Google and SpaceX Launch AI Chips Into Space as Neil deGrasse Tyson Questions the Reality of Orbital Data Centers

Neil deGrasse Tyson Questions Google’s Plan to Put AI Data Centers in Space: “Many Challenges” Remain

Google has officially sent artificial intelligence hardware into orbit, taking one of Silicon Valley’s most futuristic infrastructure ideas out of the concept stage and putting it aboard an actual spacecraft. But astrophysicist Neil deGrasse Tyson says the enormous engineering problem is not getting computers into space. It is figuring out whether operating massive AI data centers there makes any practical or economic sense.

Google’s first orbital prototype for Project Suncatcher launched October 1 aboard SpaceX’s Transporter-18 rideshare mission from Vandenberg Space Force Base in California. Built in partnership with satellite company Planet, the spacecraft carries Google Tensor Processing Units, or TPUs, and is designed to test how the company’s specialized AI hardware survives launch forces, radiation, extreme temperatures and the unforgiving thermal environment of low Earth orbit. Google said after launch that it had established contact with the satellite and that the spacecraft was operating as expected.

Speaking with NewsNation’s Dan Abrams after the launch, Tyson said orbital AI infrastructure is possible under the laws of physics but warned that the proposal faces “many challenges” that become increasingly complicated once engineers move beyond the attractive idea of nearly continuous solar energy.

Google’s Project Suncatcher Is Now in Orbit

Project Suncatcher is Google’s long term research effort to determine whether large scale machine learning infrastructure could eventually operate in space. Rather than constructing enormous terrestrial data centers drawing electricity from already strained regional grids, Google is investigating networks of satellites carrying AI processors and powered directly by solar arrays.

Google says satellites operating in favorable low Earth orbits can access near constant sunlight and potentially generate as much as eight times more solar energy than comparable systems on Earth. The long-term concept is considerably more ambitious than simply putting a server aboard a satellite. Google envisions interconnected groups of spacecraft carrying dozens of TPUs and exchanging enormous quantities of information through high-speed laser links.

The spacecraft launched Thursday is not a functioning orbital hyperscale data center. It is an experimental prototype intended to answer more fundamental questions about whether Google’s AI hardware can reliably operate in space at all. Google plans additional experiments, including two satellites expected in 2027 that will test the precision laser communications required to connect future orbital computing clusters. That distinction is important: Google has demonstrated that it can put AI processors into orbit, but it has not yet demonstrated that thousands of those processors can economically replace or meaningfully supplement the enormous data centers being constructed on Earth.

Tyson Says Space Is Not the Giant Refrigerator People Imagine

One of the most immediate problems, Tyson explained, is heat. The popular image of space as an unimaginably cold environment creates the intuitive assumption that cooling computers in orbit should be easier than cooling them on Earth. In practice, the opposite can be true.

On Earth, data centers can move heat away from processors using air, liquid cooling systems, cooling towers and other techniques that ultimately transfer thermal energy into the surrounding environment. Space offers no atmosphere through which heat can escape by convection. A computer operating in a vacuum still generates heat, but engineers must ultimately reject that heat primarily through radiation. That requires carefully designed radiators, heat pipes and spacecraft orientations capable of dumping thermal energy without simultaneously absorbing excessive heat from the Sun or other components.

Google itself identifies thermal management as one of Project Suncatcher’s central engineering problems. The company’s prototype is specifically testing cooling techniques involving heat pipes and radiators because conventional airflow cooling cannot function in a vacuum.

Tyson also pointed to the extreme temperature differences spacecraft can encounter depending on their exposure to sunlight. The engineering challenge is therefore not simply keeping processors cold. It is maintaining highly sensitive computing hardware within a controlled operating temperature while the spacecraft exists in an environment defined by radiation, vacuum and extreme thermal conditions.

Nearly Unlimited Solar Energy Comes With an Asterisk

The most compelling argument for orbital AI infrastructure is energy. Artificial intelligence requires enormous amounts of electricity, and the rapid construction of data centers has created growing concerns about grid capacity, generation, transmission infrastructure, water consumption and competition with residential and industrial electricity demand.

Space potentially offers an extraordinary alternative: solar panels operating above Earth’s atmosphere, where sunlight is stronger and certain orbital configurations can dramatically increase the amount of time a spacecraft remains illuminated. But Tyson cautioned against interpreting that advantage as meaning every satellite simply receives sunlight 24 hours a day. Ordinary low Earth orbit can repeatedly carry spacecraft through Earth’s shadow, while achieving near continuous solar exposure requires carefully selected orbital configurations.

Google’s proposed architecture accounts for that issue. Project Suncatcher is investigating orbital configurations designed to maximize solar availability, and the company says suitable satellites could produce substantially more solar power than comparable Earth-based systems. That abundant electricity, however, solves only one component of the problem. Engineers still have to launch the hardware, protect it from radiation, cool it, maintain communications between satellites, transmit information back to Earth and eventually determine what happens when equipment fails.

Radiation Is Another Problem for AI Chips

Modern AI accelerators are extraordinarily sophisticated semiconductor devices, and space radiation presents another major challenge. High energy particles can interfere with electronics and cause errors commonly known as bit flips, potentially corrupting calculations or damaging components over time. Before launch, Google subjected its TPUs to radiation testing while the chips were actively running AI workloads, including experiments at the University of California, Davis. Those terrestrial tests were encouraging enough to proceed, but Google says there is no complete substitute for exposing the hardware to the actual orbital environment. The newly launched spacecraft will therefore collect data on how the processors respond to radiation and other stresses over time.

That is one reason Project Suncatcher remains a research program rather than a commercial data center deployment. Google is attempting to identify what fails before attempting anything remotely approaching hyperscale infrastructure.

The Bigger Problem May Be Economics, Not Physics

Tyson’s larger argument is that orbital data centers do not violate any fundamental law of physics. The more difficult question is whether they can ever compete economically with infrastructure on Earth. Every server, processor, radiator, solar panel, communications system and structural component has to be manufactured, integrated into a spacecraft and launched into orbit. Repairs that might require a technician and replacement component at a terrestrial data center become vastly more complicated hundreds of miles above Earth.

Current launch economics make that difficult to ignore. Google’s first experiment was able to take advantage of SpaceX’s Transporter 18 rideshare mission rather than requiring a dedicated rocket. Transporter 18 carried roughly 130 payloads, spreading launch costs across numerous customers. A future orbital AI network operating at anything approaching hyperscale would be a radically different proposition. It could require enormous satellite constellations, massive power generation capacity, sophisticated cooling structures and extremely high bandwidth optical connections between spacecraft.

One analysis cited by The Wall Street Journal estimated that constructing a one gigawatt orbital data center using current technology could cost around $30 billion. Those economics could change dramatically if launch costs continue falling, particularly as reusable heavy lift vehicles mature, but they illustrate how far the industry remains from simply replacing terrestrial data centers with satellites.

Tyson Says Earth Still Has Some Much Simpler Options

Tyson argued that companies trying to prevent AI data centers from overwhelming conventional power grids have alternatives that do not require launching the infrastructure into orbit. One option is to build dedicated energy systems around data centers rather than relying exclusively on existing regional grids.

Large computing facilities can theoretically be colocated with dedicated generation, including hydroelectric resources or advanced nuclear power. Tyson specifically raised small modular reactors as one possible approach. The broader idea is that instead of forcing communities and electrical utilities to absorb enormous new computing loads, technology companies could increasingly develop dedicated power infrastructure alongside their data centers.

That approach has already attracted significant interest across the technology industry as AI companies search for reliable, around the clock electricity without depending entirely on intermittent renewable generation or overloaded transmission networks. The advantage is straightforward: a terrestrial data center powered by dedicated generation still allows engineers to walk through the front door when something breaks.

Google Is Not Alone in Looking Toward Orbit

Project Suncatcher is part of a broader movement toward orbital computing. SpaceX, Starcloud and other companies are exploring various forms of space based computing, while smaller orbital systems are already being designed to process satellite data before transmitting it back to Earth.

India’s TakeMe2Space, for example, also launched an orbital computing satellite aboard Transporter 18 equipped with Nvidia processors and designed to allow customers to run AI models directly in space. The immediate use case for orbital computing may therefore be considerably different from replacing giant terrestrial AI campuses. Processing Earth-observation images, scientific measurements and other satellite generated information directly in orbit could reduce the amount of raw data that must be transmitted back to Earth. Moving the equivalent of an entire hyperscale AI data center into orbit is another matter entirely.

Project Suncatcher Is a Real Experiment, Not Yet a Space Data Center

Google’s achievement this week is significant precisely because the company is beginning to collect real world data rather than relying exclusively on simulations. The prototype satellite is in orbit, Google has established contact with it, and its TPUs will now encounter radiation, thermal extremes and the physical realities of space while engineers monitor what happens.

Whether that experiment eventually leads to enormous constellations of solar powered AI computers circling Earth remains unanswered. Tyson’s skepticism is not that the technology is impossible. His argument is that every apparent advantage of space introduces another engineering tradeoff: solar energy may be abundant, but heat is difficult to remove; land is unnecessary, but rockets are expensive; and space eliminates some terrestrial constraints while introducing radiation, orbital mechanics, communications challenges, maintenance problems and an environment where sending a technician to replace a failed component is anything but routine.

For Google, that is exactly what Project Suncatcher is designed to determine. The company has successfully answered the easiest question: Can we put AI chips in space? Now comes the much harder one: Should we build entire data centers around them?

Sources & Further Reading

Google — Project Suncatcher Prototype Is in Orbit

Google — Behind Project Suncatcher

Google — Original Project Suncatcher Research Announcement

NewsNation — Neil deGrasse Tyson on Space-Based Data Centers

Reuters — Google Plans First Test of AI Chips in Space

Reuters — TakeMe2Space Launches Orbital Computing Satellite

The Wall Street Journal — Can AI Run in Space? Google Is About to Find Out

Axios — Transporter-18 and Orbital Data-Center Experiments