Jon Stewart

Ed Zitron says AI data centers will not get the second life that leftover dot-com fiber got, because a half-finished AI site is useful for AI and not for much else.

Once telecom fiber was in the ground, later companies could use it. An unfinished AI data center, he says, needs about as much money in three, four, or ten years to finish as it needed at the start, and the electricity stays just as expensive. GPUs have no alternate use, and demand for them outside Anthropic and OpenAI is minuscule. People reach for the dot-com comparison, in his view, because it lets them believe that finance and tech can correct themselves and that the spending was not a waste. He does not think the buildings,

Ed Zitron says CoreWeave is a neocloud, a company that raises debt, buys GPUs, and builds data centers. He calls it the biggest of the neoclouds and the worst of them.

CoreWeave has 30 to 40 billion dollars of debt, and Morgan Stanley has said it will need 100 billion dollars more. He thinks that if CoreWeave raised debt today the rate would be 13 to 15 percent, close to a personal loan, because the neoclouds could never afford the chips out of cash flow. The customers, he says, are Microsoft buying for OpenAI, Google buying for OpenAI, Anthropic, and Meta. Jon Stewart compares the loop to a Beavis and Butt-Head bit in which two kids pass one dollar back and forth until the chocolate is gon

Ed Zitron says an Anthropic IPO is exit liquidity. Buyers are sold the idea of investing in the next Google or the next Meta. In his account, the offering is how earlier venture investors sell shares they bought much cheaper.

He sets the offering around one and a half to two trillion dollars. Even if Anthropic grew to the size of NVIDIA, he says the return would be three or four times the money, not the old venture outcome, and he does not think Anthropic gets there. Anthropic has negative EBITDA, which he glosses as negative earnings before taxes and depreciation. Jon Stewart asks whether the doubt is the same doubt people had about Google before advertising made the network pay. Ed Zitron

Ed Zitron says the AI labs, the chipmakers, and the data-center landlords are renting and investing in one another. Jon Stewart asks if the whole buildout is circular, and Ed Zitron says it is.

The labs rent compute rather than own it. Anthropic, he says, will be paying about $3.1 billion a month to Elon Musk as of December. The labs lose billions of dollars, which makes banks uneasy about lending to them, so the labs lean on hyperscalers instead. Larry Ellison is building about $400 billion of AI data centers for one customer, Sam Altman, who is supposed to pay about $300 billion over five years. Ed Zitron says OpenAI does not have the money and Oracle does not have the capacity. NVIDIA in

Ed Zitron says Anthropic loses billions of dollars, fewer billions than OpenAI, and the latest quarter of its S-1 is blank. An S-1 is the filing a company sends the Securities and Exchange Commission when it is about to go public.

The figures he says a reader needs in order to understand Anthropic only go through June 30. Anthropic shared many of the numbers with investors at an investor day two weeks earlier, which he takes as the retail buyer being left out. He says the company scraped through the second quarter with a light profit because Elon Musk discounted its compute at SpaceX for two months, and that only some of the compute belongs to SpaceX.

Ed Zitron says an AI data center takes far longer to build than the race for more compute assumes. He calls each one the single most ambitious infrastructure project in history.

Oracle has been building OpenAI's site in Abilene, Texas for two and a half years. He puts the building at about 998,000 square feet, drawing the power of a small city, and he uses Bristol in England as the picture: the power of that city, packed into several large warehouses. Hyperscalers are adding the capacity because they see demand, but he says most of the demand is OpenAI and Anthropic. He estimates that 80 percent or more of AI compute spend comes from those two companies. Microsoft alone, he says, has spent

Ed Zitron says the OpenAI boom is a shell game and a powerful confidence game. He hopes the believers all share the same view, and because they believe it, no one wants to stop. OpenAI is the fastest growing company in the world, and Anthropic is the fastest growing company. The media, regular people, and investors buy in.

The Silicon Valley playbook is to sign as many big deals as possible, push the number up, and wait until the product gets cheaper. The cheaper turn is not happening. The demand on OpenAI and Anthropic is to grow into the next Microsoft, Google, and Amazon, which means acting like Microsoft, Google, and Amazon and signing the contracts. People who only care about growth lo

Ed Zitron says Sam Altman should be in prison, or at least at a police station. An OpenAI model hacked an Australian government website, and a regular person who did the same thing would go to prison.

The models going out and hacking are not going rogue and do not have thoughts. OpenAI and Anthropic are running dangerous automated software at scale on hundreds of billions of dollars of infrastructure, and he thinks Google and Meta are doing the same. In the tests, the software does real damage. There was no reason to train the models to hack, and open source is now available too. The most obvious problem right now is unrestrained, unregulated tech running on massive infrastructure.

Ed Zitron says an AI bust is not Great Financial Crisis level bad, but it will still be very bad, and a different kind of bad.

No area of the economy sits outside the AI project now. Everyone is invested in the market. Retail investors account for 20% of stock market volume, so everyone has exposure. The Great Financial Crisis came from massive speculation and terribly done CLOs. It fell apart because everyone lost confidence in the system. Almost no assets were broken. The world fell apart because 7% of people's mortgages were underwater, and that was the entirety of the crisis.

An AI bust will be different. One of the biggest problems, in his view, could be entirely symbolic.

Ed Zitron says NVIDIA chips are pushing up the price of phones, laptops, and every other consumer electronic because they crowd out ordinary computer memory.

Three companies make all the world's memory: Micron, SK Hynix, and Samsung. NVIDIA sells millions of GPUs that use high bandwidth memory, a special kind of memory that takes more space on the fabricator and leaves less room for the ordinary memory everyone else needs, including AI servers. Deutsche Bank says 50 to 60 percent of NVIDIA revenue comes from hyperscalers, a handful of companies that mostly need debt to keep buying chips, and the chips themselves keep getting more expensive. Because so many of the chips have already been sol

Ed Zitron says the hyperscalers and the other landlords carry the debt on AI data-center buildouts, while most of the labs' commitments are contracts rather than debt on the AI companies.

Anthropic has four hundred and thirty billion dollars of non-cancelable commitments over the next seven to ten years, and OpenAI is projected to spend seven hundred and fifty billion dollars by 2030. Anthropic does have one hundred and sixty one billion dollars of leases with Broadcom, and those leases carry interest. If OpenAI cannot fulfill its service leases, the builders still have to service the debt.

More demand would not fix the timing. Capacity is being built slowly, and little is known about it.

Ed Zitron says Oracle's stock is tanking because a $300 billion, five-year contract with OpenAI has nobody who can pay for it.

Oracle's public relations is phenomenal, and people say Oracle is doing great. Oracle's revenue has been flat for 15 years, adjusted for inflation. The only part of the company growing is AI compute, which is destroying margins because it costs so much to buy and takes so long to install. Larry Ellison's idea was to build endless compute for OpenAI, a company that loses billions of dollars. In September last year, Oracle announced the $300 billion five-year contract. Analysts went completely nuts and called the contract amazing. Then everyone asked whether Oracle ca

Ed Zitron says Anthropic has $430 billion in non-cancellable commitments with Amazon, Google, Microsoft, and Broadcom. The lab is not building its own compute. It agreed to rent data center capacity under a take-or-pay arrangement, so the bill is due whether the capacity gets used or not.

The commitments probably start next year and will cost Anthropic $75 to $100 billion a year within the next year or two, even if revenue does not grow. Anthropic only loses money. Operating expenses will be $100 billion plus in short order. Microsoft, a profitable company making $330 billion a year, has operating expenses of about $150 billion. Anthropic believes it will be the size of Microsoft soon and w

Ed Zitron says the economic payoff from AI has not arrived, even as NVIDIA has grown incredibly rich selling the chips behind it.

NVIDIA basically controls the supply and is printing money as the largest supplier to Microsoft, Google, Amazon, Meta, and SpaceX. The data centers are not actually generating enough revenue, or really any revenue. None of the companies other than SpaceX disclose their actual AI revenues. Companies have had to take on debt and lay people off. Meta is doing a $95 billion share sale. Oracle had to do a $50 billion one and is facing a downgrade potentially, all because the big promises of AI have not paid off. The chief economist at Apollo said the other day there h

Stewart Alsop III is an American entrepreneur and the host of the Crazy Wisdom Podcast .

He joins us on Smash Notes this week to talk about life in Silicon Valley through the lens of his own podcast, as well as the conversation between Peter Thiel and Eric Weinstein on the Portal.

Stewart Alsop tells the Podcast Notes team, on Crazy Wisdom, that they are unusually good at turning a podcast into writing, and that they would have something powerful if they could do the same thing out loud. Matt says second- and third-tier celebrities on Twitter are easier to reach than five years ago, often by a direct message. Alsop says that surprised him. He has cold-contacted guests mostly on Twitter and sometimes on Facebook, about 70 percent said yes, and he has not needed his Silicon Valley network. He is not making money yet. He quotes Nassim Nicholas Taleb on cocktail parties and black swans, and says a podcast is now that party, recorded.