SpaceX built its first Colossus data center at a pace Nvidia CEO Jensen Huang reportedly characterized as “superhuman,” according to Business Insider. Reports of outages at Memphis-area facilities followed. Now SpaceX has reorganized its data-center construction operation, replacing xAI-led management with engineers from its rocket and satellite-internet divisions and adding reliability measures that could slow the pace of new capacity coming online. The company holds billions in reported monthly compute contracts and is targeting more than 2 gigawatts of AI capacity by the end of 2026.
Speed Built It. Reliability Has to Keep It Running.
Trading construction records for operational resilience, SpaceX is rebuilding how its data-center program works from the inside out.
SpaceX reportedly installed executives and engineers from its rocket and Starlink divisions into roles previously held by xAI-led data-center management, according to The Information. The new priorities include additional backup power systems, redundant cooling, and more comprehensive testing before facilities become operational. Each of those additions increases construction complexity and commissioning time.
Outages at Memphis-area facilities disrupted Grok models and customers renting compute capacity, per Yahoo Finance reporting. That operational record, not any change in ambition, appears to be the trigger for the overhaul. Outages suggest the pace that earned the “superhuman” label did not deliver infrastructure prepared for sustained commercial uptime demands.
Fast and unreliable is not a product; it is a liability.
SpaceX reported 1.4 gigawatts of compute capacity in the second quarter of 2026, up from 1 gigawatt in March, according to USA Today. The company has discussed a longer-term target of up to 20 gigawatts of power and cooling capacity by the end of 2027, a figure that refers to power and cooling rather than compute capacity alone. Reuters reported that at least one large-scale Texas facility is in preparation, though whether the overhauled construction approach can sustain that trajectory remains the central operational question.
Billion-Dollar Contracts Don’t Tolerate Downtime
The reported scale of SpaceX’s commercial compute agreements means that even brief commissioning delays carry significant financial consequences.
Anthropic agreements were reported at approximately $1.25 billion per month, and a separate Google-related arrangement was reported at roughly $920 million per month, per Business Insider. Contract terms and delivery obligations are not fully public. Even modest commissioning delays create meaningful risk against commitments at that scale.
SpaceX placed Kevin Holland, a veteran associated with the Starship program, in charge of bringing up the next phases of compute expansion, per Business Insider. Rocket engineers carry genuine expertise in high-reliability systems and complex hardware integration, both relevant to data-center work. Data-center operations also require specialized knowledge in power procurement, cooling management, networking, and customer uptime obligations. Whether those skills transfer cleanly from launch infrastructure remains an open question.
The overhaul is not a retreat from SpaceX’s expansion ambitions. Watch construction timelines, outage records, GPU commissioning rates, and delivery against contracted compute in the months ahead. Those numbers will determine whether the trade-off between speed and reliability produces the dependable capacity that billions in reported monthly contracts now require.
