Michael Nicolls, a senior Starlink executive, has been handed control of SpaceX’s AI infrastructure following a series of leadership departures.
SpaceX (NASDAQ: SPCX) faces growing pressure to maintain stable operations for its compute customers, including Google and Anthropic.
What went wrong with SpaceX’s data centers?
Michael Nicolls has been appointed to oversee xAI’s infrastructure after several senior data center executives left the company over the recent weeks.
The system Nicolls will be taking over is one SpaceX has bragged about building at a low cost. The company said its second data center brought its first two computing clusters online for roughly a quarter of what is typically paid in the industry. However, the company did not spell out which expenses were included in that estimate.
The Information reported that SpaceX compromised other factors in order to build its data centers quickly. For instance, xAI’s Macrohard site used temporary equipment to get running, including mobile gas turbines, Tesla Megapack batteries, and more than 100 portable chillers.
A person who worked on the project reportedly stated that this makeshift setup caused outages that disrupted AI model training. The site’s uptime consistently fell below the company’s internal goal of 99.9%.
Some SpaceX facilities in Tennessee and Mississippi reportedly ran for months with no backup cooling or power at all, due to the company’s race to stack up compute. SpaceX has since pushed back plans for data centers beyond its Memphis hub, including proposed sites in Texas.
The delays cast doubt on the company’s plans to generate more than two gigawatts of computing capacity by the end of 2026. SpaceX reported in June that it only has about 1.4 gigawatts online.
What is SpaceX doing to fill its vacancies?
Following the exit of several executives like infrastructure chief Jake Palmer in July, SpaceX is moving employees from its rocket and Starlink programs to fill the gap in its data centers.
For instance, Wesley Salandro, who used to be a senior production manager for the Falcon and Dragon rocket lines, has joined the data center team. Logan McConnell, a product-support director at SpaceX’s Starbase site in Texas, has also been moved to a new role.
SpaceX has also brought in an engineer from Neuralink, Elon Musk’s brain-implant company, and is now hiring for several civil-engineering positions.
What happens if SpaceX’s data centers fail?
Selling computing power to Anthropic and Google increased SpaceX’s AI revenue to $2.6 billion in the second quarter of 2026, compared to $737 million in the same period last year.
Google reportedly has a compute deal worth $920 million per month for about 110,000 Nvidia GPUs, while Anthropic has a lease on SpaceX’s Colossus system that costs as much as $1.25 billion per month, although that agreement can be canceled.
An earlier report from Cryptopolitan said that SpaceX’s Colossus contracts with Anthropic, Google, and Reflection AI are expected to generate over $28 billion each year.
If SpaceX cannot keep its systems running reliably, that revenue is at risk. The company must either spend more money to fix the issues or reduce the computing power available to its paying customers.
Despite these concerns, SPCX closed near $143.36 on September 1, just below the previous day’s close and roughly 36% below its June peak of $225.64. The stock has bounced back from an August low of about $104.83.
Elon Musk is trying to take control of more of the supply chain, confirming on Saturday that SpaceX is building a foundry in Bastrop, Texas, to make its own gas-turbine blades and vanes.
He said this move could speed up natural-gas turbine deployment by as much as 18 months, but the NAACP has accused SpaceX of operating turbines without proper permits at its Memphis site.
Be the first to comment