A hypothetical one-gigawatt orbital AI data center would cost roughly $170 billion to build, more than three times what an equivalent facility on the ground would run, according to a new analysis from energy analytics firm Wood Mackenzie examining SpaceX's just-announced Starmind AI1 satellite program. The effort, developed in partnership with NVIDIA, envisions deploying as many as one million 30-meter-tall satellites equipped with the chipmaker's latest Vera CPUs and Rubin GPUs into low Earth orbit roughly 600 kilometers above the planet. Each spacecraft would draw up to 250 kilowatts at peak load and 175 kilowatts on average, powered entirely by solar arrays spanning 75 meters, and communicate via Starlink's laser link network.
Launch and satellite manufacturing would account for approximately 60% of the $170 billion price tag, the analysis finds. Each AI1 satellite is estimated to weigh 2.3 metric tons and would require SpaceX's still-under-development Starship rocket for deployment. To reach the goal of one million satellites, the two companies plan to build a standard model spacecraft at SpaceX's 11-million-square-foot Gigasat Factory, currently under construction in Bastrop County, Texas. Bringing orbital costs down to match terrestrial data centers would demand a 70% reduction in expenses, the report notes.
The technical obstacles are severe, the report warns. Heat rejection presents the largest challenge: a system running hundreds of kilowatts of AI hardware must expel nearly all of that power as waste heat through infrared radiation, a very slow process in the vacuum of space. Each Starmind satellite will carry a deployable liquid radiator system measuring 160 square meters—likely using ammonia, already employed on the International Space Station—though whether this will reliably scale to data-center-class AI deployments with their enormous heat loads remains uncertain. Networking poses another barrier: while Starlink's mini laser terminals operate at up to 25 gigabits per second across distances as long as 4,000 kilometers with roughly 25-millisecond latency, those figures don't establish that a satellite constellation can function like the tightly coupled networking fabric of a terrestrial AI supercomputer, which requires huge, predictable bandwidth and very low latency for GPU-to-GPU communications. The orbital network would also face physical propagation delays, laser-link acquisition and handoffs, routing across a moving constellation, and limits on available capacity per spacecraft.
Robert Liew, Wood Mackenzie's research director, observes that "the gap does not close without sustained and dramatic progress on launch costs," adding that the firm forecasts $9 trillion in terrestrial data center investment between now and 2040. The report concludes orbital data centers are a serious long-term proposition, but right now they remain a bet on the cost curve, with capital flowing first to ground-based facilities. SpaceX has offered a broad technical vision and a hardware partnership with NVIDIA, but few of the operational metrics that would establish commercial viability, the analysis states. The real test will be whether SpaceX Starship becomes a practical launch vehicle and can overcome its cooling and safety issues, then whether the AI1 can show enough usable compute per kilogram, kilowatt, square meter of radiator, and dollar of launch cost to outperform or complement ground-based AI infrastructure. For companies weighing where to place their next wave of computing capacity, the choice between proven terrestrial infrastructure and unproven orbital systems will ultimately hinge on whether engineering breakthroughs can overcome economics that currently favor staying grounded.