A $965 Billion Lab Needs Google to Guarantee Its Rent

The most revealing number in AI right now isn't a model benchmark or a funding round. It's a credit guarantee. Anthropic, valued at roughly $965 billion after its latest raise and reportedly weeks from a public listing, has signed more than a dozen letters of intent to lease U.S. data center space totaling over 1 gigawatt of capacity. And to make those leases work, it is in talks with Google to financially backstop the payments. A near-trillion-dollar company is asking another company to co-sign its lease.

That is not a sign of weakness so much as a window into how the economics of the buildout actually function—and who, underneath the headline valuations, is carrying the risk.

What was signed

The agreements are letters of intent, not binding leases—non-committal preliminary deals that stake out capacity while terms get negotiated. There are more than twelve of them, and together they describe more than a gigawatt of power: enough to run a serious fraction of a frontier lab's training and inference workloads on infrastructure Anthropic would control directly, rather than renting elastically from a hyperscaler.

This is the strategic shift worth noticing. Anthropic has, until now, been a cloud-native lab—its compute came through partners, principally Google and Amazon, both of which are also investors. Signing direct data center leases moves it toward a hybrid model: own (or long-term control) the physical floor for baseline capacity, and keep tapping the clouds for elastic bursts. Every lab that scales eventually confronts the same math. Renting compute at hyperscaler margins forever is enormously expensive; controlling your own footprint trades flexibility for cost and certainty. Anthropic is making that trade.

Why Google has to sign

Here is the mechanism. A data center lease is a long-term, capital-heavy commitment—the landlord and its lenders are fronting hundreds of millions in construction and power infrastructure against a tenant's promise to pay for years. The landlord prices that promise on the tenant's creditworthiness. And for all its valuation, Anthropic is a company that loses enormous sums building toward a future it hasn't yet banked. Its paper worth is a bet; its balance sheet is not yet a fortress.

So Anthropic is reportedly asking Alphabet—a multitrillion-dollar company with one of the strongest balance sheets on Earth—to stand behind the lease payments. With a guarantor of that caliber, the landlords' risk premium collapses. Better terms, lower rates, more capacity unlocked. The guarantee is leverage in the literal financial sense: it converts Google's credit rating into Anthropic's negotiating power.

It also deepens an already dense entanglement. Google is an Anthropic investor. Google supplies Anthropic compute. Anthropic recently expanded a partnership with Google and Broadcom for multiple gigawatts of next-generation TPU-class compute. Now Google may guarantee the rent on the buildings that house the rest. Each thread individually makes sense. Together they describe a frontier lab whose independence is real on the model layer and increasingly conditional on the infrastructure layer.

The circularity problem, in one deal

The AI economy has been quietly accumulating these loops—chipmakers investing in the labs that buy their chips, clouds funding the labs that rent their clouds, guarantors backing the tenants they also supply. Anthropic's leases are a clean, small-scale illustration of the pattern. Money and risk circulate among a handful of giant balance sheets, and the question of who is ultimately exposed gets harder to answer with each new arrangement.

That isn't necessarily fragile. Pooling risk among well-capitalized players is how large infrastructure has always been financed—railroads, telecoms, power grids. But it does mean the often-repeated story of scrappy labs out-innovating incumbents is incomplete. At this scale, the labs and the incumbents are financially fused. Anthropic cannot build its physical footprint without Google's credit any more than Google can win the model race without Anthropic's research.

The IPO reads it differently

The timing sharpens the picture. Anthropic confidentially filed an S-1 on June 1, 2026, following a $65 billion raise led by Altimeter, Dragoneer, Greenoaks, and Sequoia at that $965 billion post-money mark, with reported annualized revenue near $47 billion and a listing targeted for the fall. A company about to ask public markets for capital is simultaneously asking a private partner to guarantee its rent.

To a skeptic, that's a flashing light: if the business were self-evidently durable, why would landlords need Alphabet's signature? To a builder, it's simply how you move fast without waiting for your own credit to catch up to your ambition—you borrow a stronger balance sheet now and grow into independence later. Both readings are true at once, which is exactly what makes the deal interesting.

What to watch

Letters of intent are not leases, and a guarantee under discussion is not a guarantee signed. The story to track is whether these convert into binding commitments, on what terms, and whether Google actually puts its name on the line—or whether Anthropic's impending public capital makes the backstop unnecessary. Watch, too, whether this becomes the template. If a $965 billion lab needs a guarantor, every lab below it will need one more.

The headline version of the AI race is about who has the best model. The version that determines who survives is about who can finance a gigawatt of concrete and power without blinking. On that question, Anthropic just told the market the honest answer: not yet, not alone. The buildout is real, the ambition is real, and the rent—for now—has a co-signer.