SpaceX Targets 10 GW of Data Center Capacity in 2027

Elon Musk says SpaceX could add 10 GW of data center capacity in 2027 alone — roughly matching the power capacity Google has built across its entire history, in about 18 months. TechRadar reported the claim on November 10. The analysts who checked the math, SemiAnalysis (an independent research firm focused on semiconductors and AI infrastructure), judged it feasible. Conditionally.

Musk said he was "highly confident" SpaceX could deliver the 10 GW in 2027. That's a founder forecast from the executive who controls the build. SemiAnalysis then tested it against what it could see on the ground, and its verdict hinged on five inputs: construction, power availability, permitting, GPU supply and equipment deployment. Any one of them slipping can cap the whole program. A later SemiAnalysis estimate pushed the potential 2027 buildout to about 13 GW, 30% above Musk's figure.

What's confirmed versus projected?

SemiAnalysis published three distinct numbers, and mixing them up is the fastest way to misread this story:

• About 3.6 GW under construction now. The firmest figure, pointing at physical work in progress — though under construction does not mean energized or filled with working GPUs. That's 36% of the 10 GW target, leaving 6.4 GW to start and finish inside the same window.
• More than 9 GW of potential capacity mapped across sites including Colossus (the AI training campus tied to Musk's AI operations) and Starbase (the South Texas rocket development site). Mapped means plausible, not scheduled.
• About 13 GW of potential 2027 buildout, a later projection and not a confirmed figure.

None of these describes capacity that exists today. The Google comparison — matching a lifetime of building in 18 months — is an analyst scale marker, not a completed result. Claims that SpaceX's capacity exceeds Google's "entire lifetime capacity" remain unverified by anyone.

Why gigawatts aren't compute

Two caveats sit in the first line of every honest read. Gigawatts measure how much electricity a facility can draw, not how many AI operations it performs. Two campuses with the same power rating can deliver very different useful work depending on chips, networking, cooling efficiency and software. The comparison concerns power capacity only.

The distinction cuts both ways. Capacity in watts can understate performance when newer chips do more work per watt. It can overstate output when power sits idle waiting for silicon. Google, for what it's worth, designs its own accelerators and keeps shipping new generations. No source here supports a claim that SpaceX capacity would equal Google's in raw AI performance.

What could stop it?

SemiAnalysis tied its feasibility call to five gates, each with its own lead times. Grid interconnection queues in the US stretch for years in many regions. Transformers, switchgear and cooling systems all carry long lead times. Industry-wide competition for advanced accelerators and high-bandwidth memory remains tight for every buyer. Local governments now vote on data center projects with real consequences, and public opposition has grown. The reports do not name a GPU count, a construction cost, a completion date or a facility-by-facility breakdown — any such figure circulating elsewhere is unconfirmed.

How does SpaceX compare with Google?

Google built its footprint in steady stages over its entire operating history — sites, power contracts, its own silicon. SpaceX is attempting a compressed version from a much younger infrastructure base. One side of the comparison is history; the other is a plan with an analyst endorsement. That gap is where most large infrastructure projects stumble.

The site structure helps explain the pace. SpaceX isn't starting from empty land: it's building on campuses where it already controls property and can reach power infrastructure. Concentration brings advantages — simpler logistics, shorter paths from power plant to server hall — and risks. One stalled permit can hold up gigawatts at once.

Market impact

If a large share materializes, chip vendors, turbine suppliers and electrical contractors see order books lengthen, while utilities face a major new load arriving faster than planning cycles allow. Rival AI labs would compete with SpaceX for GPUs, memory and grid connections — and face a capacity-rich competitor if the sites get fully equipped. SemiAnalysis addressed feasibility of the build, not profitability of the result.

What does this change for operators and buyers? For now, nothing measured. Watch utility filings and interconnection agreements tied to Colossus and Starbase, local permit decisions, and supplier disclosures on large turbine and transformer orders. A company that's delivering reports energized capacity, not planned capacity. Until then, 10 GW stays a projection.