Anthropic's $895 Billion AI Bill Turns IPO Into Big Tech Referendum

Anthropic has committed to pay as much as $895.4 billion for AI computing power and infrastructure through 2045, according to its IPO prospectus. At least $276 billion of that total is locked in regardless of how much computing capacity the company actually uses. Those numbers, disclosed in filings ahead of a potential $100 billion-plus listing, make the Claude maker's public debut as much a stress test of Big Tech's AI business models as of Anthropic itself.

By Olivia Hart

Who is on the hook for Anthropic's compute bill?

Anthropic has become a major customer of half of the "Magnificent Seven" technology group. The company is obligated to pay Alphabet, Amazon.com and Microsoft tens of billions of dollars each for cloud capacity over the next several years, per its prospectus. It also consumes Nvidia chips through third-party deals with SpaceX and Nscale.

The exposure runs in both directions. SpaceX reportedly counts Anthropic as its biggest customer, representing 34.5% of revenue in the third quarter, according to a filing. Anthropic separately plans to buy or lease up to $271.3 billion worth of Google's tensor processing units through 2033. In plain terms, a single AI startup has become a load-bearing pillar of several trillion-dollar revenue plans — a concentration risk none of the parties names directly.

How fast is revenue actually growing?

Anthropic generated third-quarter revenue of nearly $17 billion, a 1,127% jump from a year earlier, according to an updated prospectus filed Monday. The sequential picture is less flattering: sales rose 47% from the second quarter of 2026, down from 144% growth between the first and second quarters.

“If demand continues to expand rapidly, those investments could support additional growth and improving economics,” the IPO research firm Renaissance Capital said in a note. “However, if adoption slows or computing technology changes, Anthropic could face substantial fixed obligations that are harder to offset.” That second clause is the one investors should read twice: $276 billion in take-or-pay-style commitments do not shrink if demand does.

What worries the analysts?

William Blair analyst Arjun Bhatia said investors fear revenue may not scale meaningfully enough over the years for Anthropic to meet its purchase commitments. Dauvin Peterson, an analyst at 22V, said the IPO's pricing and aftermarket performance will “depend heavily” on revenue trends in 2027.

“Our current view is that Anthropic has a significant hill to climb,” Peterson wrote in a client note, pointing to the company's need to find stickier revenue channels such as cybersecurity. Anthropic also relies heavily on a few partners — including Microsoft and Google — for nearly half its sales; 44% of revenue in the first nine months of 2026 came from customers routed through cloud partners. Anthropic itself, along with some experts, acknowledges that arrangement could create conflicts of interest as the company works to accelerate sales.

What are Alphabet and Amazon standing to gain?

Anthropic is seeking to raise as much as $100 billion or more, which could value the company at roughly $1.5 trillion, according to the Wall Street Journal. If achieved, that would surpass the record SpaceX set in June when it raised $75 billion.

The capital would help Anthropic fund its partner commitments. A successful listing would also deliver a windfall to its biggest backers. As of Sept. 30, Amazon held 164 million shares of Anthropic nonvoting preferred stock and convertible notes — a position it values at roughly $146.2 billion — though its stake is capped at 12%. Google holds the right to buy up to $10 billion of stock in the IPO, with its ownership capped at 14.99% of outstanding shares, according to a filing.

The takeaway: Anthropic's valuation case rests on hypergrowth persisting long enough to outrun nearly $900 billion in commitments, and its underwriters' largest potential buyers are also its largest suppliers.