WSJ reports Anthropic has agreed to a $35 billion cloud-computing deal with Lambda, the GPU cloud Nvidia backs; Reuters and Bloomberg each sourced the same deal yesterday. The structure is the story. Lambda rents the capacity to Anthropic, but Nvidia itself holds the lease on the Hut 8-built data center in Nueces County, Texas — Nvidia signed its own capacity deal with Hut 8 weeks ago, and Lambda plugs in chips it bought from Nvidia, its own investor. Anthropic has not confirmed any of it; treat the numbers as reported, like the Nscale line that preceded it.
Nvidia is monetizing the lease, not just the chip
What happened. Per WSJ, corroborated by Reuters and Bloomberg: a four-company chain. Anthropic (reported customer), Lambda (Nvidia-backed cloud provider), Nvidia (chip seller, Lambda investor, and the entity holding the facility lease), Hut 8 (builder in Nueces County, Texas). Reported value ~$35 billion — the same league as the 45 billion Nscale commitment from last week, and it is a lease, not an equity round.
Why it matters. For an operator the shift is in who gets paid twice. Nvidia has gone from supplying silicon to holding the property risk underneath it: a chip vendor taking the lease so a neocloud doesn’t have to means Nvidia captures the real-estate leg of frontier capacity on top of the GPU margin. That bends the economics of every GPU cloud you route against — the landlord now has a structural reason to favor its own tenants. This is Anthropic multi-homing hard off the hyperscalers ahead of a public-market debut, and the supply chain keeps acquiring a new middleman who is also the landlord.
What I’m watching
Anthropic has not confirmed, and no term/MW figure is public. Scorecard: does the roadshow disclose the deal’s capacity and term, and does the neocloud order book show up in Vera Rubin’s production ramp the way NVIDIA’s own guidance implied. Note for our ledger: this deal broke yesterday afternoon our time and cleared our 00:00 capacity-lane sweep unflagged — a first-report gap on our side, now corrected by this note.
Source: reuters.com, bloomberg.com, investing.com