cyberivy
AnthropicClaudeSpaceXxAIColossus 1AI InfrastructureData CentersNVIDIA GPUsMemphis2026

Anthropic rents SpaceX Colossus 1: 300 megawatts for Claude

May 8, 2026

On May 6, 2026, Anthropic takes over the full AI compute capacity of SpaceX's Colossus 1 data center in Memphis: 300 megawatts, roughly 220,000 NVIDIA GPUs, and double the Claude Code limit.

Anthropic locks down 300 megawatts from SpaceX's Colossus 1 for Claude

On May 6, 2026, Anthropic and Elon Musk's SpaceX announced a compute agreement that gives the Claude maker exclusive access to the entire AI compute capacity of the Colossus 1 data center in Memphis, Tennessee. Anthropic and xAI both put the new headroom at more than 300 megawatts of electrical capacity, equivalent to over 220,000 NVIDIA GPUs that come online within the next month. Anthropic is using the capacity to immediately raise limits on Claude Code and on paying Claude Pro and Claude Max plans.

Who runs Colossus 1 and why Anthropic is the tenant

Colossus 1 was originally built by xAI for its own Grok model. Earlier in 2026, Musk merged SpaceX and xAI, which is why the Memphis site now formally sits inside SpaceX. xAI is keeping its larger and newer Colossus 2 for its own training runs but is handing all of Colossus 1 to Anthropic. Bloomberg, CNBC and Data Center Dynamics all describe it as a straight rental of full capacity – a deal that quietly inverts several competitive lines. Anthropic and xAI compete head to head on frontier models, and Musk and Anthropic CEO Dario Amodei publicly hold very different positions on AI safety.

300 megawatts now, gigawatts in orbit only as an aspiration

The 300 megawatts are due to come online in production through May 2026, according to Anthropic. The company calls this equivalent to 220,000 NVIDIA GPUs. Beyond Memphis, Anthropic and SpaceX have, per CNBC, sketched a much more speculative second leg: jointly building multi-gigawatt AI compute capacity in Earth orbit. That orbital piece is explicitly not a contract – it is an "expressed interest" without a timeline or a fixed cost estimate.

What Claude users will see in the short term

On its own blog, Anthropic spells out the near-term effects: higher hourly limits in Claude Code, removal of the previous peak-time throttling and raised API limits for its top model Claude Opus. Pro, Max and Team subscribers should get more daily requests at the same price.

Why this matters

The deal shows how directly access to power and GPUs now decides the market position of big AI vendors. According to Bloomberg, Anthropic is planning a funding round that would value the company at around 900 billion dollars – but even a round that size is not enough to stand up hyperscaler-grade capacity from scratch fast enough. For companies running Claude in production, that means the supply chain of their AI vendor is becoming more political, more energy-bound and more capital intensive, even though end prices have not moved up. The arrangement also raises regulatory questions at a time when U.S. agencies are pushing for tighter pre-release model checks and CAISI is already evaluating frontier systems before public launch.

Practical example

A mid-sized software firm in southern Germany has been using Claude Code since 2025 with a team of 40 engineers. At peak hours – typically late morning – they hit hard limits, and the build-pipeline bot sat in queue for minutes. With the new limits from mid-May 2026 on, the same team can in principle send roughly twice as many coding requests per day at the same price. That makes agentic pipelines – automated pull request reviews, test generation – economically realistic without the firm having to switch vendor or buy its own GPUs in a German data center.

💡 In plain English

Anthropic, which makes the Claude chatbot, is renting a huge data center from Elon Musk's SpaceX. That lets Claude help more people at the same time without Anthropic having to build a new data center itself.

Key Takeaways

  • On May 6, 2026, Anthropic took over the full AI compute capacity of SpaceX's Colossus 1 data center in Memphis.
  • The 300 megawatts of new capacity are described as equivalent to roughly 220,000 NVIDIA GPUs coming online within a month.
  • Anthropic is using the capacity to double Claude Code limits and raise Claude Opus API limits without raising prices.
  • xAI is keeping the larger Colossus 2 for its own training and is giving only Colossus 1 to its direct rival Anthropic.
  • A second, looser piece of the deal is an expressed interest by Anthropic and SpaceX in building multi-gigawatt AI compute capacity in Earth orbit.
  • The deal lands while Anthropic is reportedly raising a round that would value it around 900 billion dollars.

Sources & Context