Anthropic’s Giant Computer Deals Are a Bet That Cuts Both Ways
It has promised to pay for vast computing power, but customers may not need it or builders may not finish it.

The AI Daily Brief: Artificial Intelligence News and Analysis · Wall Street Has No Idea What Anthropic Is WorthTo run its AI models, Anthropic has promised to rent staggering amounts of computing power years in advance, and much of that promise cannot easily be canceled.
Recent IPO paperwork shows the shape of the bet. It listed $10 billion of additional promised payments to Microsoft and Google, covering seven years.
The same filing said a separate $35 billion deal with Lambda, a smaller cloud provider known as a neocloud, could be canceled after Lambda missed a deadline. Anthropic said it has the right to walk away but does not plan to do so now.
Think of a compute commitment like a long lease on an office tower that has not been built yet. Anthropic agrees to pay for server space and chip-filled data centers, and the builder borrows money and starts construction based on that promise.
That structure creates risk in two directions. If business demand for AI slows, Anthropic could be stuck paying for computer capacity it no longer needs, a concern analysts raised when they first saw promised computing bills reaching into the hundreds of billions of dollars and largely impossible to cancel.
If demand stays strong, the danger flips to the builders. Some observers argue the larger worry is that neoclouds will hit financing or construction trouble and fail to finish the data centers on time.
The Lambda episode shows how that second risk looks in practice. A missed deadline gave Anthropic a legal reason to cancel, a reminder that the buildings behind the AI boom are not guaranteed to appear on schedule.
Questions people ask
What is a neocloud?
A neocloud is a smaller, newer cloud company that specializes in renting out AI computing power. Unlike Microsoft, Google, or Amazon, it usually owns a few large data centers and raises heavy debt to build more, which makes it more exposed if costs rise or deadlines slip.
Why does Anthropic rent computing power instead of building its own data centers?
Building data centers takes enormous capital, land, electricity, and time. Even large AI labs generally prefer to sign long leases with cloud providers rather than tie up their own money in construction and power contracts.
What usually happens when a big compute deal falls apart?
It depends on the contract, but the renter often still owes money or pays a penalty, while the builder loses expected revenue. In Anthropic's Lambda case, the filing said Anthropic was within its rights to cancel because of missed performance, and that it did not currently intend to do so.

