Version 2

On May 6, 2026, Anthropic announced it had signed an agreement with SpaceX to access the full capacity of the Colossus 1 data center — over 300 megawatts and more than 220,000 NVIDIA GPUs — with availability expected within the month.[1] The deal, framed by Anthropic as a response to severe compute constraints driven by explosive demand, also includes expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity, a moonshot ambition that signals how aggressively Anthropic is expanding its infrastructure vision.[1] As a direct customer-facing benefit, Anthropic simultaneously doubled Claude Code's five-hour rate limits for Pro, Max, Team, and Enterprise plans, and removed peak-hours restrictions for Pro and Max subscribers.[1] Colossus 1 is an xAI facility; the partnership therefore pairs Anthropic — perhaps the AI lab most publicly committed to safety — with Elon Musk's infrastructure.

Commentary published the following day added significant texture to the announcement. Analysts note that Anthropic's explosive growth — ARR climbing from $9B to $44B in 2026, gross margins rising from 38% to over 70%, and a potential funding round at a valuation above $900B — explains the urgency behind this unusual arrangement.[2] Zvi Mowshowitz characterizes the deal as driven by 80x demand growth and observes that xAI appears to be pivoting away from frontier model competition and toward becoming a compute provider, retaining the larger Colossus 2 for its own training while leasing Colossus 1 to Anthropic.[2][3] This structural reframing of xAI's competitive role is significant context that Anthropic's official announcement did not surface.

Simon Willison raised two pointed concerns that have become the focal points of skeptical discourse.[3] First, Colossus 1 carries documented environmental baggage: its gas turbines were operated without Clean Air Act permits or pollution controls by classifying them as 'temporary,' a record that sits uneasily with Anthropic's stated values and the already-charged politics of AI energy consumption.[3] Second, and more structurally alarming, the agreement reportedly includes a clause giving Elon Musk unilateral power to reclaim the compute if Anthropic's AI 'harms humanity' — with Musk himself as sole arbiter of that determination.[3] Willison frames this bluntly as 'a new form of supply chain risk,' pointing to xAI's track record of giving developers only two weeks' notice before deprecating its own models as evidence the concern is not theoretical.[3] Anthropic's official announcement, by contrast, emphasized its broader multi-provider compute portfolio totaling over 10 GW and its stated commitment to partnering only with democratic countries with secure supply chains — language that reads as aspirational positioning in some tension with the Musk clause reality.[1]