DEVOURED
Anthropic signed a deal for all 220,000 GPUs at SpaceX's Colossus 1 data center and publicly expressed interest in partnering with SpaceX to develop gigawatts of AI compute capacity in orbit.
What: Anthropic signed an agreement with SpaceX for all compute capacity at the Colossus 1 data center: over 220,000 NVIDIA GPUs (300+ megawatts) coming online within the month. The company doubled Claude Code rate limits, removed peak hours restrictions for Pro/Max, and raised Claude Opus API rate limits effective May 6, 2026. This adds to deals with Amazon (up to 5 GW, ~1 GW by end 2026), Google and Broadcom (5 GW starting 2027), Microsoft and NVIDIA ($30B Azure), and Fluidstack ($50B US investment). Anthropic also expressed interest in developing 'multiple gigawatts of orbital AI compute capacity' with SpaceX.
Why it matters: The speed of this deal—Anthropic acquiring SpaceX's entire Colossus 1 facility within a month—combined with the accumulation of multi-gigawatt partnerships years before capacity arrives, reveals acute compute constraints at the frontier of AI development. The orbital compute mention signals these constraints are severe enough to make space-based infrastructure a serious discussion topic, not science fiction.
Deep dive
- Anthropic signed an agreement with SpaceX for all compute at the Colossus 1 data center: 220,000+ NVIDIA GPUs, 300+ megawatts, coming online within the month
- Effective May 6, 2026: Claude Code rate limits doubled for Pro/Max/Team/Enterprise, peak hours restrictions removed for Pro/Max, Claude Opus API rate limits raised considerably
- The SpaceX deal adds to recent partnerships: Amazon (up to 5 GW, ~1 GW by end 2026), Google + Broadcom (5 GW starting 2027), Microsoft + NVIDIA ($30B Azure), Fluidstack ($50B US infrastructure)
- Anthropic runs Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs, continuing to explore additional capacity sources
- The company 'expressed interest' in partnering with SpaceX on 'multiple gigawatts of orbital AI compute capacity' (no firm commitment or timeline)
- International expansion planned via Amazon collaboration: additional inference in Asia and Europe for regulated industries (financial services, healthcare, government)
- Anthropic is partnering only with democratic countries whose legal and regulatory frameworks support large-scale investments and have secure supply chains (hardware, networking, facilities)
- The company committed to covering consumer electricity price increases from US data centers and is exploring extending this to new international jurisdictions
Decoder
- AWS Trainium: Amazon's custom chip designed for training AI models, an alternative to NVIDIA GPUs
- Orbital AI compute: Space-based data centers that would orbit Earth, using solar power and vacuum for cooling (currently speculative, no commercial implementations)
Original article
Anthropic increased usage limits for Claude through a new compute partnership with SpaceX, accessing over 220,000 NVIDIA GPUs. This expansion follows deals with Amazon, Google, Broadcom, Microsoft, NVIDIA, and Fluidstack for significant compute capacity. The company also plans international expansion to address compliance needs for enterprise customers in regulated industries.
Explore all 73 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi