Analysis
Anthropic signed a $35 billion computing deal with Lambda, the Nvidia-backed cloud provider, to secure roughly 350 megawatts of GPU capacity at a new data center in Nueces County, Texas, Bloomberg reported. The agreement, signed Monday, gives Anthropic dedicated training and inference capacity at a site being developed by Hut 8, a company that converted from cryptocurrency mining into AI data-center infrastructure.
Nvidia Is on Every Side of This Deal
The structure is unusual even by 2026's compute-deal standards: Nvidia holds the lease on the Nueces County facility itself, Forbes noted, which puts the chipmaker in the position of landlord, Lambda investor, and GPU supplier all at once for a single site. That circularity -- Nvidia financing the infrastructure that buys Nvidia chips that Nvidia then effectively leases capacity back from -- is the same structure critics have flagged in Nvidia's OpenAI and CoreWeave arrangements, and it means a meaningful share of Anthropic's $35 billion ultimately flows back to Nvidia's own balance sheet regardless of which cloud brand appears on the invoice.
“- This week's compute commitments -- $80B combined (Lambda plus Nscale).”
Eighty Billion Dollars in a Week
The Lambda deal doesn't stand alone. Pulse has tracked Anthropic's compute buildout closely this year:
- Nscale, West Virginia -- $45 billion, committed days before the Lambda deal, for a separate campus.
- TeraWulf, Kentucky -- $19 billion, 20 years, a lease Pulse covered in July alongside Anthropic's confidential IPO filing.
- Nexus Data Centers, Texas -- $15 billion, Google-backed financing for a separate Hubbard, Texas campus.
Add Lambda and Nscale together and Anthropic committed roughly $80 billion in cloud infrastructure inside a single week -- on top of the prior TeraWulf and Nexus deals from earlier this summer. That is not $80 billion of cash leaving the building; these are multiyear lease and take-or-pay compute commitments, closer in structure to a long-term supply contract than a capital expenditure line, but they still represent obligations a future public-company balance sheet will have to carry and disclose.
Numbers in Context
Anthropic told investors its annualized revenue run rate hit $65 billion at the end of July, Pulse reported, up sevenfold from about $9 billion a year earlier. Even against that growth, $80 billion of infrastructure commitments in a single week is a lot to underwrite:
- Anthropic revenue run-rate -- $65B in July 2026, versus roughly $9B a year earlier.
- This week's compute commitments -- $80B combined (Lambda plus Nscale).
- OpenAI's compute forecast -- roughly $750B in multiyear spending through 2030.
- OpenAI's Ohio facility -- a separate $250B Nvidia-backed buildout.
Both labs are now running infrastructure math that assumes AI demand several multiples larger than what either currently monetizes -- a bet that compounding compute commitments will be matched by compounding revenue, not merely that current revenue holds steady.
The competitive read: this is Anthropic racing to lock in power-constrained real estate and Nvidia silicon before OpenAI, Google, or Meta can claim the same sites, not a response to any near-term capacity shortfall Claude is actually hitting today. Whoever secures grid-connected megawatts first in a market where Texas just froze new data-center hookups over "ghost demand" concerns effectively locks out competitors from that same physical footprint for years.
What the total misses: none of this compute is delivered yet. The Nueces County site is still under construction, full buildout on Anthropic's other major leases runs through 2028, and every one of these numbers is a forward commitment contingent on financing partners -- in some cases banks lending against credit Anthropic doesn't yet formally carry on an audited balance sheet. A $1 trillion-plus IPO valuation and $80 billion of infrastructure promises in a week are both, in a real sense, bets on the same unproven trajectory continuing uninterrupted for years.
