SpaceX has set a new ceiling for AI infrastructure. During its inaugural earnings call as a public company on August 4, the firm announced an exclusive commitment to build its future compute capacity on Nvidia’s Vera Rubin architecture. The scale is unprecedented: 2 gigawatts (GW) of installed compute by the end of 2026, scaling to approximately 10 GW by the end of 2027, with a tentative 20 GW power and cooling infrastructure target – Musk himself acknowledged ~15 GW as the realistic outcome. This represents the largest single compute commitment in the history of artificial intelligence, effectively turning the aerospace giant into the world’s most aggressive compute landlord.
The financial architecture supporting this move is as aggressive as the hardware deployment. SpaceX reported Q2 2026 revenue of $7.8 billion, a 92% year-over-year increase that comfortably beat analyst expectations of $6.93 billion, according to CNBC. While the company’s AI division posted an operating loss of $1.26 billion, the core business model is already clear: 95% of its AI revenue is derived from GPU rental. Major tenants include Google, paying $920 million per month for roughly 110,000 GPUs, and Anthropic, which contributes approximately $1.25 billion per month at the Colossus 1 facility in Memphis. As CFO Bret Johnsen noted, the company is successfully monetizing available compute capacity to generate high incremental EBITDA margins.
The shift to the Vera Rubin architecture is a strategic bet on density and performance. Elon Musk, who characterized Vera Rubin as “the best AI computer” available, is banking on this hardware to maintain a competitive edge. According to Melius Research analyst Ben Reitzes, even the initial 2 GW deployment could add roughly $100 billion in incremental revenue for Nvidia, pushing the chipmaker toward a $1 trillion valuation milestone. This hardware pipeline is not merely speculative; SpaceX has already secured $6.7 billion in forward cloud services contracts starting in October 2026.
Beyond terrestrial data centers, SpaceX is pushing compute into orbit. The upcoming Starmind AI1 satellite, equipped with the Nvidia Space-1 Vera Rubin module, promises 25 times more AI compute per GPU compared to the H100. With an FCC filing for up to 1 million satellites in sun-synchronous low Earth orbit, SpaceX is attempting to solve the latency and memory bottlenecks that plague ground-based clusters. This orbital differentiation is central to the company’s long-term strategy to compete with its own tenants, including its internal Grok 4.5 development.
The compute landlord thesis, previously explored in the context of Nvidia’s $600 billion exposure to OpenAI, has now reached a structural extreme. By acquiring xAI in February 2026 and subsequently going public in June with an $85 billion raise and a $1.75 trillion valuation, SpaceX has effectively institutionalized the landlord model. The company is no longer just building rockets; it is building the physical substrate for the next generation of AI.
Musk’s assertion that $100 billion in annual recurring revenue by December is a baseline expectation – achievable even if the company does nothing – underscores the confidence behind this capital expenditure. With H1 2026 capex reaching $28 billion, up from $7 billion in the same period last year, the company is burning cash to secure its position as the primary host for the world’s most powerful models. The IPO was not merely a liquidity event; it was a necessary mechanism to fund a multi-hundred-billion-dollar hardware pipeline. The compute landlord thesis has officially scaled from a $10 billion experiment to a $100 billion-plus industrial reality.