SpaceX is setting an unusually large target for AI compute. According to Elon Musk, the company wants to expand capacity more than fivefold by the end of 2027 and finish that year above two gigawatts of power.
The plan matters because the buildout is tied directly to Nvidia hardware. SpaceX currently sits at 1.4 gigawatts, according to its quarterly report, and Musk said the end-of-2027 figure should be closer to ten gigawatts than five. That ambition turns the question from whether SpaceX will buy more accelerators into how many Nvidia GPUs such a buildout could require.
The compute target is measured in gigawatts
The central figure in the plan is power capacity. SpaceX wants to grow from 1.4 gigawatts today to a much larger footprint by the end of 2027. Musk described the goal as above two gigawatts and said capacity should be closer to ten gigawatts than five.
That framing shows how AI infrastructure is increasingly discussed at the scale of energy systems, not just server counts. A large AI cluster is not defined only by the number of chips inside it. It also depends on power, cooling, networking, and the ability to keep those systems running at useful utilization.
SpaceX plans to base the new buildout on Nvidia's upcoming Vera Rubin platform. Musk called Vera Rubin the best architecture available. The company's existing Colossus clusters already run on Nvidia's H100, GB200, and GB300 systems, so the plan continues a hardware path that is already centered on Nvidia.
Why the GPU count could move beyond one million
The existing and planned Colossus footprint is already large. SpaceX's IPO filings put the current and planned buildout at roughly 100,000 H100, 110,000 GB200, and 110,000 GB300 processors, plus at least 220,000 more planned GB300 units.
Together, that equals about 540,000 GPUs total. The new capacity target would push the math well past one million Rubin GPUs. If the full expansion used Vera Rubin, the number could exceed two million.
That final point is important: SpaceX has not said how much of the buildout will actually be Rubin. The two million figure is therefore a scenario based on the expansion being entirely Vera Rubin, not a confirmed purchase number.
Still, the scale implied by the target is clear. Even before the new expansion, Colossus includes multiple generations of Nvidia systems. Adding enough capacity to move from the current level toward the end-of-2027 goal would require a step change in infrastructure.
SpaceX's AI business is growing, but losses remain high
The compute push sits inside a broader business shift. Musk's AI company xAI merged into SpaceX in February 2026 through a deal funded mostly with stock. The transaction valued SpaceX at $1 trillion and xAI at $250 billion, for a combined $1.25 trillion.
SpaceX's AI segment includes xAI and X. In Q2 2026, that segment posted $2.56 billion in revenue and an operating loss of $1.26 billion. In Q1, it had recorded an operating loss of $2.47 billion on $818 million in revenue.
For the first half of 2026, operating losses in the AI segment totaled $3.73 billion. The numbers show a business that is expanding revenue quickly while still carrying steep losses.
Most of the Q2 revenue jump did not come from Grok. It came from new cloud contracts that lease out Colossus compute capacity. A single unnamed AI customer accounted for roughly $1.52 billion of that revenue, and that customer is likely Anthropic.
The Colossus strategy differs from rivals
The source article contrasts SpaceX's Nvidia-centered plan with competitors that spread compute across more providers. Anthropic uses AWS Trainium, Nvidia GPUs, and Google's TPUs to train and run Claude. It has publicly announced up to five gigawatts of additional AWS infrastructure with Trainium and Graviton systems, plus about 3.5 gigawatts of extra TPU capacity.
OpenAI has also announced compute across multiple suppliers: six gigawatts of AMD capacity, at least ten gigawatts of Nvidia capacity, and around two gigawatts of AWS Trainium capacity.
SpaceX's stated plan is different because the entire buildout will run on Nvidia chips. That makes the choice of Vera Rubin especially consequential. If the company does lean heavily on that platform, its future AI capacity would be tied closely to Nvidia's next architecture.
The logic behind leasing Colossus capacity is also visible in the revenue numbers. If customers are willing to pay for access to large clusters, then compute infrastructure becomes both a strategic resource for internal AI work and a commercial product in its own right.
Orbit remains part of the long-term vision
Musk's longer-term ambition goes beyond ground-based data centers. He wants to run data centers in orbit. The source does not provide details on timing, design, or deployment, but the statement places the current buildout inside a much larger infrastructure vision.
For now, the concrete story is the near-term expansion: SpaceX has 1.4 gigawatts of current capacity, wants to grow more than fivefold by the end of 2027, and plans to do it with Nvidia chips. If Vera Rubin carries the full expansion, the GPU count could exceed two million.
That is not just a procurement story. It is a signal about how AI companies are defining advantage: through access to power, advanced accelerators, and clusters large enough to train and serve frontier systems at scale.