SpaceX AI supercomputer plans, 2 million Nvidia GPUs, data center, server racks, advanced technology.

Editorial illustration for SpaceX's AI supercomputer plans could require over 2 million Nvidia GPUs

SpaceX's AI supercomputer plans could require over 2...

By AI Daily Post Edited by Brian Petersen, Editor-in-Chief

August 5, 2026 • 3 min read

Elon Musk laid out a compute target on SpaceX's Q2 earnings call that dwarfs anything the company has built so far. By the end of 2027, SpaceX wants its data center capacity to sit above two gigawatts, more than five times its current 1.4 gigawatts, with Musk suggesting the real number could land closer to ten gigawatts. Every bit of that expansion runs through Nvidia silicon.

The current buildout, known as Colossus, already spans roughly 540,000 GPUs across H100, GB200, and GB300 chips, according to SpaceX's IPO filings. Musk wants the next phase built on Nvidia's unreleased Vera Rubin platform, which he's called the best architecture on the market. Scaling to the gigawatt targets he described would push GPU counts past one million, and if Rubin chips make up the whole expansion, the total could clear two million units. SpaceX hasn't specified what share of the buildout will actually use Rubin hardware.

The numbers come months after xAI merged into SpaceX in a stock-heavy deal that valued the combined company at $1.25 trillion. Musk's stated ambitions don't stop on Earth, either.

SpaceX's IPO filings put the current and planned Colossus buildout at roughly 100,000 H100, 110,000 GB200, and 110,000 GB300 processors, plus at least 220,000 more planned GB300 units. That's about 540,000 GPUs total. The new capacity target would push the math well past one million Rubin GPUs.

SpaceX’s ambitious compute goals could require over two million Nvidia Rubin GPUs - THE DECODER

Why this matters

Musk's numbers keep moving, and that's the tell. Five gigawatts, then ten by 2027, run entirely on Nvidia silicon that doesn't fully exist yet in the quantities SpaceX would need. Two million Rubin GPUs isn't a procurement plan, it's a bet that Nvidia's roadmap holds and that Musk's timelines, historically generous with themselves, land closer to reality this time.

For developers and founders watching compute costs, this matters less as a SpaceX story and more as a demand signal: xAI, Tesla, and now SpaceX are all drawing from the same Nvidia supply chain simultaneously. That's the kind of concentrated demand that pushes lead times and prices for everyone else trying to book GPU capacity. Researchers should note the shift from H100 to GB200/GB300 to a hypothetical Rubin buildout inside a few years, that's the actual pace of the frontier hardware cycle now, not the leisurely multi-year refreshes we used to plan around.

Whether SpaceX gets anywhere near ten gigawatts is a fair question. Whether that appetite reshapes who can get chips at all in the meantime is the one worth tracking now.