SpaceX Gives Anthropic Full Colossus 1 Access Before IPO

SpaceX granted Anthropic full access to its Memphis Colossus 1 supercluster on Wednesday, providing more than 220,000 NVIDIA GPUs and over 300 megawatts of capacity. The compute can be deployed within a month. The agreement arrived about five weeks before SpaceX’s planned $1.75 trillion initial public offering and followed a confidential S-1 filing with the SEC on April 1. The company’s roadshow is scheduled for the week of June 8.

Under the contract, Anthropic will expand its training and inference capacity immediately. Anthropic posted that the added compute lets it double rate limits for Claude Code paid users and raise API ceilings for Opus models. The lab reached an implied private-market valuation above $1 trillion earlier this year.

The agreement contains a governance clause that lets SpaceX reclaim compute if Anthropic’s models are judged to pose a threat to humanity. Elon Musk wrote that SpaceX will provide compute to AI companies taking steps to “ensure it is good for humanity” and reserves the right to “reclaim the compute” if their AI engages in harmful actions.

SpaceX merged with xAI in February. The Anthropic deal demonstrates the Colossus cluster can serve paying external customers in addition to Grok, SpaceX’s in-house model. The two companies also flagged interest in multi-gigawatt orbital compute that would pair Starship hardware with future AI training workloads, an idea described as a longer-term prospect.

Bankers view the Colossus sale as material for the IPO roadshow and as evidence of demand for large-scale AI infrastructure. Some bankers have discussed lifting valuation guidance toward $2 trillion at the top end of the book-building range.

Investors will watch the public S-1 filing for any disclosure of AI compute revenue tied to Musk’s operations. Details on recurring AI infrastructure revenue could influence valuation discussions as SpaceX moves toward a Nasdaq listing in June.