Anthropic has struck a deal with SpaceX to take the entire compute capacity of its Memphis, Tennessee data center, unlocking more than 300 megawatts of new capacity and 220,000 NVIDIA GPUs to feed Claude Code. CEO Dario Amodei announced the agreement on stage at the Code with Claude 2026 conference in San Francisco on Wednesday, framing it as the reason Anthropic could finally raise usage limits that had been throttling its most demanding subscribers. The same day, the company doubled Claude Code's five-hour window limits for Pro and Max plan users, removed peak-hours reductions on Claude Code, and lifted API limits on its Opus model.
The Memphis facility is built around SpaceX's Colossus 1 supercomputer. "Colossus 1 features over 220,000 NVIDIA GPUs, including dense deployments of H100, H200, and next-generation GB200 accelerators," SpaceX wrote in its announcement. Routing all of that capacity to a single customer — Anthropic — is an unusual concentration of supply in a market where every frontier lab is rationing tokens.
The two companies also said Anthropic "expressed interest" in working with SpaceX on multiple gigawatts of orbital compute capacity. Anthropic's stated rationale is that the "compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter." Orbital data centers remain unproven, but the language signals where Anthropic thinks the supply ceiling lives.
Key facts
- 01Anthropic gets the entire compute capacity of SpaceX's Memphis data center, more than 300 megawatts.
- 02Colossus 1 packs over 220,000 NVIDIA GPUs spanning H100, H200, and GB200 accelerators.
- 03Anthropic doubled Claude Code's five-hour window limits for Pro and Max subscribers on May 6, 2026.
- 04The two companies discussed building multiple gigawatts of orbital compute capacity together.
- 05Deal follows recent compute agreements Anthropic signed with Microsoft, Google, Amazon, and Nvidia.
Demand for Claude Code has been climbing fast. Anthropic credits a mix of factors: developers moving off OpenAI after controversy over its US military agreements, faster adoption inside professional software shops, and a workload shift from single-agent chat sessions to multi-agent workflows that burn through tokens at a much higher rate. The result has been outages and rationing.
“Colossus 1 features over 220,000 NVIDIA GPUs, including dense deployments of H100, H200, and next-generation GB200 accelerators.”— Jaeden Schafer
The company's responses have not all gone over well. Anthropic introduced peak-hour usage limits and ran a short-lived trial that appeared to test removing Claude Code from the $20/month Pro plan entirely. Hacker News, Reddit, and X have been a steady drumbeat of developer frustration about hitting walls mid-task.
Last month, Anthropic reportedly signed compute deals with Microsoft, Google, Amazon, and Nvidia to address the gap. The SpaceX agreement is the most concrete and immediate of those, but the company says capacity from the others will take time to come online. The Memphis cluster is the supply unlock that lets Anthropic actually loosen the dial today rather than promising relief later.
The political subtext is hard to ignore. Until very recently, Elon Musk was openly hostile toward Anthropic. In February 2026 he declared on X that "Anthropic hates Western Civilization," amplifying a false claim from Trump administration official Emil Michael about Anthropic's constitution for Claude. The deal would have been unthinkable based on that posture.
Musk's tone shifted in the lead-up to Wednesday's announcement. "I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed," he wrote on X. "No one set off my evil detector." Whether that reflects genuine conversion or the gravitational pull of a customer willing to absorb 300 megawatts of capacity is left to the reader.
There are real caveats. Orbital compute is years away at best and may never pencil out economically. The Colossus 1 deal also raises a concentration question: Anthropic is now leaning on infrastructure controlled by a CEO who, two months ago, was publicly accusing it of civilizational hostility. xAI, Musk's own AI company, is a direct competitor to Claude. That counterparty risk is not trivial.
Still, this is the clearest signal yet that the binding constraint on frontier AI in 2026 is not algorithms or talent but megawatts and GPUs. Anthropic is willing to do business with a hostile-turned-friendly Musk to get them, and SpaceX is willing to rent out an entire 220,000-GPU cluster to a single tenant. The labs that solve the power and chip math first will set the pace, and for the next quarter at least, Claude Code subscribers are the ones who notice.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
