Anthropic has found another place to buy computing power, and this time it is a company barely old enough to have a logo. The AI developer has signed a $10bn, six-year deal with Volta, a cloud startup that only just launched, according to Bloomberg.

The agreement is Volta’s marquee contract and, for Anthropic, one more supplier on an already crowded list. It is the kind of deal that has become routine for a company whose appetite for compute seems to have no ceiling.

Anthropic now buys capacity almost everywhere it can. Its biggest backer, Amazon, has put up to $25bn into the company and supplies much of its compute, and that is only the start of the list.

The rest reads like a roll-call of the industry. Anthropic struck its biggest compute deal yet with Google and Broadcom as its run rate reached $30bn, and it has leaned heavily on Google’s TPUs. While, AMD is investing $5bn and deploying two gigawatts of GPUs to run Claude, while CoreWeave, Akamai, and even SpaceX have all signed on to provide compute.

Volta is not even the only $10bn deal in the mix. Anthropic has been in talks with Meta over a compute agreement of the same size, a measure of how many suppliers it takes to feed a single frontier lab.

What makes the Volta deal stand out is how new the counterparty is. Anthropic is committing $10bn over six years to a startup founded this year, valued at $2.4bn, and only just out of stealth.

Volta’s idea is to make Nvidia’s chips affordable to more than the giants, backed by Andreessen Horowitz, Altimeter, Nvidia, and Michael Dell. It plans to run Anthropic’s capacity partly through the bitcoin miner Bitdeer, starting at a site in Norway.

For Volta, the deal is validation and ballast at once. A $10bn anchor customer hands a days-old startup the credibility, and the cash flows, to raise and build against.

For Anthropic, spreading its compute across many suppliers is a form of insurance. No single provider can meet its needs, and depending too heavily on any one, including its own investors, carries risks of its own.

Diversifying is also about leverage; a lab that can shift workloads between Amazon, Google, AMD, and now a startup is less beholden to any one of them on price or priority.

Anthropic has even explored building its own chips, the logical end point for a company that has decided compute is too important to leave entirely to others.

The arrangements also feed a worry that has trailed the whole AI boom. Chipmakers and clouds increasingly help finance the customers who buy from them, an interlocking web that could magnify losses if demand disappoints.

The scale of it all is hard to fathom. Anthropic’s compute commitments now run to tens of billions of dollars, a bet that demand for Claude will keep climbing fast enough to justify the spend.

The spending tracks the business. Anthropic’s revenue has climbed sharply as Claude has won enterprise customers, which is what lets it sign commitments that would sink a smaller firm.

Committing $10bn to a days-old company is not without risk, either. Volta still has to build and deliver, and Anthropic is effectively wagering that a first-time operator can execute at a scale established clouds find hard.

The deal also hints at where the bottleneck now sits. Anthropic is chasing not just chips but the power and sites to run them, which is exactly what a firm like Volta, built by infrastructure financiers, is selling.

For a company racing to keep Claude ahead, the maths is simple. Capacity is the constraint, and Anthropic will take it wherever it can find it, even from a startup that did not exist a year ago.

Get the TNW newsletter

Get the most important tech news in your inbox each week.