SpaceX Shifts AI Data Center Buildout to Reliability, Taps Rocket Engineers

SpaceX is adjusting how it builds its AI data centers, giving greater priority to the reliability of power and cooling systems and to pre-commissioning testing. Previously, xAI's data center construction put speed and computing scale first. After xAI was folded into SpaceX, SpaceX changed that approach, shifting new campuses from rapid commissioning toward reliability-focused construction, with correspondingly longer build times.

The shift differs from how xAI built the Colossus data center in Memphis. That project installed 100,000 GPUs and brought them into service within 122 days. To compress the schedule, construction teams advanced cooling piping, power and networking work in parallel; the data center was running before some backup systems were completed and relied on temporary power sources such as gas turbines for an extended period. Its AI business built a data center at record speed in 2024. Rapid construction also brought operational risks: frequent early power outages forced AI researchers to pause projects and cost some model training progress. On September 3, 2026, construction activity affected power lines at a Memphis data center, taking some Grok models offline and affecting some external computing customers. Musk subsequently said the company was taking corrective measures to prevent similar incidents.

The strategy change follows a management reshuffle. SpaceX completed its acquisition of xAI in February 2026, and from late June 2026 a rocket manufacturing and operations team began taking over the data center business. Wesley Salandro, who has more than seven years at SpaceX and previously oversaw Falcon rocket and Dragon spacecraft production, succeeded Dan Rowland in leading the team. At the same time, more than a dozen project leaders have recently departed, spanning infrastructure, safety, procurement and finance.

SpaceX has also recruited staff from its rocket and Starlink divisions for rotational support. Of more than 1,300 volunteers, over 300 engineers have joined the data center project. The new management has added middle-management positions and tightened on-site staff authority to purchase equipment and change construction plans. As SpaceX begins leasing computing capacity to external customers such as Anthropic and Google, the impact of data center downtime extends beyond Grok's own business, raising the importance of reliability.