SpaceX Is Casting Its Own Gas Turbine Blades to Buy Back 18 Months
Four foundries control the castings that gate every new gas turbine, and SpaceX decided becoming the fifth beats waiting in GE Vernova's 116 GW line.
Four foundries control the castings that gate every new gas turbine. SpaceX decided becoming the fifth beats waiting in GE Vernova's 116 GW line.
SpaceX is building a foundry in Bastrop, Texas to cast its own gas turbine blades and vanes. Elon Musk confirmed the plan on X on August 29 and called it "a profound game-changer", saying in-house castings pull gas turbine deployments forward by as much as 18 months. The company has reason to hurry. xAI, now folded into SpaceX, runs its Colossus supercomputer in Memphis on a fleet of gas turbines, and every turbine it wants next is sitting in someone else's order book.
Why turbine blades are the choke point
A modern turbine blade spins in gas hotter than the melting point of the metal it's made from. It survives because of how it's cast. Each blade gets grown as one unbroken crystal in a vacuum furnace, with cooling channels threaded through the inside (the growing takes days per batch, which is why nobody can just add a shift). Four companies do this at industrial scale: Howmet Aerospace, Berkshire Hathaway's Precision Castparts, Consolidated Precision Products, and the UK's Doncasters. Together they hold roughly 70 to 80 percent of the market for these castings, and their order books run through 2029. A new merchant foundry takes four to six years to build, and the parts then need qualification from the turbine maker before a single blade ships.
SpaceX's answer is the same one it gave rocket engines and Starlink terminals. Build the part yourself. Between March and June the company bought roughly 830 acres near its Bastrop plant, and satellite imagery reviewed by The Information showed two new buildings standing by July. Job postings in April described a "Blades and Vanes Foundry" making nickel-superalloy castings for power generation. Morgan Stanley's August 16 note argued the foundry serves double duty, feeding data-center turbines and the turbopumps inside Starship's Raptor engines. (One detail from the reporting: no air or construction permits have surfaced in Texas records yet. The buildings came first.)
GE Vernova's gas turbine queue explains the whole move
The clean read on why a rocket company is pouring superalloy sits in GE Vernova's second quarter. Gas equipment backlog plus paid slot reservations grew from 100 GW to 116 GW in three months. About 53 GW of that is firm orders. The other 63 GW is customers paying deposits just to hold a spot in line. $GEV is now taking reservations for 2031 deliveries and told investors to expect at least 125 GW in the queue by year-end.
Prices follow the queue. Wood Mackenzie expects gas turbine prices to reach $600 per kilowatt by the end of 2027, nearly triple the 2019 level. Global turbine orders stood at 110 GW at the end of last year. Worldwide manufacturing capacity sits between 60 and 70 GW. $GEV is spending more than $160 million to push large-frame output from about 50 units a year to 70 or 80 by late 2026, with Siemens Energy and Mitsubishi expanding behind it. None of that closes the gap before 2028. So a paid reservation for a 2031 delivery slot is a product $GEV sells today, and buyers keep lining up for it.
What vertical integration in power equipment means for the buildout
SpaceX gets to skip the slowest part. A merchant foundry spends years getting each casting qualified by the turbine OEM it supplies. SpaceX is the turbine buyer, the data-center operator, and soon the foundry, so it signs off on its own blades (nobody has to approve the parts you cast for your own machine). Raptor already put single-crystal metallurgy on the Starbase bench, so the skill transfer is shorter than it'd be for anyone else contemplating this.
For the rest of the buildout, the queue doesn't shrink because SpaceX steps out of it. $GEV keeps booking equipment backlog and deposit-backed reservations, and that revenue converts as units ship over the next five years. Howmet keeps selling every airfoil it can cast, with $HWM order books already stretched to 2029. The warning in the Bastrop foundry is aimed at the years after that. Prices that nearly triple turn customers into competitors, and SpaceX is just the first customer with the metallurgists on staff to act on it. Anyone underwriting $GEV past the visible backlog is really underwriting how long the casting oligopoly stays a four-company club.
SpaceX built a foundry to get its 18 months back. Everyone else is still paying deposits for 2031.