SpaceX wants to grow its AI compute capacity more than fivefold by the end of 2027, a move that could require over two million Nvidia Rubin GPUs. Elon Musk, CEO of SpaceX and xAI, disclosed the target during the company's Q2 2026 earnings call. The plan pushes the aerospace and satellite firm deeper into the AI infrastructure race, just months after it absorbed xAI.
A Gigawatt-Scale Ambition
Musk said capacity should sit above two gigawatts of power by the end of 2027. He elaborated that it should be "closer to ten gigawatts than five" by that date. Current capacity stands at 1.4 gigawatts, according to SpaceX's quarterly report. That means the company is aiming for a jump of more than 25 percent in the near term, with a much larger leap planned over the next two years.
The expansion will build on Nvidia's upcoming Vera Rubin platform. Musk called it the best architecture available. Existing Colossus clusters run on Nvidia H100, GB200, and GB300 systems. IPO filings list the current and planned buildout at roughly 100,000 H100 GPUs, 110,000 GB200 units, 110,000 GB300 systems, plus at least 220,000 more planned GB300 units. That totals about 540,000 GPUs in the current and planned pipeline.
The Rubin Math
The new capacity target pushes the math past one million Rubin GPUs. If the entire expansion used Vera Rubin, the number could exceed two million. SpaceX hasn't specified how much of the buildout will actually be Rubin, so the exact mix remains unclear. The company's filings and public statements leave room for other Nvidia platforms or additional hardware suppliers.
The scale is staggering. Two million GPUs would dwarf most existing data center fleets. SpaceX is positioning itself as a major compute provider, not just a rocket and satellite company. The merger with xAI in February 2026 gave it the AI talent and models to make that pivot credible.
A $1.25 Trillion Combination
The merger deal was funded mostly with stock. It valued SpaceX at $1 trillion and xAI at $250 billion, for a combined $1.25 trillion. That valuation reflects the market's appetite for AI compute infrastructure. Musk has also said he wants to run data centers in orbit long-term, though no timeline has been given for that ambition.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The financial picture is mixed. SpaceX's AI segment, which includes xAI and X, posted Q2 2026 revenue of $2.56 billion. That is a sharp jump from Q1 2026 revenue of $818 million. Operating losses tell a different story. The segment lost $1.26 billion in Q2 2026, after a $2.47 billion loss in Q1. First half 2026 operating losses for the AI segment total $3.73 billion.
One Big Customer
The revenue jump mostly comes from new cloud contracts leasing Colossus compute capacity, not from Grok, the AI model developed by xAI. A single unnamed AI customer accounted for roughly $1.52 billion of Q2 revenue. That customer is likely Anthropic, the developer of Claude. The scale of the payments matches Anthropic's aggressive compute procurement strategy.
Anthropic spreads its workload across multiple providers. It uses AWS Trainium, Nvidia GPUs, and Google TPUs for Claude. The company has announced up to five gigawatts of additional AWS infrastructure with Trainium and Graviton systems. It also has about 3.5 gigawatts of extra TPU capacity from Google. That diversification contrasts with SpaceX's heavy reliance on Nvidia hardware.
The Competitive Landscape
OpenAI takes a similar multi-vendor approach. It has announced six gigawatts of AMD capacity, at least ten gigawatts of Nvidia capacity, and around two gigawatts of AWS Trainium capacity. That spread reduces dependence on any single chipmaker. SpaceX, by contrast, is betting big on Nvidia's roadmap, particularly Vera Rubin.
The race is not just about GPUs. Power is the limiting factor. Two gigawatts is enough to power a mid-sized city. Ten gigawatts would be a massive draw on any grid. Musk's orbital data center idea could sidestep some of those constraints, but that remains a long-term vision. For now, the company is building on the ground, one cluster at a time.
The numbers are still early. SpaceX hasn't confirmed how many Rubin GPUs it will actually deploy. The two million figure is an upper estimate, not a commitment. But the direction is clear. The company is aiming for a scale that would make it one of the largest AI compute providers on the planet, with all the revenue and risk that entails.