Finding Space.

This one was a bit more involved and so I decided to break it up over a month. We’ll see I may end up doing more of these in the future. I am still feeling this whole content creation thing out but this is the current plan.

“In the beginning the Universe was created. This has made a lot of people very angry and been widely regarded as a bad move.” - Douglas Adams, The Restaurant at the End of the Universe (1980)

Seemingly dominating the global zeitgeist, the next technological revolution is underway. Silicon intelligence is becoming more and more widespread and questions around AI are becoming increasingly pressing.

At the heart of this equally terrifying and impressive breakthrough are data centers. With estimated global spending at nearly $7 trillion over the next four years to keep up with demand for AI, the data center sector is huge. For context, the combined market cap of Microsoft ($MSFT ( ▼ 0.22% )) and Nvidia ($NVDA ( ▲ 0.77% )) is only $7.5 trillion. It’s also roughly equal to the entirety of US federal government spending last year. Imagine if every dime the government spent in a year went towards data center projects. Data center costs will be astronomical and how they are built and powered matters. A lot.

In response to this million (billion? trillion?) dollar question, some are looking to the stars. In the last 12 months, Google announced Project Suncatcher, Bezos predicted gigawatt-scale space-based data centers (46:39), and Elon was Elon:

Consider them marked.

If the goal of the AI race is to build out as much compute as quickly as possible, we can examine the factors currently limiting progress and the first principle reasoning that suggests space might be less sci-fi than it initially appears. Interestingly, for a $7 trillion project, the primary constraint isn’t actually capital.

The current limitations of terrestrial data centers are currently power constraints and building permits with 30-50% of large data center builds expected to be delayed in 2026.

Analyzing this problem in the context of rapid and expansive construction, we arrive at a relatively simple conclusion: pursue abundant, inexpensive power where permitting constraints are not a factor.

An idea both conceptually and physically “out there”, space actually makes a lot of sense. Primarily, it solves our abundant power problem. Consider that global power consumption was roughly 200,000 tWh in 2024 (round numbers). The solar energy that hit the planet’s surface in that same period was roughly 13,000x greater. To give context to this difference in scale, if all the power consumed on earth is represented by the height of an egg, solar irradiation is the Burj Khalifa.

We’ve established solar is a plentiful energy source, but its benefits actually increase in orbit due to more available and intense solar irradiance.

I’ll start with the simpler of the two: availability. On earth, due to weather and day-night cycles, solar panels only produce 20-30% of the energy they are theoretically capable of if left in the sun 100% of the time (this is called capacity factor and is different from efficiency). In space, satellites can move in a specific orbit called a sun-synchronous orbit (SSO). An SSO keeps the satellite in the sun 24/7/365 by spinning it around the dusk-dawn line as the earth orbits the Sun.

Avoiding the day-night cycle and inclement weather on earth can push a solar panel’s capacity factor to nearly 100%, massively increasing its utility.

Solar irradiance is more intense in space, approximately 8 times more, although the explanation is a bit nuanced. The Sun provides 1360 ^W/_m² in space, but only around 163 ^W/_m² on average at the planet's surface. This dramatic reduction is driven by two key factors. First, only 48% of the Sun’s energy actually reaches the planet's surface, with the rest being absorbed or reflected by the atmosphere. The second primary factor relates to Earth's geometry and the fact that this is average irradiance. As NASA explains:

For people who like the underlying math, keeping one side of the planet dark automatically halves our average irradiance from 1360 ^W/_m² to 680 ^W/_m². Then looking at the surface area equations for a hemisphere and a circle (assuming light hitting the planet is contained within circular beam) we can see we halve again. With an equal radius, the area of a hemisphere (2 π r²) will always be double that of a circle (π r²). Distributing the same amount of energy over double the area will halve our average irradiance again to arrive at 340 ^W/_m². Then multiplying by 0.48 (losses from atmosphere) we arrive at our original 163 ^W/_m².

Ultimately, the availability and intensity of solar irradiation are intrinsically linked. Some areas of the planet are much better suited for solar energy and can achieve much higher than average irradiance (e.g., the equator with mostly sunny days). However, all terrestrial solar projects are subject to atmospheric losses and day-night cycles, which means that, from a physics perspective, space will always be a better choice.

Claims that space is excellent for cooling due to the extremely low temperatures, while intuitively making sense, unfortunately aren't true (at least at the operating temperatures of modern GPUs). Space is a vacuum, which means the traditional methods of heat dissipation on Earth don't apply, and we are left with only radiative heat transfer. More to come on this next week in the Engineering section—get ready for some math.