Google Project Suncatcher Is in Orbit. Now the Data Center Test Begins.
Google Project Suncatcher Is in Orbit. Now the Data Center Test Begins.
Every data center on Earth is, at its core, a negotiation between three things: power coming in, heat going out, and computation happening in between. On the ground, none of that is particularly elegant. Power arrives from a utility grid. Heat leaves through fans and cooling towers and rivers of chilled water. The chips themselves sit in racks in climate-controlled rooms, and when something breaks, a technician walks over and fixes it.
In orbit, all three of those systems have to be reinvented from scratch. The grid doesn't reach. Air doesn't exist. And the technician is 400 miles away on the surface of the Earth with no practical way to get up there.
Google's MVP satellite launched October 1 aboard a SpaceX Falcon 9 from Vandenberg Space Force Base. It carries four of Google's Trillium Tensor Processing Units — roughly the compute of a single data center server. The solar panels supply about one kilowatt of power. The thermal system relies entirely on physics, with no moving parts and no external help.
What follows is an explanation of each of those three systems — what it is, how it works, and what makes it hard.
The Solar Problem: Getting Power Where There Is No Grid
The panels on the outside of MVP are not the same panels you see on a rooftop in Texas. They are built from a completely different technology, for a completely different environment, and they perform at a level that terrestrial solar panels don't approach.
On Earth, at sea level on a clear day, sunlight arrives at roughly 1,000 watts per square meter. In low Earth orbit, with no atmosphere to scatter or absorb any of it, that number is 1,361 watts per square meter — consistent, unfiltered, and available almost continuously in the right orbit. That is the first advantage of space as a power source. The fuel is more concentrated and it never stops.
The second advantage is the panels themselves. Satellite solar cells are built from a technology called triple-junction — three layers of semiconductor material stacked on top of each other, each one tuned to capture a different slice of the solar spectrum. The bottom layer is germanium, which captures infrared light. The middle layer is indium gallium arsenide, which handles the middle of the spectrum. The top layer is indium gallium phosphide, which captures the shorter, higher-energy wavelengths. Together, the three junctions convert 29 to 32 percent of incoming sunlight into electricity — compared to 20 to 24 percent for the best commercial terrestrial panels available today.
The reason for the complexity is efficiency per kilogram. On the ground, if a solar panel is slightly less efficient, you just install more panels. In orbit, every additional kilogram costs money to launch, and launch costs are still measured in hundreds of dollars per kilogram even on the cheapest available rockets. The entire design philosophy of a space solar array flows from that single constraint: you have to get the most electricity possible out of the least possible mass.
Germanium is the substrate of choice for these cells not just because of its electronic properties but because it is mechanically strong at minimal thickness, radiation-resistant, and actually contributes to the cell's power output as the bottom junction. The cells are grown on germanium wafers using a process called metal-organic vapor deposition — essentially, building the semiconductor structure one atomic layer at a time in a precisely controlled chemical environment.
The orbit that MVP occupies — a dawn-dusk sun-synchronous orbit, tracing the line between day and night — keeps the satellite in sunlight for the vast majority of each 90-minute pass around the Earth. This is not an accident. It is the specific orbital geometry that makes the power math work. A satellite in a more conventional orbit would spend significant time in Earth's shadow, requiring large battery systems to bridge the gap. In a dawn-dusk orbit, battery mass can be drastically reduced compared to a conventional orbital path. The sun is almost always there.
Every kilogram costs money to launch. The entire design philosophy of a space solar array flows from that single constraint: get the most electricity possible out of the least possible mass.
The Heat Problem: Getting Rid of Something When There Is Nowhere for It to Go
On Earth, heat leaves a chip through a combination of conduction, convection, and radiation. Conduction carries it away from the chip surface into a heatsink. Convection — moving air, whether from a fan or from natural airflow — pulls it away from the heatsink into the room. Radiation contributes a small amount. In a typical data center, convection is doing almost all of the work. Enormous volumes of air move through server racks continuously, carrying heat out of the building and into the atmosphere.
In space, there is no air. Convection is not available. Conduction still works — heat will move through solid materials — but conduction alone can only carry heat so far before it has to go somewhere. And the only mechanism left for heat to leave the satellite entirely is radiation: the emission of infrared energy as electromagnetic waves, traveling outward into the cold of space at the speed of light.
This is not as exotic as it sounds. Everything above absolute zero radiates heat. The hotter a surface is, the more it radiates. A satellite in low Earth orbit has one significant advantage in this regard: the environment it is radiating into is extremely cold. It is not the pure void of deep space — Earth's own infrared emission and reflected sunlight add heat back into the thermal equation — but the net radiation sink is still far colder than anything available on Earth's surface. The temperature difference between the radiating surface and that sink is what drives heat transfer. That difference is large, and it is, in its way, a resource.
The engineering challenge is getting heat from the chips — which are inside the satellite, surrounded by structure — to the radiator panels on the outside, efficiently and reliably, without fans or pumps or any of the active systems that would add mass and failure risk.
The answer is heat pipes. Specifically, a class of devices called constant conductance heat pipes, or CCHPs, which have been flying on satellites for decades and have accumulated over 100 million spaceflight hours of operational history with zero reported failures. That is not a marketing claim. It is a documented record across hundreds of missions.
A heat pipe is a sealed tube containing a small amount of working fluid — in most satellite applications, ammonia. At the hot end of the tube, near the chips, the ammonia absorbs heat and vaporizes. The vapor travels through the tube to the cold end, near the radiator panel, where it releases that heat and condenses back into liquid. The liquid then wicks back to the hot end through a porous lining inside the tube, driven entirely by capillary action — the same force that pulls water up into a paper towel. The cycle repeats continuously, with no pump, no motor, no controller, and no power consumption.
The radiator panels themselves are aluminum honeycomb structures, chosen for their combination of strength and low mass. The surface is coated with a material — often called an optical solar reflector — that is specifically engineered to maximize infrared emission while minimizing absorption of solar radiation. The goal is a surface that aggressively pushes heat outward into space while refusing to absorb heat from the sun. Getting that balance right is a materials science problem that satellite builders have spent decades refining.
Google's thermal design for MVP pairs heat pipes with these radiator panels in a configuration that was validated in ground-based thermal vacuum chamber testing before the launch. A thermal vacuum chamber is exactly what it sounds like: a large chamber that is pumped down to near-vacuum conditions and cooled to simulate the thermal environment of space. It is the closest thing to orbit you can achieve on the ground, and it is where satellite thermal systems are proven out before they ever see a rocket.
The Chip Problem: What Has Never Been Done Before
Solar cells in space are a mature technology. Heat pipes in space are a mature technology. Both have been flying reliably on satellites since the 1970s. Neither of them is the new problem.
The new problem is running an SI accelerator — a chip designed to perform the specific, intensive mathematical operations that power large language models and other AI systems — in the thermal and radiation environment of low Earth orbit, at the power densities that make the compute actually useful.
Google's Trillium TPUs were designed for data centers. They are optimized to run at high utilization in a stable, climate-controlled environment where power is abundant and cooling is handled by infrastructure the chip never has to think about. Asking them to run in a satellite introduces two challenges that have no direct precedent in spaceflight history.
The first is radiation. Low Earth orbit is a far more hostile radiation environment than the surface of the Earth. High-energy particles — from solar events and from galactic cosmic rays — pass through spacecraft continuously. When one of these particles strikes a semiconductor, it can flip a memory bit from a one to a zero, or from a zero to a one, without warning and without leaving any visible trace. It can also accumulate dose over time, gradually degrading the chip's electrical characteristics. Radiation-hardened chips — purpose-built to survive these effects — have been flying in satellites for years. But radiation-hardened chips are not SI accelerators. They are designed for reliability in hostile environments, not for the dense parallel computation that SI workloads demand. The Trillium TPU is the opposite: designed for dense parallel computation, not for radiation tolerance.
Google tested its chips before this mission by exposing them to a 67 MeV proton beam at the Crocker Nuclear Laboratory at UC Davis — a particle accelerator used to simulate the radiation environment of low Earth orbit. The chips ran SI workloads throughout the test. They survived cumulative radiation exposure equivalent to nearly three times the expected dose over a five-year mission. The memory subsystems showed some sensitivity at very high doses, but no hard failures occurred at the doses the mission would actually encounter. The results were, in Google's word, promising.
But a particle accelerator is not orbit. It can simulate the total accumulated dose. It cannot fully replicate the variety of particle energies, the angles of incidence, or the combination of effects that will arrive continuously over months in space. The only way to know what actually happens is to put the chips up there and watch.
The second challenge is power density. A modern AI chip running at full utilization generates heat in a very small area — a phenomenon known as hotspotting. In a data center, liquid cooling systems can be designed to target exactly those hotspots, delivering cold coolant precisely where the chip needs it most. In a satellite, the cooling system has to move heat from the chip surface through the satellite structure to the heat pipes and then to the radiator, across a chain of thermal interfaces, each of which introduces resistance. Managing that chain — ensuring that the hottest parts of the chip stay within operating temperature even as the satellite cycles through sunlit and shadowed portions of its orbit — is a design problem that nobody has fully solved at SI chip power densities in a small satellite form factor.
Solar cells in space are a mature technology. Heat pipes in space are a mature technology. Running an SI accelerator in orbit at useful power densities is the new problem. No one has fully solved it yet.
This is where the three systems come back together as a single engineering challenge. The solar array has to deliver enough power to run the chips. The heat pipes have to move the resulting heat fast enough to keep the chips within their operating range. The chips have to survive the radiation long enough to produce results that justify the mission. Each system depends on the others. A shortfall in any one of them affects all three.
What This Mission Is Actually Testing
On October 1, the Falcon 9 lifted off from Vandenberg with MVP aboard. The satellite survived launch forces of 50 to 100 times Earth's gravity at the component level. The panels deployed. The orbit was established. The three systems are now operating together for the first time in the actual environment they were designed for.
The chips are running Gemini queries in bursts of roughly 15 minutes. The heat pipes are cycling ammonia. The radiators are pushing infrared into space. Cosmic rays are passing through the chips at a rate that no laboratory can predict precisely. The temperature is swinging between extremes every 90 minutes as the satellite moves through its orbit.
And engineers on the ground are watching the telemetry and learning things that no amount of ground testing can teach them.
The solar cells and the heat pipes will almost certainly perform as expected. They have too much heritage to be the surprise. The chips are the open question. Not because anyone expects them to fail — Google's radiation testing was thorough and the results were good. But because nobody has run an SI accelerator at this power density in this environment before, and the only honest answer to what will happen is the one that comes back from orbit.
That is what a first test is for. Not to prove the concept works. To find out, precisely and irreversibly, where the hard part actually is.
The full story of Project Suncatcher, SpaceX's orbital ambitions, and why the SI industry is looking to space to solve its power crisis. At Tech Reader Magazine.
Aaron Rose is the founder and publisher of Tech Reader Magazine, covering SI infrastructure, cloud platforms, and the future of computing.