SpaceX borrowed $20 billion in debt financing weeks after closing its IPO, directing the capital not toward Starship development but toward expanding Colossus, its Memphis data center complex now generating revenue as a commercial compute platform. The company signed a $6.3 billion deal with open-source AI startup Reflection and earlier contracts with Anthropic, Google, and Cursor. AI infrastructure revenue reached $3.2 billion in 2025. One investment bank projects $322 billion by 2030.
The timing tells the story. SpaceX raised equity, immediately levered the balance sheet, and began booking multi-year compute contracts before the IPO roadshow materials went stale. The Colossus facility was built to train xAI models. It is now a for-hire GPU cluster with external customer commitments that dwarf the original internal use case. The Reflection deal alone represents twice SpaceX's total AI revenue last year. The debt structure suggests management expects compute margin to carry interest expense and fund further buildout without touching launch cash flows.
This is not a satellite company hedging into cloud. This is a deliberate reallocation of enterprise value from launch services to AI infrastructure, executed while public market appetite for compute capacity sits near all-time highs. SpaceX has 12,000 operational Starlink satellites and 5,000 more launching this year. The scale required to manage that constellation gave the company experience running distributed compute at a size most hyperscalers reach only after a decade. Colossus is that capability productized. The customer list is narrow but categorical: frontier model labs that cannot wait for Azure or GCP capacity, and code-generation tools where latency and API rate limits determine product viability.
The risk is twofold. First, the $20 billion borrowed against an asset class with no historical default curve. Data centers age faster than launch pads, and AI compute has no secondary market if model architecture shifts away from dense GPU clusters. Second, SpaceX is now a counterparty to the same companies that might, in 18 months, build their own inference infrastructure or move to edge deployments. Anthropic and Google are both designing custom silicon. If they migrate workloads off Colossus before 2028, the debt service remains but the revenue does not. The only comparable precedent is Equinix borrowing into the fiber buildout in 1999, which worked until bandwidth became abundant.
Allocators should track three items in the next six months. First, whether SpaceX closes additional compute deals above $2 billion in contract value, which would confirm Colossus as a category leader rather than a one-time arbitrage. Second, any guidance on depreciation schedules and refresh CapEx for the Memphis facility, which will clarify true margin after hardware replacement. Third, Starlink's free cash flow trajectory, since the debt covenant structure likely requires launch and connectivity revenue to remain investment-grade even if AI income stalls. If Starlink subscriptions decelerate below 3 million net adds annually, the leverage begins to matter.
The $322 billion revenue projection by 2030 assumes a 100x increase in five years, which implies either monopoly pricing or market size expansion that has not yet occurred. Neither is impossible. Both require flawless execution in a sector where the median data center operator trades at 8x EBITDA and SpaceX just mortgaged itself at launch-company cost of capital.