Anthropic Locks In $35 Billion of Nvidia-Backed Compute
Anthropic has signed a $35 billion cloud-computing deal with Lambda, an AI provider backed by Nvidia, to bring new Nvidia capacity online for Claude. The deal centers on a Texas data center whose lease Nvidia holds, with Hut 8 developing the site. It is the latest sign that frontier AI capacity is concentrating in a handful of hands.
On this page
The deal and what it buys
Anthropic has signed a $35 billion cloud-computing agreement with Lambda, the AI cloud provider backed by Nvidia, according to reports from Bloomberg, the Wall Street Journal, and Reuters [1][2][3]. The agreement centers on a new data center in Texas, and the structure is unusual: Nvidia itself holds the lease on the facility, while Hut 8, a company that pivoted from bitcoin mining into AI infrastructure, is developing the site [2][3].
A source cited by Reuters said the deal will "bring online Nvidia capacity to meet growing demand" for Anthropic's Claude assistant [1]. In plain terms: Claude's usage is growing faster than the capacity Anthropic can rent through ordinary channels, so the company is reserving an entire facility's worth of future supply before it exists.
Nvidia's new role: chip seller to capacity landlord
The structure matters as much as the size. Nvidia is no longer just selling chips into this build-out; by backing Lambda and holding the lease, it is financing and controlling the capacity that its largest customers depend on [3][4]. Coverage of the deal describes it as evidence of Nvidia's expanding role in AI infrastructure financing, from component vendor to landlord of the compute layer itself [4].
That closes a loop worth noticing. The same company designs the accelerators, finances the facilities that house them, and leases the resulting capacity back to the labs that buy them. Each step is rational on its own; together they tighten the ties between every layer of the AI stack [3][4].
Concentration keeps compounding
This is not an isolated signature. The Wall Street Journal describes it as part of a string of similarly large compute commitments by Anthropic, and it arrives days after the Bank of England's chair warned the G20 that concentrated AI capacity and valuations could amplify a future shock [2]. The pattern is consistent across the industry: frontier labs are signing very large infrastructure commitments against demand that does not exist yet.
For the people who build on these systems, the practical effect is straightforward. Capacity at the frontier is becoming a contract-negotiated commodity, allocated first to the customers with the largest agreements. Everyone else rents what remains, on terms set after the big checks have cleared [1][4].
What this means for capacity, prices, and local AI
Two takeaways for readers outside the enterprise arena. First, expect frontier model pricing and availability to track these contracts: capacity reserved for Claude's biggest buyers is capacity that does not spot-market its way to small developers, and financed build-outs eventually want repayment through usage [1][2].
Second, the independent path keeps getting comparatively more attractive. A model running on your own hardware answers to no data center lease, no financing vehicle, and no allocation queue. Deals like this one are how the cloud side of the industry competes for the future; running a capable model locally is how you opt out of that particular race entirely [4].