On Thursday, October 29, Alphabet's shares fell 12.6%. The S&P 500 fund SPY slipped just 0.4% the same day. Google's parent had just reported that its cloud business more than doubled its sales, and that it spent $51.4 billion in a single quarter, mostly on the computers that run artificial intelligence. Its operations brought in $42.3 billion of cash in the same quarter, so more went out the door than came in.
Three days earlier, on Monday, a company that is one of Google's biggest customers for those computers had put its own numbers on the table. Anthropic, the maker of the Claude chatbot and AI models, publicly filed its S-1, the registration statement a company must file with the Securities and Exchange Commission before it sells shares to the public. Buried in it: a commitment to spend at least $114.5 billion with Google and at least $110.0 billion with Amazon, "regardless of actual usage."
Put the two events side by side and you get the question that hung over markets all week. Big Tech is spending at a pace never seen before on AI data centers, and investors are losing patience with the bill. A large part of the demand that is supposed to pay it back comes from a handful of AI labs, and the biggest of them to open its books so far is Anthropic. Its filing shows how the money moves around, and how much of it moves in a circle.
The week the AI trade cracked
The Nasdaq Composite, the index heavy with technology companies, ended October with its worst month since March 2025, Investing.com reported, "notching its worst monthly performance since March 2025." Alphabet, it added, "posted its worst intraday drop on record, losing roughly half a trillion dollars in market value." On Friday, Intel's shares fell 28.6% after it warned that a memory shortage would keep shrinking PC sales. Meta's shares had fallen 7.0% the week before, on the first trading day after it raised its spending plans.
What worried investors, according to Investing.com, was not the size of AI demand but "the sustainability of issuing debt and equity to fund the buildout instead of cash flows." Amazon said its free cash flow had turned to an outflow of $42.4 billion over the past twelve months, and that it still expects to spend about $220 billion on buildings and equipment this year. Microsoft was the exception: it "left its AI spending strategy unchanged and delivered solid growth in its Azure cloud business," Investing.com reported, and its shares slipped only 0.2% after its results.
The company pages on this site show how fast the bill has grown. Alphabet, Amazon, Microsoft and Meta together spent $408.9 billion on property and equipment in their latest full fiscal years, according to their annual SEC filings. Two years earlier the four spent $140.1 billion, so the total grew 2.9 times.
Source: SEC EDGAR XBRL data
What Anthropic's filing shows
Anthropic is the other side of that spending, and its growth is the strongest argument that the spending will pay off. Its revenue went from $41 million in 2023 to $386 million in 2024 and $4.6 billion in 2025, the S-1 shows. In the first six months of 2026 alone it took in $16.3 billion. In the three months to June it reported revenue of $11.6 billion, up from $787 million a year earlier, and its first quarterly operating profit among the periods the filing shows: $1.0 billion, an operating margin of 8%.
The losses are just as striking. Anthropic lost $42.0 billion in 2025 and $80.7 billion in the first half of 2026. Most of that is accounting: the filing ties it chiefly to the rising value of convertible notes, debts that turn into shares, as investors valued the company higher. That is a paper loss, not cash out the door. The commitments are cash, and they are large.
The S-1 lists them one by one. Besides Google and Amazon, Anthropic has committed at least $44.9 billion to Microsoft. It agreed to spend up to $120.5 billion with SpaceXAI, formerly xAI and now a subsidiary of SpaceX, for computing powered by Nvidia chips: about $3.1 billion a month once all of it is delivered. Leases of chip systems from Broadcom add rent of about $161.2 billion. The filing leaves its grand total blank for now. Fortune, adding up the document, reported that Anthropic has committed to spending between $542 billion and $567 billion on future computing and infrastructure.
The filing is blunt about what that means. "Our compute commitments are substantial and generally must be satisfied regardless of our revenue or utilization," it says, while its access to more money depends on conditions "outside our control, including market conditions and investor demand." It also warns that it expects its growth rate "to decline over time, and it may decline significantly and more rapidly than we anticipate."
A circle of money
Here is where the two stories meet. "Amazon, Google, and Microsoft, through their respective affiliates, have made significant equity investments in our company," the S-1 says. "These CSPs are also our customers," it adds, using its term for cloud service providers. And: "a portion of our compute and infrastructure build-out has been funded or supported by capital or credit support provided or facilitated by certain of our CSPs and chip partners."
In plain terms: the cloud giants invest in Anthropic, Anthropic promises to spend hundreds of billions with the cloud giants, and those promises show up in the cloud giants' order books, the backlogs investors watch as proof that the data centers will be used. Google Cloud's backlog, contracts signed but not yet billed, was $560.8 billion at the end of September, Alphabet's quarterly report shows. Amazon's results carry the same loop: its profit a year ago was lifted by pre-tax gains of $9.5 billion "from our investments in Anthropic," its earnings release says. Amazon's notes in Anthropic can't convert into a stake of more than 12% of the company unless Amazon waives that limit, the filing shows.
The cloud businesses are growing fast. Google Cloud "revenues grew 103%," Alphabet said. Amazon Web Services' sales grew 39% to $46.0 billion. Microsoft Cloud revenue reached $63.9 billion, with growth of 30%. "Almost as fast as we install capacity, customers are gobbling it up," Amazon chief executive Andy Jassy said on his company's call, according to MarketBeat.
Skeptics read the same contracts differently. "Investors had become increasingly concerned about uncertain returns from the billions of dollars" going into AI, Investing.com wrote, and a backlog built partly on the promises of companies the cloud giants have invested in is exactly the kind of return that is hard to judge from outside. Anthropic itself calls its relationships with the cloud giants "multifaceted and complex."
Source: EDGAR full-text search
AI was also the talk of the filings. Of the week's 1,691 filings on Form 8-K, the report companies file for major news, 316 filings mentioned artificial intelligence, EDGAR's full-text search shows: 18.7% of them, against 9.5% in a usual week. Part of that is the calendar, because earnings releases, which talk about AI, crowd the 8-Ks in a week like this one. But the rate has risen for two weeks in a row.
Our read: the IPO is a test of the whole loop
This is analysis. Anthropic's listing, under the ticker ANT on the Nasdaq, is more than one company's debut. It is the first time public investors can put a price on the end customer of the AI build-out, the business whose sales are supposed to justify the data centers. As Yahoo Finance put it, "The company's market debut will test investors' appetite for a second public listing by a firm with an already massive valuation."
The bull case is in the numbers: a business that went from almost nothing to an operating profit in about three years, with demand its suppliers say they can barely keep up with. The bear case is in the structure. If Anthropic's growth slows the way its own filing warns it might, the contracts don't shrink with it, and the companies that would feel it first are the ones that both sell it computing and own part of it.
We think the signs to watch are plain. If the cloud giants' backlogs keep growing while their outside customers, not just the AI labs they invest in, sign up too, the circle is a flywheel. If the backlogs lean ever more on a few labs, and those labs need fresh money from the market to pay, the circle is a loop that only holds while the money keeps coming. This week, for the first time, investors can read the contracts and decide which one it is.