What happened: Anthropic has signed a reported $35 billion cloud-computing agreement with Lambda, the Nvidia-backed AI cloud provider, according to sources cited by the Wall Street Journal and confirmed by Reuters. The capacity sits in a Texas data center under development by Hut 8 in Nueces County — roughly 350 MW of planned power.
The structure is the story. Nvidia holds the lease on the facility, Lambda deploys Nvidia chips into it, and Anthropic contracts for the resulting compute. Nvidia is an investor in Lambda. The same pattern appeared in Anthropic’s recent $45 billion Nscale commitment in West Virginia. Capacity is being secured at multi-year, multi-tens-of-billions scale.
Primary sources: Reuters reporting, deal structure analysis, and earlier context on Nvidia’s ecosystem moves including the MediaTek NVLink Fusion investment.
Secure the power. Secure the chips. Then ship the models.
Why it matters
Claude demand — especially coding and agent workloads — is forcing Anthropic to lock capacity years ahead of delivery. The company is preparing for a public listing and cannot afford to be power- or GPU-constrained while competitors keep spending. The Lambda deal is one more data point that the bottleneck has moved from “can we train” to “can we keep serving at scale without rationing.”
Nvidia’s role remains central. By investing in the cloud layer, leasing the shells, and supplying the silicon, the company keeps demand for its own hardware tightly coupled to the growth of the model labs. That is not new. The scale of the numbers is. Hundreds of megawatts and tens of billions in contracted spend are now routine headlines for a single lab.
For builders watching the stack, the practical takeaway sits alongside earlier notes on the AI content stack: the tools you use run on infrastructure that is being financed and controlled at a level most product teams never see. Capacity scarcity and power constraints are structural, not cyclical noise.
What to watch next
Watch whether official confirmation or term sheets surface from any of the parties. Watch power and interconnection timelines for the Texas site. Watch how Anthropic balances Nvidia-centric capacity against any TPU or alternative silicon paths. And watch the regulatory conversation around circular financing structures that place the chipmaker at multiple layers of the same transaction.
Infrastructure is the real product right now. Models sit on top of it. The $35 billion number is large. The pattern is larger.
Not a spectator. A builder.