Opinion
AI is going to space, but can it stay there?
“Sunlight could power AI in orbit, but sunlight cannot deliver propellant or replace a failed processor,” writes Shahar Bahiri, Co-founder & COO at Moonshot Space. “The next opportunity is building the delivery service that keeps those expensive computers working long enough to justify putting them there.”
The moment AI takes a tedious task off your hands, you start looking for something else to give it. You try it on an email, then a presentation, then the report you have been putting off all week. Before long, it becomes part of your working day.
Now multiply that experience by millions of people and businesses. Companies are using AI to write software, answer customers and handle work that once occupied whole teams. Every useful application gives us another reason to use it. And we are only beginning to discover what it can do.
On our screens, this growth looks effortless. Behind every answer, computers are working, and those computers need electricity. The more we ask AI to do, the more power we need to supply.
We can introduce an AI service to millions of users almost overnight. Building the power plants and transmission lines to support it takes years. In major data-center hubs, demand is already moving faster than electricity infrastructure can expand. Keep widening that gap, and the limit on AI's growth becomes simple: we can build computers faster than we can supply the power to run them.
So where will the electricity come from? Commercial fusion reactors could eventually provide a vast new source, but we cannot yet build them at scale. Meanwhile, the largest fusion reactor in the solar system is already working: the sun.
What if we put some of our computers in space, where they could use that energy without clouds, atmospheric interference or the daily interruption of sunset?
A carefully chosen orbit can follow Earth's boundary between day and night, known as the terminator. At the right altitude and alignment, a satellite can remain in sunlight almost continuously as Earth turns beneath it.
Equip that satellite with computers and large solar panels, and you have the outline of an orbital data center. The panels generate electricity, the computers do the work, and the results travel back to Earth. Nearly 24/7 sunlight means more energy collected over time and less need for batteries to cover the night.
When I talk to investors about orbital AI data centers, the conversation usually turns to radiation and cooling. Satellite builders already design for radiation and heat, although orbital AI takes those demands to a new scale. The question I hear less often is what happens once they work: how do we keep these expensive platforms in orbit long enough to pay for themselves?
Related articles:
- Secretive Israeli startup Moonshot Space raises $12M, unveils electric launcher for hypersonic and orbital missions
- Israel expands eyes in the sky with Ofek-19: “Space allows us to make the entire Middle East accessible to the IDF”
- Israel expands space influence with new partnerships at International Astronautical Congress
SpaceX's Starlink network follows a model built around replacing satellites when their working lives end. To understand why replacement can make sense, consider a hypothetical satellite costing $600,000 to build and launch and generating $1.2 million annually. Its initial cost equals 6 months of revenue, before expenses. Years of operation can then justify retiring and deorbiting it instead of paying to refuel or repair it.
An AI platform carrying far more expensive computing equipment changes that calculation. Suppose its business plan requires 5 to 7 years to recover the investment. Losing it after 3 years means it has not even recovered its cost. Running out of propellant or suffering a replaceable component failure could end an otherwise productive investment. Keeping it supplied and maintained becomes essential to making the business work.
Getting into space is hard. But staying there is even harder.
Staying there takes more than electricity. Satellites use thrusters to adjust their orbits and avoid collisions. In low Earth orbit, traces of atmosphere create drag that gradually lowers their altitude unless it is counteracted. Solar panels can supply energy, but they cannot replenish the propellant those thrusters consume. A functioning computer can therefore outlast the spacecraft's ability to maintain its intended orbit.
Nor should a new generation of processors force an operator to discard useful solar arrays, cooling systems and communications equipment. You can upload new software. A new chip has to be physically delivered and installed.
Rockets are excellent at moving infrastructure into space, carrying satellites weighing tonnes. But consider a platform that later needs 200 kg of propellant to extend its working life. Booking an entire heavy-lift launch for that delivery would make little economic sense. The operator needs a service sized and scheduled for replenishment.
Think of someone opening a coffee shop. Bringing in the expensive coffee machine may require a large truck and a crane to put it in place. That delivery makes sense: it installs the equipment the business will depend on.
Now the shop is running, and the barista realizes there are not enough beans for tomorrow. The order is 10 bags weighing 1 kg each. Calling back the truck and crane would turn a simple restocking job into an expensive operation. Waiting for their availability could leave the shop unable to serve customers. What the barista needs is a delivery service that brings those beans when needed, at an affordable price.
Orbital AI faces the same distinction between installing equipment and keeping it supplied. Shared rocket launches can carry small payloads, but a place on a launch manifest is not a delivery to a particular satellite. The missing service must get propellant and replacement parts to the asset, when needed, and support refueling or installation once they arrive.
That service must cost less than the productive life it preserves. Satellites designed to receive supplies, dependable delivery schedules and affordable small shipments must become part of the business from the beginning.
This is why I have spent the past 4 years working on a space delivery system. The opportunity goes beyond putting valuable machinery in orbit. It is building the recurring service that keeps that machinery supplied, maintained and commercially useful.
Sunlight could power AI in orbit. But sunlight cannot deliver propellant or replace a failed processor. The next opportunity is building the delivery service that keeps those expensive computers working long enough to justify putting them there.
Shahar Bahiri is Co-founder & COO at Moonshot Space.
