FULL STORY

Anthropic's Compute Binge: From Rentals to $517 Billion in Deals

After reports of a $45B Nscale rental deal and a $35B Lambda cloud pact, The Information estimates Anthropic has signed over $517B in compute agreements, locking in 14.8GW of capacity.

2026-08-27 ~ 2026-09-14 · 4 episodes · 28 posts

Episode 1 · Anthropic Reportedly to Pay Nscale $45 Billion Over Six Years for 460MW of Nvidia Vera Rubin Compute (2026-08-27, 9 posts)

On August 27, Bloomberg reported that Anthropic has agreed to pay data center operator Nscale $45 billion over six years for roughly 460MW of capacity at its West Virginia campus, to be powered by Nvidia Vera Rubin chips and expected online by late 2027.

Confirmed

  • $45 billion over 6 years, 460MW capacity, per Bloomberg; multiple outlets relayed the report
  • Facility located at Nscale's West Virginia campus, using Nvidia Vera Rubin architecture
  • A calculation relayed by @zephyrz9 puts the implied cost at about $5.84 per usable GPU-hour ($6.87 at 85% utilization)
  • @rohanpaulai noted the average annual spend mainly reserves scarce compute capacity

Unconfirmed

  • FirstSquawk initially framed it as market speculation; final signing and contract terms await official confirmation
  • A few posts (e.g., BethKindig and parts of rohanpaulai's text) cited $4.5 billion, inconsistent with Bloomberg's $45 billion figure
  • Ed Zitron criticized Anthropic for spending on chips not yet shipped and data centers not yet built, allegedly to support an IPO; this is personal opinion, with no response from Anthropic and no official word on any IPO

Why it matters

  • The $45 billion, 460MW scale shows top AI labs' long-term compute lock-in has entered a multi-billion-dollar-per-year race, giving emerging operators like Nscale major backing
  • If IPO speculation holds, such infrastructure commitments could anchor valuation narratives while drawing more scrutiny over capex sustainability

Episode 2 · Anthropic Signs $35B Cloud Compute Deal with Nvidia-Backed Lambda (2026-09-01, 11 posts)

According to the Wall Street Journal, Anthropic has signed a $35B cloud computing deal with Nvidia-backed Lambda to expand compute for AI training and inference, with Lambda renting out capacity at a Texas datacenter. This marks another massive compute commitment by Anthropic, with Nvidia deeply embedded in the deal chain across multiple roles.

Confirmed

  • The $35B deal with Lambda was reported by the WSJ and corroborated across multiple independent posts (@rohanpaulai, @beffjezos, @afurgs).
  • The structure is a vendor-financing loop: per @unconventionalbook, Nvidia signed a 15-year master lease with Lambda, enabling the startup cloud provider to take on the large contract; Nvidia is simultaneously an investor in Lambda, the GPU supplier, and holder of the Texas datacenter lease.
  • Per @beffjezos and @rohanpaulai, the Texas campus is leased by Nvidia from Hut 8, which builds the facility; the campus is roughly 700MW, and Hut 8 had previously disclosed a 15-year lease on it.
  • In early August, Anthropic reportedly also signed a $45B deal with another Nvidia-backed cloud provider, Nscale, bringing its monthly cloud capacity commitments to roughly $80B (@kimmonismus, @eyishazyer).

Unconfirmed

  • @emmanuelvivier and @pstAsiatech note the deal had not been officially confirmed by Anthropic or Lambda at the time, remaining at the level of media reporting and unnamed sources.

Why it matters

  • Nvidia's multi-role involvement—investor, chip supplier, and leaseholder—makes it the central actor in a vendor-financing model of compute expansion.
  • Anthropic's consecutive mega-deals show the GPU compute race among top AI labs has reached hundred-billion-dollar scale, pulling in infrastructure firms like Hut 8.
  • @eyishazyer points out that even as consumer users complain about Claude's reliability, usage limits, and overall product experience—with some posting cancellation screenshots—Anthropic is betting big on enterprise and AI coding agents via massive compute investment.

Episode 3 · Anthropic Locks In Over $517B in Compute Deals, 14.8GW of Capacity (2026-09-07, 6 posts)

Per The Information's estimate, Anthropic has signed over $517B in compute agreements since October 25—nearly triple what it told investors—locking in at least 14.8GW of new capacity. Its annualized revenue reportedly tops $65B, exceeding OpenAI's $40B+. The figure has become a focal point for its IPO investors.

Confirmed

  • The Information's estimate, consistently relayed by multiple accounts (@Hesamation, @rohanpaulai, @stephpalazzolo): deals exceeding $517B total, roughly 14.8GW of capacity
  • The deals include roughly $300B in combined Amazon and Google compute commitments, plus a SpaceX agreement (amount unspecified in the posts)
  • Per @rohanpaulai, Anthropic's annualized revenue reportedly exceeds $65B, surpassing OpenAI's $40B+

Why it matters

  • @GaryMarcus notes the $517B is roughly a thousand times Anthropic's best (possibly only) profitable quarter, underscoring the gap between spending and profitability
  • @pstAsiatech calls the pace—$517B in 11 months—a 'moonshot-style expansion' far beyond industry norms, signaling compute is now the core resource and largest cost item in frontier AI competition

Episode 4 · Anthropic's Compute Commitments Reportedly Reach $517 Billion (2026-09-13, 2 posts)

Anthropic has signed compute deals worth up to $517 billion, far exceeding the $180 billion spending plan it previously disclosed to investors through 2029, The Information reports.