Zitron · EZPR

  Regional bank selloff over AI agents moving deposits is baseless and unexplained.

  Ed calls the bank selloff on the thesis that people will use AI agents to reallocate their capital a little silly. He says most people have one checking account and no brokerage, so an agent cannot easily move money into Treasuries or CDs. Jack cites a regional bank with an average account of $3,000, where moving from 1% to 4% barely matters, and banks would not let an LLM access accounts for regulatory and liability reasons. Ed says no one can explain how the reallocation would happen, so the selloff reflects a narrative rather than a measurable risk.

  “But the idea that banks are selling off because of the off chance that people use agents to do something that no one can really explain what it is.”at 39:06

• Ed Zitron · EZPR

  Goldman Sachs and Japanese banks are exposed in nearly every data center deal and share the blame.

  Ed says the people to blame for the data center debt build-out are the private credit funds and the banks that finance them, naming Goldman Sachs, Sumitomo, Mitsubishi UFJ and SMBC. He says they are in basically every data center deal and keep enabling large borrowers. He notes he has heard the two Japanese banks are selling off the risk, and says Goldman and Morgan Stanley are in a lot of these deals, so the risk is everywhere.

  “the people to blame for this are the private credit funds, and they are the Goldman Sachs', the Japanese banks, Sumitomo and Mitsubishi, MUFJ, SMBC. They are in basically every data center deal.”at 1:01:51

AI infrastructure & data centers

• Ed Zitron · EZPR

  Data center debt will break: expect dead projects, failed bond sales and downgrades within a year.

  Ed argues the build-out has exhausted private capital sources and that data center financing is becoming unworkable. Borrowing costs are rising to 8-13%, Treasury yields are high, and the spread of depreciation over gross margins means the first five years of revenue won't cover costs. He expects a failed or barely covered bond sale (Amazon's recent deal drew only 1.6 times interest, versus three to four times normally), more downgrades such as Oracle, and dead Stargate projects. He thinks the problem will show up once someone is not paid.

  “So I think that we're gonna see a bunch of dead data center projects. I think we see Oracle downgraded.”at 1:07:51

• Ed Zitron · EZPR

  Data center buildout is outpacing proven demand, and stated AI capacity far exceeds what is actually running.

  Zitron argues there is no proof of enough demand for even a quarter of the data centers being built. Data centers are also taking too long to finish. As evidence he cites Microsoft claiming 12 gigawatts of data center capacity, while his and Bloomberg's reporting found only about 2 gigawatts of AI capacity. He concludes that companies are holding massive stockpiles of chips that are not being turned on.

  “there is no proof that there is enough demand for even a quarter of the data centers being built... Microsoft claimed they had 12 gigawatts... They only have two gigawatts of AI capacity.”at 1:14:57

AI: overall outlook

• Ed Zitron · EZPR

  AI has not improved productivity or GDP outside data centers; its gains are construction and GPU sales.

  Ed argues AI was meant to be the next productivity and GDP booster, but only the infrastructure side (GPUs and construction) has worked. Productivity is basically the same and growth is the same, and GDP outside data centers and GPUs has not grown. Software sales growth is no more remarkable than before. AI has instead created a complex financial crisis inside the shadow banking system, and the bet only works as long as the buildout continues.

  “What has AI actually improved? Has it improved productivity? No. Productivity is basically the same. Growth's the same.”at 1:10:41

• Ed Zitron · EZPR

  AI was sold as a panacea, but its products reflect hype and growth-at-all-costs management rather than stable businesses.

  Zitron says AI was meant to fix everything going wrong in tech, including becoming a new infrastructure layer and a platform for startups. Instead, he argues, the tech industry has been taken over by management consultants focused on growth. As a result, products reflect marketing and hype more than businesses people would pay for, and the Silicon Valley playbook is dead, so everybody will pay.

  “I think AI was meant to be the panacea for everything that was going wrong in tech... I think what's ended up happening is we have seen what happens when the tech industry gets taken over by management consultants”at 1:20:11

US stock market outlook

• Ed Zitron · EZPR

  Markets look bleak, with the S&P 500 too heavily weighted toward big tech and weak cash flows elsewhere.

  Zitron says the broader financial markets are in 'dog shit season' and look very bleak. He argues the S&P 500 is too weighted toward large tech companies, while the rest of the index lacks the cash flows to justify its prices. He links the cynicism to the delays in the Anthropic IPO and the wider concerns about AI profitability.

  “The S and P 500 is way too weighed on the big tech companies, the rest of the S and P 500, the cash flows aren't there.”at 27:29

Market sentiment, positioning & volatility

• Ed Zitron · EZPR

  Markets are driven by AI narratives and ghost stories, disconnected from fundamentals.

  Ed says markets move on vague AI narratives rather than fundamentals. Examples include the Citrini global intelligence crisis report, PayPal and Mastercard selling off because agents would replace commerce, and a Matt Schumer post that overstated a model's capabilities. He thinks no one really knows what is going on, so everything is marketing-driven, and that someone could manipulate the market this way. When the real bad news on AI arrives, he expects it to be tangible and impossible to narrate away, as with Oracle's power problems.

  “I think it's part of a broader problem that the markets are no longer connected to reality.”at 38:36

US inflation

• Ed Zitron · EZPR

  Inflation is broad price gouging that rate hikes cannot control; regulators must use price-gouging laws.

  Ed points to rising costs: his mortgage is around 8%, basic loans and goods are more expensive than ever, and inflation has been everywhere for years. He argues it is not being controlled by rate hikes and is a result of price gouging across the board. He thinks the media has failed to report this, and that the only real tool is financial regulation and enforcing existing price-gouging laws, not the Fed.

  “When inflation is everywhere and has been for years and is not being controlled by rate hikes at this point. Inflation is a result of price gouging across the board.”at 42:40

Consumer AI products

• Ed Zitron · EZPR

  Consumer AI agents lack an iPhone-level moment; they won't become essential to everyday life.

  Ed argues consumer AI has not offered a clear reason to depend on it. He tested an agent for restaurants and got poor results, and says the real problem with reservations is automated bots, not reservation tools. Flights and wedding cakes are high-touch decisions that need human input. Muse is a nicer chat interface, but there is no single product that makes consumers say this is essential to their life, and a few bad incidents could push people away.

  “But where does that turn into something that makes a consumer go, this is essential to my life.”at 47:22

US Treasuries & yields

• Ed Zitron · EZPR

  Rising 10-year Treasury yields near 5.3% make AI data center debt increasingly unaffordable.

  Ed says Treasury rates are only getting worse. Ten-year yields at 5.3% are already an insane base for data center debt, which must clear that rate plus a risk premium. Even if the yield falls a bit, a rate base this high raises the cost of every new data center financing, and interest rates keep going up, adding to the wall the sector will hit.

  “And the Treasury rates are only getting worse... this is bad, but it's also 5.3% on ten years.”at 56:27

Private credit

• Ed Zitron · EZPR

  Private credit is already a brewing crisis, having lent heavily into volatile, low-margin data center projects.

  Ed argues private credit has invested heavily in some of the most volatile and difficult-to-complete projects in the world, which also have poor margins. Data centers do not print money when switched on. He says the risk could give private credit its own financial crisis, and points to reporting from Nick Nemeth and the Financial Times as the best coverage of the risk, which he says the broader market has not absorbed.

  “But this private credit is already a brewing crisis, and private credit has invested heavily in what I would describe as some of the most volatile and difficult to complete projects in the world, which also have piss poor margins.”at 1:04:59

Enterprise AI adoption & AI software

• Ed Zitron · EZPR

  Software is getting too much valuation credit for bolted-on AI features that are mostly hype.

  Zitron argues software is being given too much credit because vendors call AI additions new features that justify a higher valuation multiple. He notes that right now everyone is adding AI to everything, so the new features are not evidence of durable value.

  “software is being given way too much credit because they say, oh, it's a new feature. So this is new. The new value here. Need to the hedonic multiplier needs to make this more valuable.”at 1:13:02

AI pricing & unit economics

• Ed Zitron · EZPR

  AI inference is a brittle, airline-like business where renting GPUs ahead of demand makes it unprofitable.

  Zitron compares AI inference to an airline. A provider rents a fixed number of GPUs and needs a set customer load to break even. Too many customers degrade service and drive churn, while too few means paying for idle GPUs. He says this makes the model very brittle, and that it could collapse for OpenAI, Anthropic, and smaller players, leaving data centers with no use.

  “So it's already quite a brittle business model. So this means this all collapses, OpenAI, Anthropic, [Ghost hits up] or just much smaller.”at 1:14:49

AI bubble & tech valuations

• Ed Zitron · EZPR

  An AI bust would likely hit tech stocks hard, since the S&P 500 rests on AI-linked megacaps.

  Zitron says he does not see how an AI bust fails to produce some form of minor or major depression in the tech sector. He notes the S&P 500 is being held up by Microsoft, Google, Meta, Amazon, Nvidia, Broadcom, Micron, and SK Hynix. He argues these companies are intrinsically tied to the AI bubble, so the index's stability depends on AI spending holding up.

  “I don't see how that doesn't work out to some form of minor or major depression in the tech sector. And considering that the S and P 500 is pretty much being held up by Microsoft, Google, Meta, Amazon”at 1:15:26

Central banks outside the US

• Ed Zitron · EZPR

  Rising Japanese interest rates will push Japanese investors to repatriate money from US equities.

  Zitron notes interest rates are going up further in Japan. He argues this will lead Japanese investors to start repatriating money instead of investing in American equities, a flow that would weigh on US stocks.

  “interest rates are going up even more in Japan. Which means that Japanese investors are gonna start repatriating their money instead of investing in them in American equities.”at 1:17:52

Venture capital & startups

• Ed Zitron · EZPR

  Roughly 80% of venture capital could be wiped out by an AI bust, worse than the dotcom crash.

  Zitron says he sees a scenario where 80% of venture capital is washed out, worse than the dotcom bubble. He argues there is no post-dotcom recovery story this time. Incomplete data centers still cost as much to build, and operating costs stay high for years, so there is no cheap way out of the spending.

  “there is a scenario here where 80% of venture capital gets washed out with this, worse than the dotcom bubble. And there is no post dotcom bubble thing here”at 1:17:52