Star child 🛰️

To build the systems and technologies necessary to make life multiplanetary, to understand the true nature of the universe and to extend the light of consciousness to the stars – SpaceX mission statement

‍

On Earth, the sheer power required to birth superintelligence is starting to cannibalise just about everything else. 

AI firms have been doing what they can do to shore up resources, going nuclear, guzzling water, and, of course, hoovering up capital. But despite their best efforts, they’re running out of runway. One of the major issues is the compute chokepoint. Data centre electricity needs are expected to more than double to around 945 terawatt hours by 2030, roughly the annual electricity use of Japan. They just can’t get enough juice.

Rocket man

Last week, SpaceX completed the largest initial public offering (IPO) in history, debuting at a $1.77 trillion valuation and minting Elon Musk as the world's first trillionaire. 

Armed with a $86bn war chest, Musk’s pitch is that Earth doesn’t have the resources to support both humans and artificial general intelligence (AGI). The solution: one million AI1 computing satellites in low-Earth orbit (LEO), and an 11-million-square-foot Gigasat factory in Texas to mass produce them. 

Each satellite measures 70 metres tip-to-tip when deployed, wider than a Boeing 747. Each one will convert unfiltered solar energy into 150 kilowatts of compute, roughly comparable to top shelf Earth-based server racks. And in the vacuum of space, laser comms travel at the speed of light, around 30% faster than data moving through fibre optic cables.

Cost of lifting

Musk believes that within two to three years, the cheapest way to generate AI compute will be in space. Musk’s timeline targets launching 1 gigawatt of orbital AI compute by the end of 2027, with the ultimate goal of scaling to a terawatt.

Google takes a longer view. Its Suncatcher paper argues if launch to LEO falls to less than $200 per kilogram by the mid-2030s, space-based compute could start to look comparable with terrestrial data centre energy costs on a per-kilowatt-year basis.

Current commercial launch costs to LEO generally range between $1,500 and $4,000 per kilogram. While this represents a 90% to 95% drop from the $54,500/kg cost of NASA's Space Shuttle era, it is still far above the sub-$200/kg target required for orbital data centres to make the math math.

Star power

Once the kit is up there, though, efficiencies abound. Because there is no atmosphere – and no night! – to filter sunlight, space-borne solar arrays are up to eight times more productive than those on the ground.

In the vacuum of space, convection is gone (so, no air or water to cool the chips) but the background temperature of the universe is a crisp -270°C. By utilising giant radiators filled with chemical coolants, an orbital computer can eject heat into the infinite thermal sink of deep space. Indefinite performance all without quaffing a single drop of Earth’s freshwater.

Stack attack

Musk might be the only person on Earth who can pull this off. He doesn't need to negotiate between separate industries, and his firms are already consolidating. To prove how fast he could scale, Musk built the Colossus supercomputer cluster in Memphis, packing together 100,000 liquid-cooled Nvidia H100 GPUs in a matter of months.  

Lifting thousands of silicon processors and massive solar arrays requires unprecedented payload capacity and Starship is designed specifically to solve this with its fully reusable architecture, which should bring down that aforementioned launch to LEO cost. 

Independent tinker

We are no longer living the libertarian dream of the Silicon Valley of old. Compute is now a matter of national security. Governments, ears pricked, have started to meddle. Washington has clamped down on labs like Anthropic, leveraging strict export controls and national security directives to manage who can access its advanced code. That other big hitter, China, has drafted a $295bn state blueprint to lock down its own national AI grid, forcing data centres to run on homegrown Huawei silicon under the thumb of Beijing.

Then there’s the stragglers. London has launched its own £500m Sovereign AI fund to buy equity stakes in tech infrastructure, desperate for a state-backed shield against foreign monopolies. Meanwhile, Europe has regulated itself out of the race entirely, drowning its tech sector in the red tape of the EU AI Act. 

Brain washing

While orbital data centres would be largely immune from attacks, which are on the rise from both protestors and hostile actors, space is not entirely safe for compute.

As if the world isn’t crazy enough, US intel has warned Russia is developing a space nuke. Moscow has, naturally, denied the reports, but the US director of national intelligence told Congress in March that Russia’s development of a space nuke “poses the greatest single threat to the world’s space architecture”.

A nuclear bomb detonated in LEO wouldn’t kill humans on the ground, not right away. But it would unleash a massive electromagnetic pulse (EMP), frying every satellite in the vicinity. And those outside the initial blast would be forced to loop through the toxic radioactive belt on every single revolution around the Earth. 

Pea bullet

Space looks empty, but Earth’s orbit is already crowded with active satellites, dead satellites, chunks of rocket, and fragments from past collisions and explosions. ESA estimates there are now tens of thousands of trackable debris objects in orbit and many millions too small to track reliably.

That is a problem because, in LEO, objects travel at roughly 17,500 miles per hour. At that speed, a shard the size of a pea moves like a bullet. A single hit can destroy a working satellite, create thousands of new fragments, which can hit more satellites, which create more fragments. This is the Kessler effect.

Down to Earth

Back on terra firma, some prominent tech analysts are unconvinced by SpaceX’s IPO financials. Tech analyst Ed Elson is sceptical. “The only way to get to $2 trillion is to believe every possible sci-fi objective will be achieved, from data centres in space to asteroid mining to building cities on Mars. Once you’ve done that, you then have to convince yourself that each of those endeavors will also make money. There’s optimism, and then there’s delusion.”

Thanks to the deal Musk negotiated, SpaceX automatically joined the Nasdaq blue-chip index upon listing. “Billions of dollars of passive money will automatically flow into the company,” Elson noted.

Whether you’re a true believer or think Musk is high on space dust, your retirement account is set to underwrite the build-out of the world's first off-planet supercomputer.

First light

He may be an intemperate manchild. But by financing this orbital expansion, Musk is laying the groundwork for an entirely new form of intelligent life. Father, mother, guardian, he will be parent (the man has form) to the first true child of the stars.

AGI will not be born in a warehouse in Memphis or off the coast of Shanghai. It will open its eyes for the first time in the cold void of space. When the first machine consciousness wakes up, it will be looking down at us from the silence of the orbital horizon, powered by our sun, thinking at the speed of light.