With the rapidly rising demand for electricity in the AI data centre, the supply of electricity is becoming an additional constraint beyond the chip. Mask has recently indicated that SpaceX is building in-house casting capacity for critical components of gas turbines in Bastropu, Texas, in an attempt to shorten the equipment delivery cycle and mitigate the problem of slow powering in the data centre.
On August 30th, on platform X, it was stated that SpaceX and Tesla were building a solar energy capacity of 100 GW per year as soon as possible, but that natural gas would continue to be used in the coming years to replenish solar energy and support the initial power supply. He stated that the main bottleneck in the current expansion of gas turbines was the casting of leaves and directional blades, which, if instead produced by SpaceX itself, could go online at most 18 months in advance.
AI Data centre to self-supply
The background to this move is that the infrastructure building of AI is facing both chip and grid constraints. The International Energy Agency expects that global data centre power will nearly double by 2030. The gas turbine manufacturer, GE Vernova, also indicated that its capacity had been largely reserved for 2030 and that a large part of the demand came from AI infrastructure.
In this context, an increasing number of megatech companies have started to build gas power facilities directly near data centres rather than waiting for public grid expansion. The report mentions that Amazon, Google, Meta, OpenAI and Microsoft are using this approach to facilitate faster transport of data centres.
Highly concentrated supply of critical spare parts
The most difficult part of gas turbines is the leaves and directional leaves used in high-temperature areas. It was reported that the components work at temperatures of approximately 300 to 3600 degrees Fahrenheit, higher than the melting point of the material itself, and therefore have to rely on internal cooling corridors, heat barrier coatings and very precise casting processes.
Even more difficult, such leaves usually need to be found in single-crystal form, slowly in a vacuum furnace and as little as possible to avoid cracks. The size of the leaves used in power generation equipment is greater than that of the air engine, and the volume is more difficult to produce. It is reported that only four companies globally now possess sufficiently mature industrial casting capacity and that existing capacity is close to full.
If SpaceX can make this breakthrough, it means that the system controlled by Mask will have a scarce capacity in the AI power infrastructure. This would allow it to move faster on the data centre package and raise the threshold for replication by other competitors.
Faster expansion is associated with emission pressure.
However, faster deployment of gas turbines would also lead to more direct environmental disputes. It is reported that SpaceXAI has been powering the Colossus data centre in Memphis since 2024. Local NAACP has repeatedly alleged that it operated the equipment under conditions of licence or pollution control that did not meet the requirements of federal law.
Organizations are concerned about the release of smoke-producing compounds and harmful substances such as formaldehyde, which are associated with asthma, respiratory diseases and some cancer risks. Researchers at Memphis University have also indicated in a limited analysis that air pollution around data centres “slightly deteriorated”.
Similar controversy is not confined to Memphis. It was also mentioned that a study based on the US EPA Health Impact Model (HIM) in the data centre concentration area of Virginia had shown that if a single facility operated on a long-term basis,8 gas turbines could have an emission impact of more than 2.5 million people, with additional health losses and economic costs.
On the whole, Mask is trying to solve the real power constraints in AI ' s expansion, but to introduce gas units more quickly and to further bind data centres to emissions.