Elon Musk is trying to manufacture his way around one of the AI boom’s biggest constraints: finding enough electricity to switch on all those expensive Nvidia Corp. (NASDAQ:NVDA) chips.
SpaceX (NASDAQ:SPCX) is building a foundry in Bastrop, Texas, to cast blades and vanes for industrial gas turbines, according to The Information, which cited SpaceX job postings describing the construction and ramp-up of the facility.
Musk later confirmed on X that SpaceX is bringing the casting process in-house, saying it could accelerate natural-gas turbines coming online by up to 18 months.
The striking part is that Musk is targeting the same part of the supply chain that GE Vernova Inc. (NYSE:GEV) has been warning investors about.
GE Vernova Says the Bottleneck Is Real
GE Vernova CEO Scott Strazik has said the company has had to push suppliers to add capacity just to keep up with turbine demand.
At a Bernstein conference in May, Strazik said GE went to its castings and forgings suppliers in the summer of 2024 and told them it needed substantially more output. In some cases, GE even provided the capital for suppliers to build new furnaces.
The pressure has only increased. GE is mostly sold out through 2030 and is trying to lift annual turbine production from about 20 gigawatts this year to 30 GW by the end of the decade.
That ramp still depends on the part of the supply chain Musk is now trying to bring in-house. On GE’s July earnings call, Strazik said the step up to 24 GW of annual output in 2028 depends on additional castings and forgings arriving in 2027.
Musk said in February that SpaceX and Tesla might have to manufacture blades and vanes themselves because casting capacity was so constrained.
The AI Race Is Becoming a Power Race
SpaceX has committed more than $2.8 billion to gas turbines over three years to support its Colossus data centers near Memphis. Building blades and vanes in-house would push that effort further upstream, into the supply chain limiting how quickly new turbines can be delivered.
Data Center Map currently lists 4,767 U.S. data centers. Kalshi traders put a 66% chance on that figure reaching at least 5,300 by year-end, implying hundreds more facilities competing for power over the coming months.
Faster turbine components would solve only part of the problem. SpaceX still has to prove it can make them reliably at scale, while Strazik has said construction, permitting, fuel supply and other infrastructure can also delay projects.
For years the question in AI was who could get the chips. Now it is increasingly who can get the power to run them.
Image: Shutterstock
Kalshi and Benzinga have an existing data collaboration agreement.
Login to comment