SpaceX's confirmation that orbital AI data centers would launch in 2027 has drawn warnings from two technical communities. Astronomers warn the constellation would interfere catastrophically with ground-based observations. Grid regulators point to electrical engineering challenges that terrestrial data centers already create—challenges orbital facilities cannot escape if they draw power from Earth-based grids during development or for ground infrastructure.
SpaceX president Gwynne Shotwell told CNBC on June 12 that the company expects to launch its first AI1 satellites in late 2027, with earlier test payloads flying on Starlink broadband and mobile satellites. The AI1 design stretches 70 meters long and 20 meters tall when deployed, carrying solar arrays that generate up to 150 kilowatts of peak power to support an average 120-kilowatt computing payload. Ian Dahl, SpaceX's director of satellite engineering, framed the core challenge as delivering power and rejecting waste heat—simpler, Elon Musk argued, than building Starlink's complex phased-array antennas.
The company has filed with the Federal Communications Commission for authorization to deploy up to one million satellites. SpaceX claims AI satellites will be easier to manufacture than Starlink units because they rely on laser intersatellite links rather than phased arrays. Production will center on a new facility in Bastrop, Texas, where solar array manufacturing has already begun. Musk said the company expects AI satellite production and solar fabrication to reach reasonable volume by the end of 2027.
Astronomers immediately raised alarm. At a June 4 National Academies meeting, Tony Tyson—distinguished research professor at UC Davis and chief scientist for the Vera C. Rubin Observatory—warned that orbital data centers would compound interference problems already created by broadband constellations. Despite mitigation efforts, many current Starlink V2 Mini satellites shine at fifth magnitude, well above the recommended seventh-magnitude limit for protecting sensitive telescopes. The larger V3 Starlink satellites will be brighter still. AI satellites would dwarf them both in size and reflectivity.
Tyson warned that at a million satellites, continuous 'bright lanes' would exist in low Earth orbit due to the high launch cadence required to build out and replenish the constellation. The AI satellites would first be launched into low parking orbits for checkout before raising to operational altitude. The satellites are extremely bright while being tested in those lower orbits, creating lanes that ground-based observers cannot avoid.
Even at operational altitude, the spacecraft would create glints as bright as Venus—averaging magnitude zero—that interfere with time-domain astronomy monitoring supernovas, gamma-ray bursts, and other transient events. In a preprint paper, astronomer Geoffrey Marcy calculated that a 4-by-4-kilometer solar array at 500 kilometers altitude would span roughly 0.4 degrees on the sky, comparable to the Moon's angular size, and shine between magnitude negative-five and negative-seven—about 100 times brighter than the brightest stars. Dozens of such structures would appear as a north-south chain visible for roughly 90 minutes after sunset and before sunrise, blocking stars and deep-sky objects for minutes at a time.
Tyson expressed pessimism about both technical and policy solutions, calling orbital data centers a failed business model unable to compete with terrestrial facilities powered by renewable energy. He warned that major impacts on optical astronomy would begin in 2027, leaving the next generation of astronomers to cope with a degraded observational environment. The Rubin Observatory, equipped with an 8.4-meter mirror and 3.2-gigapixel camera designed to conduct the Legacy Survey of Space and Time, is particularly vulnerable—its wide field of view and sensitivity to faint objects make it highly susceptible to moving light pollution that masks pixels, triggers false detections, and contaminates statistical measurements of dark matter and cosmic expansion.
Parallel warnings emerged from grid regulators about a different failure mode: the electrical engineering challenges posed by massive computational loads. The North American Electric Reliability Corporation issued what the regulatory world characterized as a 'level three' warning about AI data centers threatening grid stability. Colin McCormick and Doug Bryan of Carbon Direct explained on the Volts podcast that computational loads behave fundamentally differently from traditional industrial facilities. Data centers interface with the grid through power electronics and uninterruptible power supplies designed to protect expensive computing equipment by tripping offline within 20 milliseconds if they detect minor voltage or frequency fluctuations.
When a 100-megawatt data center suddenly disconnects, it creates an abrupt change that grid operators must compensate for instantly to maintain frequency regulation at 60 hertz. While grids are designed to handle some load loss, the proliferation of gigawatt-scale computational facilities—each capable of dropping hundreds of megawatts in a blink—introduces risks of cascading failures and widespread blackouts. McCormick noted that unlike rotating machinery in traditional industrial plants, power electronics actively sense grid conditions and respond protectively, making them simultaneously more sensitive and more disruptive. The concern centers on loads large enough and unpredictable enough to destabilize frequency before backup generation can respond.
Natural gas turbines whose shafts are designed to last seven years are failing in 10 months due to excessive cycling.
Jigar Shah, former head of the Department of Energy's Loan Programs Office, told Volts that data centers running AI training have been operating in violation of grid standards, breaking rules established in the 1950s that prohibit loads from ramping between 20 and 80 percent loading five times per minute. The load volatility is wrecking equipment: natural gas turbines whose shafts are designed to last seven years are failing in 10 months due to excessive cycling. Shah argued that the rush to build behind-the-meter gas plants is a fairy tale concocted on the back of a napkin—bad for emissions, bad for economics, and bad from the data center's own perspective because it introduces equipment failure and schedule risk.
The grid stability problem creates an unexpected constraint on orbital data center development. While space-based facilities would eventually operate independently on solar power, the ground infrastructure supporting them—laser communication terminals, mission control, data uplinks and downlinks—would still draw from terrestrial grids. Any plan to scale production in Bastrop or elsewhere would require massive manufacturing facilities whose power demands exhibit the same volatility characteristics NERC is warning about. Casey Handmer's technical analysis argued that pure solar-plus-battery systems could eventually bypass gas turbines entirely, but noted that turbine manufacturing constraints are already limiting AI buildout on Earth, and orbital facilities cannot accelerate faster than the ground infrastructure that supports them.
SpaceX's IPO valuation reportedly rides heavily on orbital data centers proving economically viable. The company's S-1 filing acknowledged that the initiatives involve significant technical complexity and unproven technologies that may not achieve commercial viability—language that echoes the optimistic autonomous vehicle predictions in Lyft's 2019 prospectus, two years before that company sold its Level 5 engineering division to Toyota. The orbital data center concept faces questions about profitable unit economics given launch costs, hardware replacement cycles, and latency penalties. Recent developments add two concrete risks: astronomical interference that could trigger regulatory limits on deployment, and electrical engineering challenges that constrain the terrestrial infrastructure required to build, launch, and operate the constellation.
The parallel to terrestrial AI compute buildout is striking. Just as data center developers are discovering that gas turbine supply chains cannot support unlimited growth—forcing a shift toward solar, batteries, and grid-interactive software—orbital data center plans are colliding with physical limits on how much infrastructure low Earth orbit can absorb before rendering ground-based astronomy unusable. The International Telecommunication Union's WRC-27 treaty summit looms as a potential regulatory checkpoint where astronomers will push for binding limits on orbital carrying capacity and spacecraft reflectivity. Whether those limits arrive before or after SpaceX's late-2027 launch target may determine whether the concept scales to millions of satellites or stalls in the thousands, stranded between technical feasibility and operational viability.