Musk projects at least 13 gigawatts of AI compute by end of next year

Elon Musk says his AI operation will be running on at least 13 gigawatts of compute by the end of next year. For scale, the current figure is 1.7 GW.

Musk laid out the underlying numbers in more detail on SpaceX’s Q3 2026 earnings call on October 20, and those numbers describe SpaceX’s own AI buildout. The goal is a roughly sevenfold jump in about 15 months.

The math behind 13 gigawatts

On the call, Musk pointed to a potential increase of approximately 10 GW in 2027. That would lift total compute capacity to around 13 GW by year-end.

SpaceX’s AI compute capacity stood at 1.7 GW as of September 30, 2026, up from 1.4 GW in Q2. The near-term target is more than 2 GW by the end of 2026. Hitting 13 GW a year later means the buildout has to accelerate dramatically.

From Q2 to Q3, capacity grew by about 0.3 GW. The 2027 plan calls for roughly 10 GW in a single year, which works out to about 2.5 GW per quarter if spread evenly.

Nvidia chips, by the hundreds of thousands

Musk said SpaceX is installing about 200,000 Nvidia GB300 GPUs every month for the rest of 2026. After that, the deployment is expected to shift toward Nvidia’s upcoming Vera Rubin architecture.

The headline project is a large-scale training cluster. Musk described a system of 440,000 GPUs that is expected to be operational by mid-2027.

SpaceX’s longer-term plans call for power and cooling infrastructure reaching 15 to 20 GW by the end of 2027. That range sits above the 13 GW compute target.

The revenue engine paying for it

SpaceX’s Super Intelligence segment, which includes xAI cloud services and other offerings, generated $7.56 billion in revenue in Q3 2026. That was up 830% year-over-year, making it SpaceX’s largest business area.

What this means

For Nvidia, a customer installing about 200,000 GB300 GPUs a month, then moving to Vera Rubin, represents a large and sustained order book. For companies that supply power, cooling and data center construction, the 15 to 20 GW power and cooling target may matter even more than the chip count.

The 440,000-GPU cluster is the milestone to watch first. It is expected to come online by mid-2027, so its status will show whether the broader buildout is on schedule well before year-end.

The year-end 2026 figure is the other near-term checkpoint. Clearing more than 2 GW would show the current deployment pace is holding, while falling short would raise questions about the far steeper climb that follows.