To scale AI the world needs more data centers to house and run the necessary IT equipment. The current data center supply is 70 gigawatts but in five years or so, we’ll use an estimated 220 gigawatts, according to McKinsey. That means we’ll need to find space to build more data centers and the energy to power them.
That is where U.S. startup Starcloud comes in. It wants to enable a transition from terrestrial data centers to building them in space, a shift that promises to have a major impact on how we store and protect the world’s data.
The transition is likely to take about 20 years but “in ten years almost all new data centers will be in space,” predicts Philip Johnston, CEO of Starcloud.
Customers won’t have to wait that long. If all goes according to plan, Starcloud will begin by selling compute to NASA and U.S. Department of Defense (DOD) earth constellation satellites and start serving enterprise customers from space later this year, Johnston said during an interview with The Innovator. He expects Starcloud’s orbital data centers to be able to slash the energy cost of inference by 10X.
Starcloud is one of 100 young companies from 28 countries selected as 2025 Technology Pioneers by The World Economic Forum. The latest cohort, selected for their potential to transform industries and tackle pressing global challenges, was announced at the Forum’s Annual Meeting of The New Champions in Tianjin, China.
The Case for Space
Terrestrial data centers can’t keep up with demand. Building new data centers terrestrially requires land, permitting can take up to a decade, and you need to factor in the cost of solar cells, battery storage, backup power and generators, says Johnston.
The amount of energy needed to run them is a major constraint. The surge in data centers is pushing up prices as well as overall electricity consumption And, if current trends continue, multi-GW clusters will be required from 2027 to train next-generation AI models of the future. A 5 GW cluster will be needed to train models like Llama 5 or GPT-6. This would exceed the capacity of the largest power plant in the U.S and some of the largest operational power plants in the world, making these clusters simply not possible with today’s energy infrastructure, according to a white paper on Starcloud’s site.
“In space you don’t need land, so the cost is zero, you don’t need battery storage or backup because the sun is always shining, and you need less solar, so the three biggest costs are slashed,” says Johnston.
The biggest inhibitor to putting data centers in space to date has been the cost of the launch but that is set to rapidly decline, Johnson says.
Space X is working on the first every fully reusable rocket, reducing the marginal cost of a launch from around $60 million to as low as $2 million. The payload will increase 5x, enabling Starship to economically send enormous amounts of compute and solar equipment to space, says Johnston. The U.S. startup has patents pending on large, low cost radiators for cooling in deep space. It has also developed hardware to shield against radiation and software to override the effects of radiation.
In 2027 when launch costs comes down Johnston says he expects the company to be competitive with terrestrial data centers on energy costs, charging less than 0.5 cents per kwh.
If you compare the cost of a single 40 MW cluster operated for 10 years on land versus space the advantages are clear, says Johnston. Operating a terrestrial data center over 10 years would incur energy costs of around $167 million including $140 million at $0.04 per kwh, the use of 1.7 million tons of water, $2 million worth of solar array, $7 million in cooling (the chiller energy cost about 5% of overall power) and $20 million for backup power supply at commercial equipment pricing.
The same cluster in space over 10 years would cost $8.2 million, according to Starcloud’s calculations. This includes $5 million for a single launch of compute module, solar and radiators, $2 million for solar arrays, $1.2 million for radiation shielding per kwh of compute and a $39/kg launch cost. Water and backup power supply are not required. As newer AI chips emerge these can be sent to space and added to the mix, Johnson says.
Advantages include more than cost. To scale to gigawatts in orbit, compute modules, power, cooling and networking can be assembled together in a modular fashion. Compute modules can also be assembled with architectures that scale in 3D rather than 2D as on Earth, ensuring the cluster is as tightly coupled with as low latency within the cluster as possible (a critical property of AI training clusters).
Starcloud’s approach also offers better cybersecurity, says Johnston. It prioritizes direct satellite-to-ground communication, bypassing terrestrial networks and reducing exposure to common cyber threats, a model that appeals to governments and companies with strict cybersecurity and data sovereignty mandates, says Johnston.
The Business Model
Johnston says there will be two phases to Starcloud’s business. The first phase will be correlated to the cost of rocket launches. While the cost is still high it will sell compute to earth observation constellations run by the U.S. DOD and NASA, running inference, and downloading the insights, as a service.
In the second phase Starcloud intends to compete with AI infrastructure providers like Coreweave and Lamda, selling H100 and other GPU capacity for AI training and inference to AI Cloud providers like AWS, Azure and smaller cloud brokers, Johnston says. Corporate clients of the cloud providers will benefit from the drastically reduced energy price of inference, he says,
The Data Center Space Race
Starcloud is not alone in its ambition to send data centers into space. In May Chinese companies Chengdu Guoxing Aerospace Technology (also known as ADA Space) and Zhejiang Lab launched 12 satellites for an in orbit AI compute project, as part of a wider Star-Compute Program, which aims to build a huge on-orbit network of 2,800 satellites.
European companies are also aiming for the space. Last year Thales Alenia Space, a joint venture between French and Italian aerospace groups Thales (67%) and Leonardo (33%), announced promising results of its European Commission-funded ASCEND (Advanced Space Cloud for European Net zero emission and Data sovereignty) feasibility study. ASCEND envisions orbiting data platforms twice as large as the International Space Station Like Starcloud’s approach the server racks at the heart of the ASCEND platforms would be powered by vast solar arrays/
Meanwhile, Florida-based Lonestar Data Holdings recently claimed to have successfully tested a tiny data center the size of a hardback book in space. It was sent to the Moon on the Athena Lunar Lander from US space exploration firm Intuitive Machines, which was launched by a rocket from Elon Musk’s SpaceX.
Johnston says Starcloud’s key differentiators are the amount of compute it plans to offer from the start, the cost savings it can offer government and enterprise clients and the quality of its team, which includes former NASA and Space X employees as well as data center and satellite design experts.
The company has raised a total of $23.5 million, including a $21 million seed round in September of last year.
This article is content that would normally only be available to subscribers. Sign up for a four-week free trial to see what you have been missing.
To access more of The Innovator’s Startup of the Week articles click here.
