Three weeks before it is expected to begin marketing one of the largest technology IPOs in history, Anthropic is quietly converting itself from a tenant into a landlord. The maker of Claude has signed more than a dozen letters of intent to lease US data centers with a combined capacity of over one gigawatt, according to a report from The Information, a move that for the first time gives the company direct control over the physical infrastructure it has so far rented from cloud giants. The non-binding agreements arrive as Anthropic races toward an October 2026 public listing at a valuation now pegged near $965 billion, and they underscore a single, unrelenting reality of the frontier AI business: the appetite for compute has outrun anyone's ability to buy it off the shelf.
From Renter to Owner
Until now, Anthropic's strategy has been to rent, and rent enormously. The company has committed to more than 10 gigawatts of cloud capacity across deals that read like a who's who of the infrastructure economy: a $200 billion arrangement with Google, an $18 billion contract with Akamai, a 5 GW capacity agreement tied to Amazon's $5 billion investment, a multi-year deal with CoreWeave, and a $50 billion partnership with AI cloud firm Fluidstack spanning sites in Texas and New York.
The newly reported leases mark a deliberate pivot toward a hybrid model in which Anthropic owns more of its stack to drive down long-term compute costs while still tapping cloud partners for elastic capacity. The letters of intent are non-binding, giving the company room to perform due diligence on each developer and site before committing to multi-decade obligations. The locations and developers have not been disclosed.
Crucially, Anthropic is not shouldering the credit risk alone. According to The Information, the company is in talks with Google for the search and cloud giant to provide a financial guarantee on its lease payments, an arrangement that would let landlords and their lenders price the deals against Alphabet's roughly $4.3 trillion balance sheet rather than a pre-IPO startup's. "Anthropic is worth $965 billion and still needs Google to pay its rent," ran one headline at BeInCrypto, capturing the irony of a company valued near a trillion dollars leaning on a backer to secure favorable terms.
The $1.25 Billion-a-Month Backstop
The leasing push sits atop an already staggering compute bill. Anthropic is paying SpaceX roughly $1.25 billion a month to rent the entirety of Colossus 1, the Memphis supercomputer originally built for Elon Musk's xAI, in a deal that runs through May 2029 and totals more than $40 billion. The facility houses over 220,000 Nvidia GPUs, a mix of H100, H200, and next-generation GB200 accelerators delivering some 300 megawatts of compute.
In its announcement of the partnership, xAI described Colossus 1 as "one of the world's largest and fastest-deployed AI supercomputers," adding that "Anthropic plans to use this additional compute to directly improve capacity for Claude Pro and Claude Max subscribers." Reports suggest SpaceX rented the site to a rival in part because it struggled with latency issues linking the Memphis campus to other facilities more than 10 miles away, an awkward marriage of competitors that Musk reportedly waved through with the quip that "no one set off my evil detector."
Why This Matters
The numbers behind the buildout explain the urgency. Anthropic's run-rate revenue crossed roughly $47 billion this month, up from about $9 billion at the end of 2025 and $30 billion in March, a growth curve that few companies in history have matched. That demand has to land somewhere physical, and terrestrial power, land, and cooling are increasingly the binding constraint. xAI put the problem bluntly in its own statement: "The compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter." That same logic is what is pushing Anthropic to lock down gigawatts of capacity now, before its competitors and the IPO clock force its hand.
The timing is conspicuous. Anthropic confidentially filed its S-1 with the SEC on June 1, 2026, shortly after closing a $65 billion Series H round at a $965 billion post-money valuation, with Goldman Sachs, JPMorgan, and Morgan Stanley reportedly steering a listing expected to raise more than $60 billion. Locking in long-term, owned capacity is precisely the kind of margin-of-the-future story that public-market investors will scrutinize.
The Export-Control Shadow
There is a sharper tension lurking beneath the expansion. Even as Anthropic pours capital into US compute, the federal government has just demonstrated how quickly it can switch that compute off. On June 12, the same day the data-center leases surfaced in reporting, Anthropic disabled its flagship Fable 5 and Mythos 5 models worldwide after the government issued an export-control directive barring access by any foreign national, citing a purported jailbreak.
Anthropic pushed back hard. "We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company wrote. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers." The episode crystallizes the paradox of Anthropic's moment: it is spending tens of billions to expand US compute while Washington reserves the right to restrict who its most powerful models can serve, a regulatory wildcard that no balance sheet can fully hedge.
What to Watch
Three things will tell the story from here. First, whether the non-binding letters of intent convert into signed long-term leases, and whether Google's guarantee materializes on terms favorable enough to justify the strategy. Second, how Anthropic frames its compute obligations and its concentration risk with SpaceX and Google in the public S-1 when it flips from confidential to disclosed. And third, whether the Fable 5 suspension proves a one-off misunderstanding or the first of a recurring pattern of government intervention that reprices the entire frontier-model business. For now, Anthropic is betting that owning the picks and shovels is the surest path through an October IPO and the compute crunch beyond it.
“The compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter.”— xAI, Statement on Colossus 1 compute partnership