Nvidia put a number on Anthropic’s appetite for computing power on September 28, and the number is measured in gigawatts. The chipmaker said Anthropic has signed up for 2.6 gigawatts of AI infrastructure that will be delivered through 2028, a scale of power that would have been unthinkable for a single customer only a few years ago.
The contracts that underpin that capacity now total more than $180 billion, according to Nvidia, and they run through a mix of established cloud providers and newer entrants built specifically to supply AI computing. Nvidia is in a position to know: it sells the chips that fill those data centers, and it is an investor in many of the companies doing the building.
The tally is Nvidia’s way of describing a market it is helping to create. The company also said one of its newer cloud provider partners in Australia is expected to deliver up to 2 gigawatts of computing capacity by 2027, a measure of how quickly the geography of AI infrastructure is expanding beyond the United States.
Anthropic has been racing to lock in capacity ahead of its expected initial public offering, and the individual deals behind the $180 billion total have come fast. In April it committed more than $100 billion to Amazon Web Services over the next decade. Since then it has signed a six-year, $35 billion deal with Lambda, a San Jose provider backed by Nvidia, and a $45 billion agreement with the London-based Nscale, among others. Most of those arrangements are difficult to walk away from once signed.
The structure of the newer deals shows how layered the market has become. In the Lambda case, the computing will sit in a data center in Texas being developed by Hut 8, a former bitcoin miner that turned to hosting. Nvidia leases the facility, Lambda pays Nvidia for access to the chips, and Anthropic buys the resulting capacity from Lambda. Each player takes a slice, and Nvidia sits in the middle of the transaction.
The Australian figure points to a second shift. Cloud computing was once concentrated in a handful of American regions, but AI has pushed providers to build wherever land and power are available. Australia, with its abundant energy resources, has become one of the new frontiers, and Nvidia is backing local operators to build there rather than waiting for American capacity to come online.
The scale becomes clearer when set against the rest of the industry. OpenAI and Nvidia announced a partnership to deploy at least 10 gigawatts of systems, a figure that made Anthropic’s 2.6 gigawatts look modest by comparison but still enormous in absolute terms. The race has moved from individual data centers to fleets of them, and the unit of account is now power, not servers.
Nvidia’s position in all of this is unusual. It is at once the supplier of the essential component, a landlord in some deals, and a shareholder in the companies on both sides of the transaction. The company said its investment portfolio now includes 13 publicly traded companies and 229 private companies, a sprawling set of stakes that tie its fortunes to the entire ecosystem it feeds. By disclosing the tally, the company is also making an argument about its own centrality: the $180 billion in commitments is a measure of demand flowing through its components, and a reason for investors to keep watching the chipmaker.
For Anthropic, the contracts are a bet that demand for its models will grow fast enough to justify the spending. Each gigawatt of capacity carries with it an obligation to pay, and the commitments now stretch for years. Analysts said the scale of the outlay has raised questions about whether the company can convert the capacity into revenue quickly enough.
The newer providers now account for a growing share of the total. Anthropic signed a roughly $45 billion deal with Nscale for capacity at a West Virginia campus, a $10 billion agreement with Volta Infra Holdings in Norway, and a 20-year lease with TeraWulf for computing load in Kentucky. Nvidia has backed several of these operators, betting that a new class of cloud company, built narrowly around AI workloads, can deliver capacity faster than the incumbents.
That is the point of Nvidia’s arithmetic. The company uses the term new cloud providers to describe this crop of startups, and it has spent the past two years seeding them with capital and chips. The $180 billion figure is, in its telling, proof that the strategy is working: the demand exists, and the supply is materializing around Nvidia’s own components.
Nvidia, for its part, benefits either way. Its chips sit inside the data centers no matter which model wins, and its investments give it a share of the upside when the builders succeed. The $180 billion tally is, in effect, a statement of how much of the AI economy now runs through a single chipmaker’s balance sheet.
