Elon Musk spent the weekend describing a solution and its limit.
On Saturday, he confirmed that SpaceX is preparing to cast the hot-section blades and vanes used inside industrial gas turbines at a new operation in Bastrop, Texas. Doing the casting in-house, he said, could bring gas turbines online up to 18 months earlier.
Hours later, he endorsed an estimate that roughly 15 GW of AI computing hardware produced in 2027 "cannot be turned on in 2027." Turbines are only one link. The rest of his list was transformers, wiring, liquid cooling, large chillers and networking.
The claims fit together. SpaceX is attacking the first visible bottleneck. If it succeeds, Musk's own list says where the constraint moves next.
The summary table keeps four different kinds of figure separate: Musk's 15 GW and 18-month claims, SpaceX's 10 GW target, APR Energy's disclosed fleet and our 4.89 GW model output.
A hiring plan is not yet turbine supply
The Information reported job listings for the construction and ramp of a new blades-and-vanes foundry in Bastrop, plus a machining role explicitly tied to a new industrial-gas-turbine manufacturing line. The same report found about 830 acres acquired near SpaceX's existing Starlink operation between March and June, while noting that no foundry permit or construction filing had been located.
The evidence supports a manufacturing line being staffed and laid out. It does not establish when a qualified production blade will leave it, how many blades it can make or which turbine design it will serve.
The part is difficult for a reason. A turbine's hottest blades operate above the temperature at which their underlying alloy would ordinarily melt. They survive through internal cooling passages, thermal coatings and single-crystal casting without the grain boundaries where cracks can begin. SpaceX knows adjacent physics from Raptor turbopumps. Adjacent is not identical: industrial turbine airfoils are larger, and a casting line does not become qualified production because its vacuum furnace switches on.
Musk had already named the constraint before the Bastrop plan became the headline: turbines are allocated through 2030, and his companies may have to make the blades and vanes themselves.
The useful limit is what SpaceX has not disclosed: output, yield, customers, cost or a production start date. Without those figures, "up to 18 months" cannot honestly be converted into gigawatts.
APR is a bridge, not the answer
A Federal Trade Commission notice dated May 14 identifies Elon Musk as the acquiring party for New APR Energy. APR said in January that its mobile turbine fleet had expanded from 850 MW to more than 1.1 GW. Its GE TM2500 and Mitsubishi FT8 units are rated at 20–35 MW each, and APR markets deployment in 30 to 90 days. The FTC notice records antitrust clearance; The Information reports that Fortress sold the business to Musk.
The sequence is clear: buy mobile generation that already exists while building the capability to relieve the component shortage on the permanent fleet.
Now put the bridge against the job:
- APR's >1.1 GW is more than 7.3% of the 15 GW Musk says may sit dark across the industry in 2027.
- It is more than 11% of SpaceX's stated 10 GW online by the end of 2027 target.
- It is more than 22.5% of the 4.89 GW of end-2027 terrestrial capacity in our SpaceX model.
Those ratios are ours, and they are ceilings on relevance rather than forecasts. The FTC filing names Musk personally, not SpaceX; APR has other customers; and owning 1.1 GW of turbines does not mean all 1.1 GW is idle, available or destined for a Musk-controlled site.
The acquisition still matters. Existing fleet first, castings second, permanent generation later. But an entire mobile-power company covers only a small fraction of the capacity in Musk's own warning.
Saving 18 months on blades does not save 18 months on the campus
GE Vernova's own answer is the cleanest check on the thesis. In April, CEO Scott Strazik said the company had roughly 10 GW of remaining turbine capacity across 2029 and 2030 and a lead time of about three years. In the same answer, he said turbines often are not the gating item in a three-year project once engineering, construction, permitting and fuel availability are counted.
That is not an argument against the foundry. It is the reason to look two steps ahead.
A turbine turns fuel into electricity. A data centre still needs transformers and substations to change and distribute the voltage, switchgear and cabling to move it, cooling equipment to remove the heat, and networking to turn powered racks into a cluster. All of it must reach the same site, in the right order, and pass commissioning. Moving one component forward by 18 months does not move the whole chain forward by 18 months.
It can also move the constraint into permits. Faster gas generation means faster fuel and emissions decisions. TechCrunch documented the dispute around mobile turbines at the Memphis campuses and a Virginia study estimating health damage from continuously operated on-site turbines. The point is not that every turbine has the same emissions profile. It is that manufacturing speed does not waive air permits, pollution controls or local opposition. Those clocks belong in the activation schedule too.
Tesla supplies storage and solar, not the gas
Tesla sits in this story twice.
First, Musk said Tesla and SpaceX are each trying to build 100 GW a year of US solar manufacturing. Tesla's reported Brookshire panel plant carries that same 100 GW design target, but our earlier work found no Tesla filing with a production date, volume target or capex figure. It is an ambition, not available generation.
Second, Megapacks solve a different problem. They smooth power, bridge interruptions and replace some backup generation; they do not create the primary energy that charges them. Musk made the distinction explicit on Sunday: Megapacks can solve Terafab's power quality, while there is not enough nearby land to run the facility entirely on solar. It will be only partially solar-powered.
That is why SpaceX's $835 million of disclosed Megapack purchases and a SpaceX turbine foundry are complementary rather than competing plans. Tesla supplies storage and, eventually, more solar hardware. Gas supplies dispatchable bridge power. SpaceX is trying to own the scarce casting between the two.
The foundry does not change either model yet
Our SpaceX model begins with 1.8 GW energised in 2026 Q2 and adds 400 MW a quarter, with the addition itself growing 10% each quarter. That reaches 4.89 GW at the end of 2027 — less than half SpaceX's 10 GW target, and 32.6% of the 15 GW Musk says the industry may fail to power.
We are not changing it for the foundry. The model already treats energised capacity as the constraint, and no disclosed foundry output can replace its build-rate assumption. The news raises confidence that SpaceX understands the constraint; it supplies no number that says SpaceX has solved it.
Our Tesla model also does not change. It already carries storage and solar separately: batteries sold by the gigawatt-hour, solar as a much smaller line whose reported Texas expansion arrives late in the horizon. A turbine foundry owned by SpaceX is neither Tesla revenue nor Tesla capex. Megapack purchases become Tesla revenue only when the related-party orders are placed and delivered.
What to watch
- A first qualified blade and vane from Bastrop. A hiring page proves intent; a production part accepted into a named turbine proves manufacturing.
- Annual capacity and yield. Without parts per year and an accepted yield, "18 months" cannot be converted into turbines or gigawatts.
- Where APR's >1.1 GW is deployed. A SpaceX, Terafab or Colossus allocation would turn a personal acquisition into visible group capacity.
- The next bottleneck disclosed. Transformer deliveries, substation energisation, cooling capacity and network commissioning will show whether the constraint migrated as expected.
- Permits and emissions controls. A faster turbine that cannot legally operate is still dark capacity.
SpaceX is doing what Musk's companies do when a supplier sets the schedule: buy what exists, then try to manufacture the bottleneck. The foundry may work. If it does, the 15 GW warning says exactly what happens next. The scarce object stops being the blade and becomes the completed system.
Sources and provenance. The Bastrop foundry, job listings, land purchases, reported APR sale and absence of located permits are reported by The Information, 29 August. The 18-month, 15 GW, two 100 GW solar and Terafab comments are Musk's claims on X from 29–30 August, not company guidance. SpaceX's 10 GW target is from its 2026 Q2 results call. The FTC identifies Musk as APR's acquiring party and records clearance rather than closing; APR's fleet capacity, unit range and deployment time are APR disclosures. GE Vernova's available slots and project-cycle comments are from its 22 April earnings call. The pollution and health-impact discussion is reported by TechCrunch and remains site- and equipment-specific. The 4.89 GW model case and every percentage comparison are R40 estimates or arithmetic, not company figures.
