Signals Inbox·September 2, 2026·SpaceTech

Why is SpaceX building a turbine-blade factory?

SpaceX is building a turbine-blade factory because AI has made electricity a growth bottleneck, and waiting roughly three years for the existing gas-turbine supply chain no longer matches how fast it wants to add compute. The unusual part is how far upstream SpaceX is now willing to go to fix that problem.

Q1Is SpaceX really building a turbine-blade factory?

Yes. SpaceX is currently building a specialized turbine-blade and vane foundry in Bastrop, Texas, mainly to make parts for gas turbines that can power AI data centers.

SpaceX’s current job postings are unusually specific about what the factory will do. The company is hiring engineers to build and ramp a new blades-and-vanes foundry next to its existing Bastrop operations. The roles cover investment casting, nickel superalloys, directionally solidified parts and single-crystal components, all technologies used for the hottest sections of modern gas turbines.

The factory therefore goes much further than producing ordinary metal parts. SpaceX is moving into one of the hardest manufacturing steps in the power-generation supply chain.

The timing also removes much of the mystery. SpaceX has been spending heavily on AI infrastructure, buying mobile gas turbines and expanding its data-center power capacity. The foundry appeared as the company started running into the industrial limits of that expansion.

Q2Why does SpaceX suddenly need its own power plants?

SpaceX now needs enormous amounts of electricity because AI compute has become a major part of its business, and the company wants to add computing capacity much faster than utilities can normally add grid power.

The change happened quickly. After SpaceX absorbed xAI, the company inherited the Colossus data-center infrastructure and became a large AI-compute operator alongside its rocket and Starlink businesses.

SpaceX then accelerated again. On its latest earnings call, Elon Musk said the company expects to end next year closer to 10 gigawatts of compute capacity than five. He also said SpaceX is developing projects representing roughly 20 gigawatts of power and cooling infrastructure, with something closer to 15 gigawatts of actual power capacity looking achievable on that timeline.

Those are power-system numbers, not normal corporate data-center numbers. One gigawatt is already the scale of a large power station.

SpaceX also spent almost $16 billion on AI infrastructure in its latest reported quarter. Once a company is deploying capital at that speed, waiting several years for electricity can become more expensive than building power-generation capacity itself.

Q3Are turbine blades really what is slowing SpaceX’s gas turbines down?

Yes, according to SpaceX. Elon Musk says casting turbine blades and vanes has become a major bottleneck in getting more natural-gas generation online, and SpaceX believes internal production could remove as much as 18 months of delay.

That claim makes sense when we look at how these parts are manufactured. The hottest turbine blades operate under extreme temperatures and mechanical stress. High-performance versions use nickel-based superalloys and can be cast as directionally solidified or single-crystal structures to survive those conditions.

Very few factories can produce these components reliably at scale.

Musk’s argument is that ordering more complete turbines does not fully solve SpaceX’s problem if the companies assembling those turbines are themselves waiting for hot-section castings. SpaceX has therefore gone one level deeper into the supply chain.

It has identified a specific manufacturing process that it believes is slowing the whole machine down, and it is trying to increase that process directly.

Q4How bad is the gas-turbine shortage right now?

Gas turbines are genuinely hard to get quickly these days, with major manufacturers already allocating production several years ahead.

GE Vernova gives us one of the clearest benchmarks. In its latest quarterly results, the company said its gas-power equipment backlog and slot reservations had jumped from 100 gigawatts to 116 gigawatts in just one quarter. GE Vernova now expects at least 125 gigawatts under contract by year-end.

Its manufacturing capacity is much smaller than that backlog. GE Vernova is moving toward an annual production rate of roughly 20 gigawatts, then plans to reach 24 gigawatts in 2028 and 30 gigawatts in 2030.

Using current output as a rough benchmark, the 116-gigawatt pipeline equals almost six years of production. Actual delivery schedules are more complicated because capacity is rising and projects are spread across several years, but the order of magnitude shows why customers are scrambling for slots.

GE Vernova CEO Scott Strazik has also said lead times are currently around three years. Customers have already been booking turbines for 2029 and 2030.

Current gas-turbine supply indicators

Q5Why can’t SpaceX just wait for grid power?

SpaceX cannot grow AI infrastructure at its current pace if every new data center has to wait several years for a normal grid connection.

APR Energy, the mobile-power company SpaceX recently acquired, explains the attraction of the alternative. APR says its modular gas-turbine plants can normally be delivered, installed and commissioned in 30 to 90 days. Its systems can scale from tens of megawatts to several hundred megawatts at a site.

Permanent power infrastructure moves much more slowly. APR itself compares its 30-to-90-day deployments with the two to five years often required for permanent grid capacity.

That time gap is enormous for an AI company. Nvidia can ship GPUs, contractors can build data halls and networking equipment can be installed while the utility is still working through substations, transmission upgrades, generation and interconnection.

Mobile gas generation lets SpaceX run those processes in parallel instead of waiting for the grid to finish first.

There are limits. Permits, fuel pipelines, switchgear, transformers and local environmental rules can still slow a project. But mobile generation gives SpaceX much more control over the schedule than a normal utility connection does.

Q6How much power does SpaceX actually need for AI?

SpaceX is currently planning AI infrastructure on a scale where even a billion-dollar mobile-turbine fleet only covers a small piece of what the company wants.

APR Energy owned more than 1.1 gigawatts of mobile generation capacity before SpaceX acquired it. That is already one of the larger mobile gas-turbine fleets in the world.

Compare that with SpaceX’s own plans. Management has discussed getting closer to 10 gigawatts of compute capacity by the end of next year and having roughly 15 gigawatts of power infrastructure available around the same period.

Even if APR’s entire 1.1-gigawatt fleet could be redirected to SpaceX, it would equal only about 7% of a 15-gigawatt power buildout.

The comparison is approximate because generator nameplate capacity and usable compute power are different things. Cooling, power conversion and other systems consume electricity too. But we do not need perfect equivalence to see the scale problem.

SpaceX is trying to add several APR-sized fleets worth of power.

SpaceX AI infrastructure scale

Q7How aggressively has SpaceX already been buying gas turbines?

SpaceX has already committed billions of dollars to gas turbines, so the new blade factory continues a strategy that was well underway before the foundry became public.

SpaceX disclosed in its IPO filings that its AI business expects to buy roughly $2.8 billion of turbine infrastructure over three years. At least $2 billion of that commitment is specifically for mobile gas turbines.

The company also acquired APR Energy. The exact purchase price was never formally announced, but the transaction can be partly reconstructed from Duos Technologies, which owned 5% of APR’s parent company. Duos disclosed receiving $50.4 million in cash plus additional escrow proceeds from its stake, putting the implied equity value somewhere around $1 billion to $1.2 billion depending on how the escrow is treated.

APR had already expanded aggressively before the deal. It bought eight additional turbines earlier this year, taking its fleet from 850 megawatts to more than 1.1 gigawatts. The company was also supplying 375 megawatts to a single large AI data center.

That sequence tells us SpaceX did not suddenly discover gas power when it announced the foundry. It had already been spending heavily to secure turbines themselves.

Q8Why buy APR Energy and still build a turbine foundry?

SpaceX bought APR Energy for power it can deploy now, while the turbine foundry is meant to make the next wave of power equipment arrive faster.

APR gives SpaceX an existing fleet, operational teams and experience deploying temporary or bridge power. That solves part of the immediate problem.

Owning APR does not magically create another thousand turbines. APR still depends on manufacturers such as GE and Mitsubishi and on the specialist suppliers that produce hot-section parts for those machines.

SpaceX is now trying to loosen that dependency.

This is a familiar SpaceX pattern. The company often starts by buying from an established supplier, learns which part of the supply chain controls speed or cost, and then brings unusually painful parts in-house. SpaceX followed versions of that playbook with rocket structures, engines, avionics and Starlink hardware.

Power generation is getting the same treatment.

Q9Can SpaceX actually make turbine blades good enough for power plants?

SpaceX has unusually relevant manufacturing experience, but making reliable power-turbine blades at scale will still be hard even for SpaceX.

Raptor rocket engines already force SpaceX to work with extreme temperatures, turbomachinery and advanced metal parts. Musk has spoken publicly about years spent solving turbine-blade cracking problems inside SpaceX’s rocket program.

The overlap extends into manufacturing. Current SpaceX foundry roles call for expertise in nickel superalloys, directional solidification, single-crystal casting, investment casting, process automation and advanced inspection. Those are directly relevant to high-performance gas-turbine blades.

SpaceX therefore starts much further ahead than a normal company entering this industry.

Power-generation blades still face a different operating environment. A rocket engine runs under brutal conditions for relatively short periods. Industrial gas turbines may run for thousands of hours and need predictable maintenance intervals, consistent efficiency and very low failure rates.

SpaceX must prove that its castings can survive that kind of service repeatedly, not just that it knows how to pour sophisticated alloys.

Q10Can SpaceX really save 18 months by casting turbine blades itself?

SpaceX could plausibly remove months from turbine delivery schedules, but the full 18-month saving remains an internal target that has not yet been demonstrated.

The surrounding market gives the claim some credibility. GE Vernova is currently talking about roughly three-year turbine lead times. Specialist casting capacity is tight enough that turbine suppliers are expanding manufacturing years into the future.

Removing a major casting queue could therefore have a large effect.

Still, a new foundry has its own ramp. SpaceX has to install equipment, stabilize casting processes, improve yields, inspect components and qualify them for use. A badly cast turbine blade can destroy much more expensive equipment around it, so qualification cannot simply disappear in the name of speed.

GE Vernova has also made an important point about the broader industry. On many conventional projects, turbines themselves are not the only reason a plant takes three years. Permitting, engineering, construction and fuel availability can take just as long.

SpaceX has a better chance of realizing large time savings because it is using mobile and behind-the-meter generation, where some of those other steps can move much faster.

For now, the 18-month figure is believable enough to take seriously, but too new to treat as a proven result.

Q11Is saving 18 months on power really worth building a whole factory?

For SpaceX, getting AI power online earlier can be worth billions of dollars, which makes a specialized turbine foundry economically plausible even before we count strategic benefits.

The clearest evidence comes from SpaceX’s compute contracts.

Google has agreed to pay SpaceX $920 million per month for access to roughly 110,000 Nvidia GPUs plus related infrastructure once the capacity is fully ramped. SpaceX also has a separate Anthropic agreement worth $1.25 billion per month for Colossus I compute capacity.

Those contracts show how much money can sit behind an energized data center.

We should not assume every future SpaceX GPU earns revenue at the same rate. The Google contract nevertheless gives us a useful scale reference. Six months at $920 million per month equals $5.52 billion in gross contractual revenue. Twelve months equals just over $11 billion.

As seen above, SpaceX says internal blade casting could remove as much as 18 months from turbine delivery in the best case. At the revenue density of the Google contract, 18 months represents $16.56 billion.

That does not mean the foundry automatically creates another $16.56 billion of revenue. It shows why the timing is worth fighting over: when powered compute can support contracts approaching $1 billion per month, losing months gets very expensive.

Value of earlier compute availability at $920 million per month

Q12Is SpaceX building the turbine foundry for Starship too?

Starship probably benefits from the same foundry expertise, although the immediate reason for SpaceX’s expansion is clearly AI power generation.

There is real technical overlap. Raptor engines use high-temperature turbomachinery, while gas turbines use many of the same families of advanced alloys and casting equipment.

Morgan Stanley analyst Adam Jonas recently pointed out that Raptor turbopump castings and gas-turbine airfoils can use similar alloys and furnaces. A large advanced foundry can therefore spread equipment, engineers and metallurgical know-how across both businesses.

SpaceX is also trying to increase Starship production dramatically. Better internal casting capacity could help rocket manufacturing even if the first blades coming from the Bastrop expansion are aimed at terrestrial power equipment.

We should keep the order of causality straight, though. The new factory appeared while SpaceX was rushing to secure gigawatts of electricity, and Musk himself explained the project in terms of speeding up natural-gas turbines.

Starship makes the investment more useful. AI power appears to be what made it urgent.

Q13Is SpaceX trying to become a gas-turbine company?

SpaceX is currently behaving more like a huge turbine customer that wants control over its supply chain than a new competitor to GE Vernova.

A complete industrial gas turbine requires far more than blades and vanes. Manufacturers also have to build compressors, combustion systems, shafts, bearings, generators, controls and packaging, then test complete machines and support them for decades.

SpaceX has announced a foundry. It has not announced a SpaceX-branded commercial turbine, a utility sales business or a full turbine product line.

Financial markets still reacted immediately to the news. Howmet Aerospace, a major producer of complex turbine castings, fell about 7.5% after Musk highlighted the casting bottleneck. GE Vernova also dropped.

The latest analyst reaction has been much calmer than the initial stock move. Citi, Bernstein and Deutsche Bank analysts have argued that SpaceX could end up working with existing turbine companies instead of displacing them. That fits what we currently know about the project.

APR Energy creates another possibility. Because APR already supplies outside customers, SpaceX could eventually use its manufacturing capability to expand an external power business without ever becoming a traditional turbine OEM.

Today, though, the strongest evidence still points to vertical integration for SpaceX’s own power needs.

Q14Does SpaceX’s turbine factory mean it is choosing gas over solar?

No. SpaceX currently sees natural gas as the fastest bridge to huge amounts of AI power while it also tries to scale solar much further.

Musk has said SpaceX and Tesla are each working toward solar-manufacturing capacity of 100 gigawatts per year. That is an ambition rather than current output, but it shows how large the solar plan is.

Gas solves a different problem.

A mobile gas turbine can run day and night, fit into a relatively compact site and come online within months. Large solar installations need land and enough batteries, grid support or another generation source to keep a dense AI data center running continuously.

Musk has described natural gas as a way to “supplement and bootstrap” solar for the next several years. That wording matches what SpaceX is actually building.

Longer term, SpaceX is even considering AI compute in orbit, where solar power becomes far more attractive because satellites can access abundant sunlight without terrestrial land or grid constraints.

Gas is currently the speed solution. Solar remains central to the larger energy plan.

Q15What could still derail SpaceX’s turbine plan?

SpaceX can solve turbine casting and still run into several other hard limits before those blades translate into working AI data centers.

The first risk is manufacturing yield. Single-crystal and directionally solidified turbine parts are difficult to produce consistently. A factory can technically make a blade while still throwing away too many defective parts to achieve useful volume.

Qualification comes next. SpaceX will have to prove that its components last under long-duration power-generation conditions.

Then the bottleneck can simply move somewhere else. Musk himself has mentioned transformers, switchgear, cabling, cooling equipment and networking gear while discussing the AI expansion. Fuel infrastructure can become another constraint once gas generation reaches gigawatt scale.

Permitting is already a live problem around SpaceX’s AI power strategy. Gas turbines used around xAI facilities have faced lawsuits and complaints over emissions and noise. Mobile equipment can arrive quickly, but local and federal rules can still slow its operation.

There is also a strategic risk. If utilities, solar, batteries or other power sources catch up faster than SpaceX expects, the value of squeezing every possible month from gas-turbine production will fall.

The foundry gives SpaceX more control over one bottleneck. It does not give SpaceX control over the whole energy system.

Q16So why is SpaceX building a turbine-blade factory?

SpaceX is building the turbine-blade factory because AI has turned electricity into a growth constraint, and waiting years for the existing gas-turbine supply chain no longer fits the speed at which SpaceX wants to add compute.

Put the recent moves together and the strategy becomes fairly clear.

SpaceX absorbed xAI and became a major data-center operator. Its AI infrastructure spending then jumped to almost $16 billion in a single quarter. Management started talking about approaching 10 gigawatts of compute and building power infrastructure on an even larger scale. SpaceX committed billions to gas turbines and acquired APR Energy to gain fast-deployment generation capacity.

The foundry takes the company one step further upstream.

SpaceX believes turbine-blade casting is slowing new power generation enough to justify bringing the process inside the company. That is especially valuable when GE Vernova is carrying a 116-gigawatt pipeline and talking about roughly three-year turbine lead times.

There is also a simple economic reason for the urgency. SpaceX now has individual compute contracts worth close to or above $1 billion per month. A data center that gets electricity six months earlier can start earning money six months earlier.

That is why a seemingly obscure turbine-blade factory suddenly makes sense inside a rocket company.

SpaceX wants to build AI compute faster than the power industry can currently supply it, so it is starting to manufacture one of the components holding the power industry back.