A turbine-parts maker fell 7.5% because Elon Musk wants to make the parts himself
Summary
Elon Musk said SpaceX will cast its own gas-turbine blades because the handful of foundries that can make them are the reason AI data centers cannot get power fast enough, and the biggest listed one lost 7.5% in a day. The company that just had the shortage confirmed by its largest potential customer got sold as if the shortage were over — and a newly listed competitor with the same bottleneck barely gets covered.
On August 29–30, 2026 Elon Musk confirmed on X that SpaceX will build a foundry in Bastrop, Texas to cast the hot-section blades and vanes used in industrial gas turbines, saying in-house casting could bring turbines online up to 18 months sooner. The Information first reported the plan and the Wall Street Journal later confirmed it. On September 1, 2026 Musk told the G20 that roughly 15 gigawatts of AI computing capacity built in 2027 may not be able to switch on that year for lack of power equipment.
Howmet Aerospace, the largest listed maker of such castings, fell 7.5% on August 31, 2026 to $244.95 — its steepest single-day drop since April 2025 — after falling as much as 9% during the session. On September 1 Bernstein called the reaction an overinterpretation, noting that Howmet’s customer agreements run through 2030 and that six expansion projects due by the end of 2026 could lift its turbine-blade capacity by about 38% versus the first quarter of 2025; it raised its price target to $328 from $248.
The casting field is concentrated among four firms: Howmet, Berkshire Hathaway’s Precision Castparts, Consolidated Precision Products and Doncasters. Doncasters, a UK-based maker of turbine castings and nickel- and cobalt-based superalloys with $837 million of 2025 revenue, listed on the NYSE on June 25, 2026 at $33 a share in an upsized $700 million IPO; its order backlog grew from $725 million in December 2025 to $930 million by March 29, 2026. GE Vernova’s gas-turbine production slots are reported as largely sold out through 2030.
Opportunity
The obvious reading — a deep-pocketed new entrant is coming for the incumbents’ business — treats blade casting like a factory you can simply build. It is closer to a craft: single-crystal and directionally solidified superalloy airfoils have to be qualified part by part by the turbine maker, a process that has historically taken years, and the entrant has announced a groundbreaking, not a qualified part.
Two things are more likely to be true at once. First, the announcement is the strongest confirmation yet, from the largest buyer of turbines outside the utilities, that casting capacity is the binding constraint on new gas generation — which is the incumbents’ pricing power, stated out loud. Second, any additional foundry, SpaceX’s included, still has to buy vacuum-melted nickel superalloy from a small number of alloy producers, so more casting capacity means more alloy demand rather than less.
Hypothesis: the market sold the confirmation of a bottleneck as if it were the removal of one. If qualification timelines hold, the hot-section economics of Howmet and Doncasters through 2030 are unchanged by SpaceX’s plan, and the newer listing, Doncasters, carries the same bottleneck exposure with far less coverage, while the superalloy suppliers upstream are the part of this chain that gains from a new entrant. That is inference; how quickly SpaceX can qualify parts is exactly the unknown.
How it could play out
AI data centers keep ordering gas turbines faster than blades can be cast, so turbine makers stay sold out through the decade. The casting bottleneck persists whatever SpaceX builds, because a new foundry needs years of qualification before its parts fly in a customer’s machine. Incumbents keep their pricing power and expand capacity against contracts that already run to 2030, and alloy melters see incremental demand from every new foundry that breaks ground. The sell-off in the largest incumbent then reads as a misread once quarterly results show the backlog intact — or, if SpaceX qualifies parts faster than any precedent, the turbine makers gain a new supplier and the incumbents’ terms weaken from 2028 onward.
Questions worth asking
- How long does it take a new foundry to qualify a hot-section blade with GE Vernova, Siemens Energy or Mitsubishi Power, and has any new entrant ever done it in under three years? This is the question that decides the idea.
- What share of Howmet’s and Doncasters’ revenue and profit actually comes from industrial gas turbines rather than aerospace, and how much of the turbine work sits under agreements that run to 2030?
- Who melts the nickel superalloy a new foundry would need, and is that capacity already spoken for? If it is, SpaceX’s foundry competes with the incumbents for alloy rather than displacing them.
- Does SpaceX intend to sell castings to others or only feed its own turbines? A captive supplier removes a customer from the market; a merchant supplier adds capacity to it.
- Doncasters listed in June 2026 with a backlog that grew 28% in a single quarter. Is that growth priced, or is a three-month-old small cap simply unfollowed?
Where to look
- Howmet Aerospace — the largest listed maker of hot-section castings, and the stock that absorbed the reaction
- DPC Holdings (Doncasters) — the June 2026 listing with turbine castings and its own superalloy plants, and the least-covered expression
- ATI and Carpenter Technology — nickel superalloy melters that supply any foundry, including a new one
- GE Vernova and Siemens Energy — the turbine makers whose delivery times measure whether the bottleneck is actually easing
- Berkshire Hathaway — owns Precision Castparts, the other large incumbent, inside a conglomerate
Thesis check
The facts are unusually clean: Musk’s own words identify casting as the constraint, the incumbents’ backlogs and 2030 agreements are disclosed, and the price reaction is a matter of record. The weakness is that Howmet is a widely followed large cap that Bernstein has already publicly defended, so the easy part of any misprice may be closing, and the whole idea rests on qualification timelines that are known to be long but that a company with SpaceX’s engineering record could compress in ways no precedent covers. For Doncasters specifically, industrial gas turbines are one segment alongside aerospace, so that exposure has to be sized before the argument means anything.
Timing
Howmet and Doncasters report third-quarter results in late October and early November 2026
Sources
TechCrunch, Aug 30 2026 · Yahoo Finance / Stocktwits, Aug 31 2026 · GuruFocus on Bernstein, Sep 1 2026 · Quartz, Sep 2 2026 · Business Wire, Doncasters IPO pricing, Jun 24 2026 · Benzinga, Aug 2026 · Yahoo Finance, Musk at the G20, Sep 2026