SpaceX Claims 18-Month AI Build to Match Google's Lifetime

The 10GW claim, explained

SpaceX's target is to go from roughly 3 gigawatts of compute infrastructure to more than 10GW, according to SemiAnalysis — a jump of about 10GW in 18 months.

For context, 10 gigawatts is roughly the output of ten large nuclear reactors. Google, which has been building data centers since the late 1990s, has accumulated a comparable total over its entire existence. SpaceX says it can match that in a year and a half.

The claim isn't coming from a press release. SemiAnalysis, a research firm that tracks AI infrastructure closely, examined SpaceX's construction pipeline across several sites and concluded the schedule is achievable. That's a notable vote of confidence for a company whose CEO is known for setting deadlines that slip.

The evidence is Colossus 2, SpaceX's flagship AI facility. Its power generation capacity reportedly climbed from 0.4GW in January to a projected 2.4GW by December. That's a sixfold increase in under a year — the kind of curve that makes 10GW sound less like fantasy.

SpaceX's AI data center plans: key numbers

What's actually powering Colossus 2

Colossus 2 runs on a mix of on-site gas turbines, Tesla Megapack batteries and grid power — a combination that reveals both the ambition and the constraints.

The facility reportedly houses 89 gas turbines capable of generating about 2.5GW, plus 1,377 Tesla Megapack battery systems. Those batteries aren't just backup; they help smooth demand fluctuations and fill gaps when turbines or grid power fall short.

Environmental permits will eventually cap the turbines at roughly 1.24GW, according to the report. The site is also expected to draw about 1.2GW from the grid, leaving an estimated 300MW shortfall that batteries and external sources may cover.

SpaceX has temporary permission from Mississippi authorities to run additional turbines, provided it phases them out by July 2027. That deadline adds a layer of regulatory pressure to an already tight construction schedule.

Tesla's manufacturing edge — and the GPU bottleneck

SemiAnalysis suggests Tesla may be producing modular mechanical and electrical equipment, letting SpaceX install finished systems instead of waiting on conventional suppliers.

That vertical integration could be a real advantage. Power equipment like transformers and switchgear is in short supply across the industry, and lead times can stretch years. If Tesla can churn out modular components, SpaceX skips the queue.

But buildings aren't compute. Microsoft CEO Satya Nadella has described the problem bluntly: “You may actually have a bunch of chips sitting in inventory that I can't plug in.” SpaceX still needs GPUs, memory and cooling systems — and those supply chains are stretched thin across the entire AI industry.

Musk said Colossus 2 already operated 550,000 GPUs as of September 25, though the distribution across buildings is unclear. Other projects, including Gun Range and Project SPAM, are also under construction, with aerial photos showing significant activity. Whether they translate into operational capacity by 2027 is the open question.

What it means for you

For US, UK and EU readers, the immediate impact is indirect: more AI compute capacity could mean faster, cheaper AI services down the line. But the buildout also strains local power grids and raises questions about emissions from gas turbines, especially in regions with strict environmental rules.

SpaceX's Mississippi facility operates under temporary permits that expire in July 2027. Similar projects in Europe would face far tougher permitting hurdles, so the 10GW target is likely to be concentrated in the US, where rules are looser and land is cheaper.

SpaceX's 10GW claim is the kind of number that sounds absurd until you look at the construction photos. The company has already shown it can scale power capacity at Colossus 2 faster than almost anyone expected. The question now is whether chips, permits and grid connections can keep up with the bulldozers.