Yesterday, when I first read the Bloomberg headline “Meta Is Planning a Cloud Business to Sell AI Computing Power,” the first thing I instinctively thought was, “Selling AI computing power to whom, exactly?” AI labs are currently split into two categories: those who built their own infrastructure and those who did not. In the first category, you find the likes of Microsoft, xAI (merged into SpaceX), and Google’s parent company, Alphabet. In the second category, you find two companies that together account for the vast majority of the group’s computing demand, up to 90% according to reliable estimates: OpenAI and Anthropic.
Both OpenAI and Anthropic either purchase computing capacity directly or, most of the time so far, indirectly via their respective investors such as Microsoft, Amazon, and Alphabet. These investors can either build the infrastructure themselves and supply OpenAI and Anthropic, or decide to purchase that computing capacity in the market from a third-party provider: the famous “neocloud companies.” So far, everything has been fairly straightforward. However, META just threw a wrench into this well-oiled circular scheme upon which the whole narrative supporting the AI bandwagon relied.
Both hyperscalers and AI labs constantly lamented, like a broken record, how their growth and path to profitability were restrained by the limited availability of computing power. Any investor who was presented with projections of the expected returns and profitability of building data centers to supply this insatiable demand rarely thought twice before jumping in. It was considered a no-brainer, effectively the holy grail for those seeking stable and high, long-term returns on capital, with strong collateral on top, like insurance and pension funds. However, these investors do not have the skills to fund and eventually run the construction of data centers, which is why Private Equity and Private Credit firms found themselves in the right place at the right time. The flow of private investors’ capital into the sector has been biblical for years; then circular financing and revenue round-tripping did the rest, contributing to the impression of historic growth that no business could achieve before.
Obvious signs that something was wrong have been plentiful for months. I am saying “obvious” because if anyone dared to look closely at what was happening, as I did, they would have realized how the whole AI narrative pushed by the trio Huang-Altman-Nadella since 2022 was a complete scam, hidden within one of the most sophisticated efforts of financial engineering. Here are some examples of recent and obvious warning signs that, of course, I duly documented:
- HAS THE NVIDIA CASH COW STOPPED PRODUCING MILK?
- ALPHABET EQUITY RAISE: A BOLD MOVE OR A GIANT MISTAKE?
- NVIDIA IS NOW VERY CLOSE TO HITTING THE CIRCULAR FINANCING BRICK WALL
- MAG7s NOW ISSUING DEBT TO CONTINUE BUILDING AI IS A LOUD WARNING
- COREWEAVE: THE WEWORK OF AI
- AI IS BECOMING ORACLE’S LITTLE BIGHORN
- THE DATA CENTERS FRENZY WILL BE REMEMBERED AS THE LARGEST WASTE OF CAPITAL IN HISTORY
Putting all this together, you can easily understand how this whole Ponzi-like scheme was standing on two legs all the time: continue sucking private investors’ capital and push the narrative as far as possible to feed as much as possible fear of missing out. That private investor capital isn’t infinite should be a point of no argument, but somehow many thought there were $3 trillion available in the world to fulfill all Sam Altman’s pipe dreams. Sadly, the money is running out and, not surprisingly, all hyperscalers, including Nvidia, that issued debt itself (perhaps because it isn’t able to “print cash” anymore), are rushing to suck out of the system what’s left. Yesterday, the whole narrative claiming that “AI isn’t making any profit yet because we don’t have enough computing power” imploded. Why? If META is now setting up a cloud business to sell computing capacity, when the company has been one of the largest buyers of GPUs and sponsors of new data centers itself, this means one thing and one thing only: META not only does not suffer from a computing constraint anymore, but it also has spare capacity to sell in the market. If META has spare capacity, will it continue to be one of the largest buyers of GPUs and sponsors of data centers? Obvious answer: no.
However, do not expect META to admit failure, to admit that the tens of billions of CAPEX it spent so far won’t bring any more of the biblical revenues AI models were supposed to bring, to admit that it is now pivoting to selling computing power in an effort to attach some revenues to all that CAPEX that was squandered. What’s the point of admitting any failure when the market still celebrates whatever headline comes its way, without making sure whether they are eating chocolate or human excrement, sending your stock up 10%? As a consequence, I expect the upcoming earnings season will still be “fine,” although more and more concerned investors are now starting to pay attention to the real numbers rather than enjoying listening to storytelling. There is no doubt the market still believes in the whole original AI dream. The reason why MAG7 shares aren’t performing well lately is that investors are jumping from the hyperscalers ship onto the memory companies ship. Memory companies do not lose a day telling the world how much demand they have and how much money they will make from now to eternity, a world that mostly forgot that the very same Micron, SK Hynix, and Samsung already formed a cartel to boost memory prices years ago during the dot-com bubble and were heavily fined once they got caught:
- “Chipmakers fined by EU for price-fixing”.
- “Korean Company Pays Second Largest Criminal Antitrust Fine in U.S. History”
What are the chances that the gang is now doing exactly the same? In my opinion, pretty high, considering they have all been sued again: “DRAM price-fixing allegations return: Samsung, SK Hynix, Micron sued in US”
So let’s stop pretending we’re watching an inevitable march toward infinite AI profits. What we’re really watching is the moment the circular financing machine starts eating its own tail. When a company as capital-hungry and GPU-obsessed as META suddenly discovers it has compute to spare and decides it’s worth monetizing, this is not a victory lap. It’s an admission, disguised as “strategy,” that the CAPEX party has overshot the real, paying demand. From here, the math gets ugly fast. Spare capacity turns into price pressure; price pressure turns into margin compression; margin compression turns into “discipline” and “prioritization”: corporate euphemisms for cutting spend. And once the biggest buyers start acting like sellers, the entire narrative that justified the data-center frenzy collapses into what it always was: storytelling built on leverage, vanity metrics, and investors’ willingness to suspend disbelief. The next chapter won’t be written by headlines, but by cash flows. If the market wants to keep cheering slogans, it will. But the moment compute becomes a commodity and demand stops being mythical, the “holy grail” turns into stranded capital. And stranded capital has a way of revealing, very quickly, who was building the future, and who was merely recycling money in a very expensive circle.
JUSTDARIO Premium is now live on Slice Finance
Dear readers and followers, after many requests over the years for more direct access to me and premium content, I accepted the invitation to join as a mentor and launched a Premium Subscription on Slice Finance this Sunday. Slice Finance not only lets me share premium content, commentary, and trade ideas for a market-professional audience, but also provides many tools you can use to directly interact with me easily: live chats, group chats, and weekly live streams. I’ve set the subscription fee at a fair, accessible price with no fixed time commitment. If you’re interested, you can subscribe to my premium content by clicking this link: https://bit.ly/JustDario