An AI is leaving the planet
An artificial intelligence is leaving the planet. On 1 October, a SpaceX Falcon 9 rocket will carry a single, small satellite into low Earth orbit. The payload belongs to Google. It contains the company’s Tensor Processing Units, the specialised processors that run its AI. The mission has a name. Project Suncatcher.
Google’s public explanation for this launch is straightforward. The company wants to test its hardware against the extremes of space. It is a harsh place for electronics. The goal, according to Google, is to measure how its chips cope with the physical shock of spaceflight, the constant bath of radiation and the severe thermal changes found just a few hundred kilometres above the Earth’s surface.
This is a plausible story. Space is unforgiving. But it is not the complete picture. The experiment seems almost trivial for a corporation of Google’s scale, a firm that routinely operates at the very edge of computing and spends billions on research every quarter. Sending a few chips into the void is an extraordinarily expensive way to check for radiation damage. There must be another reason.
The real prize is hinted at in the project’s name and its internal purpose. Project Suncatcher. The name suggests energy. This is not just a scientific test of silicon fortitude against cosmic rays. It is the beginning of something much larger, a quiet first move in a strategy that could eventually see entire data centres built in orbit, floating server farms powered by the one resource space has in abundance, constant, unfiltered sunlight.
That ambition is enormous. It changes everything. An orbital data centre exists beyond borders, potentially outside the jurisdiction of terrestrial laws which govern how personal and state information must be handled. Who has legal authority over data that never touches the ground? The 1 October mission is only one satellite. A proof of concept. But the questions it raises are not about physics. They are about power, security and control. They are about who owns the future.
What is a Tensor Processing Unit?
A Tensor Processing Unit is not the chip inside your laptop. It is a specialist. It does one thing. A TPU is an application specific integrated circuit, a sliver of silicon custom built by Google to accelerate the mathematical operations at the heart of machine learning programmes. Think of it not as a brain but as a muscle, designed for the repetitive heavy lifting that allows an AI to recognise a face in a photo or translate a sentence from one language to another. These chips make AI calculations faster and far more energy efficient than a standard processor ever could.
Putting one in space is difficult. The environment is hostile. Low Earth orbit is not empty, but is instead flooded with high energy particles from the sun and from deep space. This radiation is a constant threat to electronics. A single particle striking a processor in the right place can flip a bit, changing a zero to a one, corrupting data or causing the entire system to crash. Then there is the temperature. A satellite swings from extreme heat in direct sunlight to extreme cold in the Earth’s shadow, all while sitting in a vacuum that makes dissipating heat exceptionally difficult.
The traditional solution is expensive. It is slow. Space agencies have historically used ‘radiation hardened’ components, or rad hardened for short. These are processors designed from the ground up to withstand the space environment, often using older, more robust manufacturing techniques and containing layers of physical shielding. A single rad hardened chip can cost hundreds of thousands of pounds yet offer the performance of a decade old commercial processor. Building a data centre from such components would be financially impossible.
This is where Google’s experiment becomes interesting. The company is not launching a rad hardened chip. Its satellite contains a ‘lightly modified’ commercial TPU. This is the core of Project Suncatcher. The gamble is whether mass produced commercial hardware, with only what Google calls ‘light modifications’, can do a job that has historically required components costing hundreds of times more. If Google can prove that cheap, powerful, off the shelf hardware can survive the void, the economics of computing in space are rewritten completely.
The search for free energy
The commercial logic behind Project Suncatcher is simple. It is about money. Artificial intelligence is phenomenally expensive. The biggest line item is not the hardware or the researchers, but the raw electrical power needed to make the algorithms think. AI consumes energy on a vast scale. A single data centre dedicated to training large models can draw as much power as a small city, placing new and unwelcome demands on terrestrial electricity grids. Companies are now building AI infrastructure where the electricity is cheapest, not where the engineers are. This is a problem.
A huge fraction of that power is simply wasted. It becomes heat. Processors running complex mathematical operations become hot, and thousands of them packed into racks in a server farm become incredibly hot. So operators spend fortunes on cooling. They build entire secondary infrastructures of industrial scale chillers, pumps and vast fans just to blast the heat away from the silicon and stop the servers from melting. This is a constant battle. It is a battle against physics. The cooling systems themselves consume enormous amounts of energy, adding significantly to the overall running cost and the environmental footprint of the facility.
Space offers a fix. An elegant one. An orbital data centre would have access to two resources that are free and abundant beyond the atmosphere. The first is solar power. A satellite equipped with solar panels can draw on a constant stream of energy from the sun, freeing it from any reliance on a terrestrial grid. No more bills. No more negotiating with utility companies. The second resource is the cold. The near perfect vacuum of space is an astonishingly effective place to get rid of heat. An orbital server could be designed to radiate its waste energy directly into the void, a cooling system with no moving parts, requiring no power whatsoever. It is the perfect heat sink.
This is the real prize. This is the ultimate goal of Project Suncatcher. The long term vision is a data centre powered by the sun and cooled by the vacuum, a facility that could operate with running costs an order of magnitude lower than anything achievable on the ground. The economic advantage would be immense. It would rewrite the financial model of the entire industry. The 1 October launch is a small step. It is a feasibility study. But if Google’s lightly modified chip can function in the radiation and thermal extremes of orbit, it proves the most basic component of this vision is viable. It unlocks the business case for taking AI off the planet.
A server beyond government
The physics is one thing. The law is another. A data centre on the ground is bound by geography, meaning it is bound by the laws of the nation in which it sits. European regulators can enforce the General Data Protection Regulation, or GDPR, on a server farm in Dublin precisely because it is in Dublin. But who has legal authority over data that is generated, processed and stored on a satellite that never touches the ground? The existing international agreements, such as the Outer Space Treaty from 1967, were conceived to prevent nations from placing nuclear weapons in orbit, not to govern the intricacies of cloud computing and data sovereignty. They are completely silent on the issue. A legal vacuum.
The benefits are not just about avoiding regulators. The security implications are enormous. Currently, spy satellites collect colossal volumes of imagery and signals intelligence, raw data which must then be transmitted down to ground stations for processing and analysis. This download is a bottleneck. It is slow. It can be intercepted. An artificial intelligence in space changes everything. Google’s TPUs could sift through hours of satellite video in real time, identifying military hardware, tracking ship movements or detecting the preparations for a missile launch as they happen. Only the crucial result, the final piece of intelligence, would need to be sent to Earth, a tiny and easily encrypted file arriving almost instantly. This would grant an extraordinary advantage to any military or intelligence agency that controlled it, shortening the loop between observation and action from hours or days to mere seconds. The tactical edge is obvious. It is immense.
Such a server is also physically secure in a way that no terrestrial facility can ever be. There is no perimeter fence to breach. You cannot cut the power cable, you cannot physically seize the hard drives and you cannot serve a warrant to a machine travelling at five miles a second. This combination of legal ambiguity and physical remoteness creates a powerful new kind of digital haven, a server beyond government. It is a move towards a form of corporate sovereignty. The organisations building this infrastructure could establish a domain for data that exists entirely outside the established rules that govern nations on Earth. They would become the gatekeepers to this new realm, setting the terms of access and use far from any parliament or court. A private jurisdiction. Written in the stars.
A small step for an algorithm
This launch is just one small step. A single satellite. A simple test. Google calls Project Suncatcher a proof of concept, a method for checking how its commercial hardware performs when exposed to the harshness of low Earth orbit. But the company is not alone in its ambition. It is not even the first. The race to put the cloud into space has been running for some time, and Google's rivals are already established in a market that could one day define the entire global technology sector. This experiment, though modest, is a clear signal that the search giant does not intend to be left behind on the ground. It is playing catch up.
The competition is formidable. Microsoft has invested heavily in its Azure Space initiative, which aims to connect its vast terrestrial cloud network to hardware in orbit, creating what the company calls a 'seamless' computing fabric for the space industry. Amazon Web Services, the dominant player in cloud computing on Earth, also has a dedicated Aerospace and Satellite division working to extend its services beyond the atmosphere. These are not speculative ventures. They are core strategic priorities for two of the world's most powerful corporations, both of whom recognise that the next frontier for data is not on Earth at all. They see a future where satellite ground stations are replaced by orbital data relays, and information is processed at the source, thousands of miles above the planet. Google’s Suncatcher is a late, but significant, entry into this contest. A toe in the water.
This is not simply about market share. It is about power. The corporate struggle is only one part of a much larger contest playing out between the United States and China. Both nations see dominance in space as critical to their future economic prosperity and national security. The country that masters space based computing will not just gain a military edge, it will control the essential infrastructure of the twenty first century economy, setting the standards and controlling the flow of data for a new era. Chinese companies are also pursuing their own ambitious space programmes, backed by the state, with the explicit goal of challenging American supremacy in this domain. The Falcon 9 carrying Google's chips is therefore more than a commercial rocket. It is a move on a global chessboard.
What happens after the launch
The launch on 1 October is a beginning. It is not an end. Once the satellite reaches its low Earth orbit, the real experiment begins, far from any press conference or public announcement. Google’s engineers will spend the coming months monitoring their Tensor Processing Units, searching for the tiny digital scars left by high energy particles and the performance drops caused by extreme temperatures. The company has been clear about its objective. It wants to measure how the chips handle space.
What it has not shared are its metrics for success. Nobody outside the project knows the acceptable error rate for a processor struck by cosmic radiation, nor the thermal tolerance the company expects from its lightly modified hardware. That information is a prize. It is the key to the entire enterprise. Do not expect a detailed public report. The results will almost certainly remain a closely guarded corporate secret, a data point worth billions of pounds if the concept proves viable. The real signal of success will not be found in a Google blog post. It will be found on a launch schedule. The true sign of victory for Project Suncatcher will be the quiet commissioning of a second, more ambitious mission.
A server in the sky is still a distant prospect. The journey from this single proof of concept to a fully functioning orbital data centre, powered by constant sunlight and cooled by the vacuum of space, is immensely long. It is filled with engineering problems that have not yet been solved. This is a small step. But it is a step. An orbital data centre sounds like science fiction. Yet a rocket will soon carry the core components of an artificial intelligence into space to see if it can live there. This is not a simulation. A physical object is moving from a cleanroom to the void above the planet. The age of the orbital cloud may not be here, but this week, we can watch it begin.
Sources. The Register: Google's TPUs to catch some rays in orbit next week. The Verge: Google is sending an AI satellite into space next week.
Analysis. Drafted with AI assistance from the sources listed above and reviewed by an editor before publication. Jnews links to the organisations it writes about.
