SpaceX Cuts Data Center Build Times to 4-5 Months, Analysts Claim

SpaceX completed its MiniHard data center in roughly five months, and its Gun Range plot south of the state line is on track for about four, according to SemiAnalysis — figures that, if accurate, halve the 8-12 month build times most operators report. The claim comes from a single analyst outlet. SpaceX has published no construction records to confirm it.

What do the build numbers actually show?

The SemiAnalysis thread, posted November 14, lays out a pace few rivals match:

• MiniHard: construction finished in approximately five months.
• Gun Range plot: expected build of roughly four months.
• Colossus 1 first retrofit: 122 days for 150 MW.
• Colossus 2: 91 days for 210 MW.
• Project SPAM: started October 5, targeting first tranches in December.

The retrofit numbers are the hardest data in the set. The headline build times are approximations, marked with tildes, and unaudited. Treat the four-month Gun Range figure as a forecast, not a result.

Where does the speed come from?

SemiAnalysis credits two factors. First, staffing density: the Elon Musk companies run at roughly 3,000 workers per gigawatt, which the analysts attribute to top-tier engineering headcount. Second, and more speculative, they believe Tesla is manufacturing modular mechanical-electrical-plumbing equipment for SpaceX deployment. No contract, shipping manifest or Tesla statement supports that belief — it is inference, and SemiAnalysis flags it as such with the word "believe."

The analysts compare the arrangement to Crusoe building its own switchgear to skip lead times, but at much larger scale. They predict Tesla will move beyond its current role as a battery-energy-storage vendor to SpaceX. That is a projection, not an announcement.

How much storage is on site?

The storage numbers are concrete. Colossus 2 hosts 1,377 Megapacks, nearly 6 GWh — which SemiAnalysis calls the largest battery storage field in the United States. The deployment pattern the analysts describe is roughly 1,000 Megapacks, or 250 Megablocks, per gigawatt.

At 20 GW of AI data center capacity, SpaceX would need about 20,000 Megapacks per year. That is a demand figure the analysts themselves generated; Tesla has not committed to supplying it.

The proposed Arizona data center would carry four times the Megapack count of the current site.

Can the power math hold?

SemiAnalysis says gas turbines and batteries are orchestrated together: turbines cycle through charging the Megapacks while both sources raise effective baseload. With a PUE of roughly 1.10-1.15, the site may still fall about 300 MW short of demand. That shortfall caveat matters. Fast construction does not solve an energy deficit, and the analysts bury the gap under the speed narrative.

The same outlet projects SpaceX will operate the most AI data centers by the end of 2027. That is a two-year forecast resting entirely on continued execution at the observed pace — and on Tesla's supply chain holding.

What is not said?

Absent from the reporting: capital costs per megawatt, confirmed vendor agreements, grid interconnection status for Arizona, and any failure or delay data. Every figure traces back to one analyst account's thread. The build-speed story is plausible given the retrofit record. The 20 GW ambition and the Tesla manufacturing thesis remain claims awaiting paperwork.