SpaceX Turns AI Compute Into a Rentable Infrastructure Business

The Colossus data center in Memphis was originally an internal facility built to train Grok, but it is now beginning to operate like a cloud provider and rent capacity to external customers. In its public filings and supplemental materials, SpaceX disclosed that Anthropic and Google have together reserved large amounts of NVIDIA GPU compute, with contract value reaching about $2.17 billion per month during the overlap period.\n\n## What happened\n\nIn a free writing prospectus filed on June 5, 2026, SpaceX said it had signed a cloud services agreement with Google to provide the latter with about 110,000 NVIDIA GPUs plus supporting CPU, memory, and other resources. Under the agreement, Google will pay $920 million per month from October 2026 through June 2029; if SpaceX fails to deliver the promised number of GPUs by September 30, 2026, Google may terminate the agreement or reduce payments proportionally based on actual deliveries.(sec.gov)\n\nAnother SEC filing from SpaceX shows that an agreement signed by Anthropic in May 2026 covers about 325,000 NVIDIA GPUs, as well as servers, storage, and high-speed networking infrastructure. Anthropic agreed to pay $1.25 billion per month through May 2029, with a lower rate applying during the initial expansion phase; both parties have a 90-day termination right after the initial phase.(sec.gov) Anthropic later publicly confirmed that one direct result of the partnership was higher usage limits for Claude Code and the Claude API.(anthropic.com)\n\n## Why it matters\n\nThis is not a typical cloud computing order, but rather a sign that compute assets are beginning to be priced as long-term capacity contracts. If the overlapping contract months are simply added together, SpaceX’s monthly contract revenue would be about $2.17 billion, or an annualized $26 billion; but this is only an arithmetic projection of contract value, not realized revenue, and it does not deduct construction, chip, electricity, operations, or financing costs.\n\nThe more important change is at the mechanism level. In the past, the scarcest resources at AI companies were usually described as model capability and user scale; now, entities that can secure power, build high-density data centers, obtain GPUs, and deliver them on schedule can package compute as rentable, financeable, and valuably priceable infrastructure. After SpaceX folded xAI into its own structure, Colossus is no longer just a cost center serving Grok, but also has a path to monetization through external customers.\n\nThis expands the competitive boundary between “new cloud providers” and traditional cloud platforms. Google itself has a cloud business and proprietary AI chips, yet it still chose to rent NVIDIA compute from SpaceX under contract, showing that when model demand is growing rapidly, near-term deliverable capacity may be more valuable than a long-term technology roadmap. For public markets, investors need to distinguish among three types of revenue: model subscription revenue, cloud platform service revenue, and capacity rental revenue backed by power and hardware. Their gross margins, capital expenditures, and contract risks are not the same.\n\n## What to watch next\n\nFirst, whether SpaceX can deliver the GPUs on schedule. The Google agreement already includes termination and payment reduction provisions for shortfalls, meaning contract size does not equal unconditional revenue. Second, how Anthropic’s and Google’s 90-day termination rights will affect lenders’ assessment of cash flow stability. Third, whether the customer base for Colossus will continue to expand: if more model companies choose to rent SpaceX’s compute, SpaceX could gradually shift from being a space company with AI assets to an infrastructure platform with significant customer concentration risk.\n\nWhat is clear for now is that AI compute is moving from internal corporate spending to commercial contracts with terms around duration, capacity, delivery, and default. The next phase of competition may not be just about “who has the strongest model,” but about who can combine power, chips, and data centers into more reliable cash flow.\n\n## Sources\n\n- SpaceX’s SEC filing on the Google compute agreement\n- SpaceX’s S-1 filing with the SEC, disclosing the Anthropic compute agreement\n- Anthropic’s announcement on its partnership with SpaceX and Claude usage limits\n\n## Sources\n\n- SpaceX SEC listing documents and Anthropic official announcement