Models & Releases

Anthropic pays SpaceX $1.25B/month for Colossus 1; Musk can pull compute if Claude harms humanity

SpaceX IPO filing reveals $1.25B/month compute deal with odd 'harm humanity' clause.

Deep Dive

SpaceX's S-1 IPO filing revealed Anthropic is paying $1.25 billion per month for exclusive access to Colossus 1 — a 300MW data center in Tennessee housing over 220,000 Nvidia GPUs. The deal runs through May 2029. In a controversial clause, Elon Musk retains the right to revoke compute access if Claude is deemed to 'engage in actions that harm humanity.' This puts Musk in a dual role: competitor to Anthropic and critical infrastructure provider.

Meanwhile, Moonshot AI's Kimi K3 released open weights on July 27, wiping $314 billion from Anthropic and OpenAI pre-IPO valuations. The Future of Life Institute's AI Safety Index gave Anthropic the highest grade (C+), OpenAI and Google scored C, while Meta, xAI, and DeepSeek effectively failed. The White House also announced a 30-day pre-release AI review framework and plans to redirect $200B in federal R&D toward AI.

Key Points
  • Anthropic pays $1.25B/month for Colossus 1 compute (300MW, 220K GPUs) through May 2029, per SpaceX IPO filing.
  • Musk retains a clause to cut compute access if Claude 'harms humanity,' mixing competitor and provider roles.
  • Kimi K3 open-weight release erased $314B from Anthropic and OpenAI valuations; Moonshot paused new subscriptions.

Why It Matters

Musk controls both AI model competition and critical compute infrastructure, creating unprecedented leverage over rival Anthropic.

📬 Get the top 10 AI stories daily