Anthropic’s cloud partnerships are getting messy

According to an S-1 filing leaked to The Wall Street Journal and summarized by TechCrunch, Anthropic is struggling to secure enough cloud capacity from AWS, Google and Microsoft, which increasingly see it as a competitor rather than a customer. The company says it could spend more than $750 billion on cloud contracts, some still under negotiation, and is leasing its own data centers and working with third-party operators it worries have a limited operating history.

The numbers are enormous on both sides. Anthropic reported $16.3 billion in revenue in the first half of the year against an $80.7 billion net loss, with two customers making up about a fifth of sales. It has told investors third-quarter revenue hit $16.8 billion, up 45 percent from the prior quarter. The company is reportedly aiming for a November IPO.

The filing also warns that damages from harmful AI behavior could exceed its insurance coverage. That is the most consequential AI lab in the market putting its biggest dependencies and its tail risk in writing.

We think the cloud story is more important than the loss number. Every frontier lab runs on rented compute from companies that also sell their own models and assistants. For years that looked like a comfortable arrangement: the hyperscalers got a marquee customer, the lab got scale. What the filing describes is that arrangement fraying. When your landlord is also your competitor, capacity becomes a negotiating weapon, and growth can be capped by someone else's priorities.

The revenue trajectory is real and impressive. A 45 percent quarter-over-quarter jump at this size says customers are paying for Claude at a pace few software businesses have ever seen. We do not want to wave that away. But revenue concentration, with two customers near a fifth of sales, combined with a cost base that produced an $80.7 billion half-year loss, means the business is exposed to a small number of relationships on both the buying and the supplying end.

The workaround, leasing its own facilities and leaning on younger data center operators, carries its own risk, and Anthropic says so. Newer operators may be less reliable. Outages or construction delays would land directly on a company whose product is now embedded in other companies' workflows.

Then there is the liability language. A frontier lab warning that harms from its models could outrun its insurance is unusually candid, and we credit it for that. It is also a reminder that public investors will now be pricing catastrophic-risk disclosures alongside growth. That could be healthy, forcing clearer accounting of safety risk. It could also create pressure to downplay those risks once quarterly expectations set in.

Where this leads if it scales: a handful of labs whose costs run through the same three clouds, each partly at the mercy of the others. I'm watching whether the IPO prospectus, once public, softens or sharpens these warnings.

An S-1 is the registration statement a company files with the SEC before going public. Reuters and the Financial Times reported on the filing's broad numbers in late September; the Journal's account adds detail on cloud capacity and customer concentration. Anthropic did not immediately respond to TechCrunch's request for comment.