AI Data Centers in Space?
Posted on | August 16, 2026 | No Comments
Citation APA (7th Edition)
Pennings, A.J. (2026, Apr 16) AI Data Centers in Space? apennings.com https://apennings.com/space-systems/ai-data-centers-in-space/
Introduction
I drove past the SpaceX GigaSat factory being built in Bastrop, Texas, the other day with my wife, after we hiked in the area. It’s big. SpaceX is big news after the IPO, and more so with the news of TeraFab, the mile-long chip fab breaking ground near Texas A&M.
Part of the recent hoopla is debate about putting AI centers in orbit. I’ve had my doubts about the possibility, but SpaceX is betting big and many others are going along for the ride. This post asks what SpaceX’s AI1 is, how it overcomes challenges of power and heat, and how it can offer AI services while circling Earth.
AI1 is SpaceX’s proposed first-generation orbital data-center satellite, designed to host AI compute payloads in low Earth orbit. Unveiled in detail by SpaceX/Elon Musk in June 2026 (ahead of the company’s IPO), AI1 is a large spacecraft optimized for running artificial-intelligence workloads in space rather than providing broadband like Starlink.
Key published specifications for the AI1 satellite include a solar power system that uses about 150 kW peak and about 120 kW average sustained compute capacity. This is roughly comparable to one high-end terrestrial AI rack, such as an Nvidia GB300-class system.
The AI1 satellite is about 70 meters (230 feet) wide with a deployed wingspan, and about 20 meters tall. It will be wider than the wings on a Boeing 747 and about as tall.
It is planned for low Earth orbit, about 600 km above Earth. Its architecture will be dominated by the large deployable solar arrays for power generation and radiators for heat rejection into space. Like other Starlink satellites, it uses laser inter-satellite links for data movement and has an interchangeable compute payload bay, so different processors can be installed. As would be expected, it is designed to be compatible with the Starlink constellation of spacecraft.
SpaceX describes its design as simpler than a Starlink broadband satellite because it omits the complex phased-array antennas needed for direct ground user links. Much of the solar, thermal, and structural technology builds on Starlink V3 experience, engineered to provide a 10-fold increase in downlink capacity, around 1 Tbps, and 160 Gbps Uplink. About 22 times more than the Starlink V2 satellites.
SpaceX describes its design as simpler than a Starlink broadband satellite because it omits the complex phased-array antennas needed for direct ground user links. Much of the solar, thermal, and structural technology builds on Starlink V3’s experience, engineered to provide a 10-fold increase in downlink capacity, around 1 Tbps. It’s uplink is projected to be 160 Gbps, or about 22 times more than the Starlink V2 satellites.
SpaceX will likely deliver them into space with its Starship rockets. SpaceX has stated plans to begin launching the first AI1 units around late 2027, after flying smaller “canary” compute payloads on regular Starlink satellites to validate operations. Longer-term ambitions include scaling toward a very large constellation of AI1 satellites (filings have referenced up to a million satellites in some concepts).
The core idea for using data centers in space is to move AI training and inference workloads off Earth’s power grids and into orbit, where continuous solar power and radiative cooling into the vacuum of space become available.
This offers several advantages over land-based AI/data centers. SpaceX emphasizes the abundant, continuous solar energy without weather, night, or land-use constraints. It will also reject heat by radiating waste heat directly into space using large deployable radiators (See more below).
This arrangement provides potential long-term cost and scalability advantages once launch costs fall (especially with Starship) and manufacturing scales. This is big, because launch prices are currently too high for the AI1 to be profitable. A lot will depend on mass-producing Starships and recovering them quickly to get them back into space. Delivery costs need to get down to $20-130 per kilogram.
Another critical component is the ability to form a distributed orbital compute network linked by lasers, with data downlinked via Starlink or dedicated links. Light travels faster in space than through fiber-optic cabling.
Practical operation will involve solar arrays to power the computations. The compute module (initially flexible on-chip supplier; longer-term plans involve radiation-tolerant, higher-temperature chips such as those targeted by the related Terafab/D3 efforts) runs AI workloads.
Waste heat is rejected through liquid-loop radiators oriented edge-on to the Sun. Inter-satellite laser links connect multiple AI1 (and Starlink) satellites into a larger orbital fabric. Ground users or terrestrial data centers can access the compute capacity remotely.
This forms the first step in SpaceX’s broader vision of space-based AI infrastructure, intended to complement (and eventually potentially rival) ground-based hyperscale data centers for certain workloads. The approach remains early-stage; thermal management, radiation hardening, launch cadence, reliability, and overall economics are still being proven at scale.
Solar panels were invented for satellites, so the power systems have been tested over time. The major issue is the additional heat created by the GPUs. The solution is likely not convection, but rather the radiation of the heat through UV light.
The Stefan-Boltzmann Law states that the total heat power emitted from a blackbody’s surface is directly proportional to the fourth power of its absolute temperature. As an object gets hotter, the thermal radiation it gives off increases very quickly.
In other words, make those radiators really hot, and you will get something like 16x the heat radiation. Run the liquids past the chips and let the heat collect at the radiators. Each AI center will use a 110 m² deployable liquid radiator, depending on how much compute it performs.
Notes
[1] Farzad provides excellent videos on SpaceX.
AI Prompt(s) What is AI1 and how can be used for data centers in space?
© ALL RIGHTS RESERVED
Not to be considered financial advice. AI is often used, and results are thoroughly interrogated. Links are used for some citations.
Anthony J. Pennings, PhD is a Professor at the Department of Technology and Society, State University of New York, Korea and a Research Professor for Stony Brook University. He teaches AI and broadband policy. From 2002-2012 he taught digital economics and information systems management at New York University. He also taught in the Digital Media MBA at St. Edwards University in Austin, Texas, where he lives when not in Korea.
Tags: AI1 Satellite > GigaSat > solar panel array
Comments
Categories
Referencing this Material
About Me
Professor (full) at State University of New York (SUNY) Korea since 2016. Research Professor for Stony Brook University. Moved to Austin, Texas in August 2012 to join the Digital Media Management program at St. Edwards University. Spent the previous decade on the faculty at New York University teaching and researching information systems, digital economics, and global political economy
You can reach me at:
anthony.pennings@gmail.com
apennings70@gmail.com
anthony.pennings@sunykorea.ac.kr
About me
Writings by Category
- artificial intelligence
- automatrix
- blockchain
- books and films
- characteristics of digital media
- creative cities
- creative industries
- crisis communications
- cryptocurrencies
- cultural Industries
- data analytics and meaning
- democratic political economies
- digital coordination
- digital destruction
- digital geography
- digital media economics
- digital media management
- digital monetarism
- digital spreadsheets
- disaster risk management
- disruption
- dystopian economies
- educational innovations
- electric money
- enabling frameworks
- enterprise systems
- financial news
- financial technology
- food policy and security
- futurists
- geopolitical risk
- global communications
- global e-commerce
- health IT
- how IT came to rule the world
- intellectual property
- journalism
- meaning makers
- meaningful play
- media strategies
- mobile technologies
- multimedia
- political economies in sf
- political economy of media
- public administration
- science and technology studies
- sensing technologies
- smart new deal
- social media
- space systems
- spreadsheet logic
- strategic communications
- sustainable development
- technologies of meaning
- technology management
- telecom policy and net neutrality
- telegraphic political economy
- telelearn/telework
- the cold war
- the smith effect
- uncategorized
- visual literacy and rhetoric
