SpaceX Fully Adjusts AI Data Center Construction Model: Expansion Slows, Reliability Strengthened

SpaceX is undergoing a systematic overhaul of its data center construction approach, shifting its strategy from prioritizing speed to emphasizing reliability—a significant turning point for Musk's AI venture in its infrastructure expansion path.

According to The Information, citing insiders, SpaceX’s new management team is intensifying efforts to install more backup power and cooling systems, requiring more comprehensive testing before data centers go online, and beginning to reduce reliance on temporary power and cooling equipment. This could significantly slow the pace of data center expansion.

This approach forms a stark contrast to xAI’s previous rapid deployment model of “building and operating simultaneously, supplementing redundancy later.” At the same time, management is advancing redesign projects at several existing data centers.

Last week, construction work inadvertently affected a power transmission line, leading to a high-profile outage at SpaceX’s large data center in Memphis, Tennessee.

SpaceX stated that this failure caused portions of the Grok model to go offline and triggered cascading effects for compute rental clients such as Anthropic and Google. Musk later posted on X, stating that the company is “taking corrective measures to ensure such incidents do not recur.”

As of press time, SpaceX rose 1.63% in intraday US trading on Thursday, having surged more than 3% earlier in the session on news of a roughly $40 million satellite services order from the UK government before pulling back somewhat.

SpaceX previously considered rapid data center construction as a core advantage in the AI race.

xAI once claimed that its first data center completed the online deployment of 100,000 GPUs in only 122 days—a feat, in an industry where similar-sized centers take years to build, that Nvidia CEO Jensen Huang praised as “superhuman” and “incredible.”

To achieve this speed, xAI mobilized thousands of contractors and adopted multi-process parallel construction—cooling pipe installation, electrical and network cabling, and other processes often took place in the same area simultaneously, rather than waiting for prior stages to complete before starting the next.

SpaceX even highlighted the rapid construction capabilities as a major selling point in its IPO filing this June. However, this aggressive pace has come at a clear cost to reliability.

According to insiders, the first two data centers launched with redundancy levels only a fraction of those at other hyperscale cloud providers, meaning any single device failure could cripple the entire compute cluster. Early on, frequent power outages sometimes forced AI researchers to halt projects for hours at a time, resulting in partial loss of model training progress.

Additionally, because team members were previously granted considerable purchasing autonomy, there was inconsistency in equipment specifications within the data centers, further undermining system stability.

At the end of June this year, Musk implemented a significant overhaul of the data center management team, appointing SpaceX veteran Wesley Salandro to replace Dan Rowland as team lead.

Rowland previously led infrastructure engineering and at times reported directly to Musk; prior to joining xAI he was involved in Tesla’s Dojo supercomputer project. He left the team soon after Salandro took over. Salandro has been with SpaceX for over seven years, formerly serving as Vice President of Production for the Falcon and Dragon rocket programs.

Several core leaders departed in the weeks that followed:

Jake Palmer (physical infrastructure), Logan Beach (data center security), Tiffany Wilson (procurement), and Pablo Mendoza (finance) have all left.

Duke Energy site project engineer Aakash Verma moved to OpenAI in August, with site construction manager Brent Mayo following suit.

Data center design contributor Liz Balke has now joined Anthropic’s infrastructure team.

To fill the manpower gap, SpaceX issued a call for volunteers among rocket and Starlink engineers in Bastrop, Texas and Cape Canaveral, Florida, recruiting for rotations of six to eight weeks. Reportedly, over 1,300 engineers signed up, with more than 300 formally joining the data center project in the past month.

The new management not only shifted the technical direction but also reshaped the organizational structure. According to insiders, SpaceX has added several mid-management positions to what was previously a relatively flat hierarchy, with employees’ equipment procurement and on-site decision-making powers now tightened—a move that will objectively slow construction further.

Musk himself appears to support this new strategy. According to two sources, he has visited the Memphis data center nearly every week over the past few months, typically stopping by on Sundays—the latest visit being just last Sunday.

Sources said that while constructing backup systems ahead of time and strengthening testing will surely slow down the overall pace, SpaceX is unlikely to completely abandon xAI’s previous rapid construction mode.

At the same time, balancing the rebuilding of reliability with maintaining the pace of expansion remains the core challenge for the new team—especially as the compute rental business scales up and outside client dependency increases.