The European Union and the United Arab Emirates are taking AI computing off the ground with a $1 billion plan to turn a satellite constellation into an orbital data center complete with, of all things, an app-style "LLM store." The program, announced by UAE-backed Marlan Space and France-headquartered Loft Orbital, grows the Altair Earth observation constellation from 10 satellites to 50 and is the latest sign that model inference is no longer confined to server racks on land.
French President Emmanuel Macron unveiled the initiative at the International Space Summit in Paris, positioning it as a strategic bet that gives both Europe and its Gulf investor an independent path in frontier AI at a time when US and Chinese labs dominate the leading models.
What the constellation actually does
This is not a typical imaging satellite program. The Altair-Next Gen fleet is built around the idea of interpreting data in space rather than beaming raw imagery down to Earth for processing. Each satellite carries radar, optical and other sensors and is designed to make sense of what it sees before anything reaches the ground.
The stated goal is speed: the system can send an alert within seconds, whereas traditional setups that download raw imagery and process it on the ground can lose hours in the pipeline. The partners describe the resulting capability as "permanent," a step up from the delayed, batch-style processing that defines most current Earth observation operations.
The "LLM store" is the genuinely novel piece
The most interesting part of the proposal is the centralized AI application store orbiting aboard the constellation. Customers do not simply receive processed data. They can choose which models to run on satellite-collected information, execute them on shared spacecraft compute, and export the results to external models if needed.
- An in-orbit "store" lets users pick which LLMs run on satellite-collected data.
- Customers pay for inference on shared spacecraft "compute units" rather than for fixed sensor output.
- Data can be exported to external models for further processing on the ground.
French AI lab Mistral is the anchor model provider. It supplies the onboard models that give the constellation its reasoning and natural-language tasking capability in orbit, and it will power the services in the AI application store. It remains unclear whether the store will be limited to open-source or open-weight models, a question the partners have left open and one that will matter for how widely the platform gets adopted.
"Working together, Loft Orbital and Marlan Space have assembled a consortium required to build out a 50-satellite, AI-enabled, multi-modal constellation," Alex Greenberg, co-founder and COO at Loft Orbital, told TechRadar Pro. "The satellites are produced by Orbitworks at its facility in Abu Dhabi, the region's largest high-volume site for commercial constellations. Mistral joins the consortium to provide the onboard models giving the constellation its reasoning and natural-language tasking capabilities in orbit, and will power services in the AI application store."
None of the first 10 satellites, which are already in production, have launched yet. The stated timeline for full deployment of the 50-satellite fleet is five years.
Why the orbital push matters
The plan slots into a broader race to build space data centers. US players including SpaceX and Blue Origin have filed plans to host general-purpose compute in orbit and rent capacity to buyers, and the Europe-UAE project is the clearest transatlantic parallel so far.
For Mistral, the deal is a meaningful step. As one of the most powerful European AI players, it remains behind US and China on the very largest frontier models, and the orbital constellation gives it a flagship deployment that ties European and Gulf capital together around its models.
The strategic logic for both backers is straightforward. With US labs at the center of the frontier-model race and Washington debating how fast that frontier should move, an orbital computing platform built around a European model is a hedge that keeps sovereign access to capable AI separate from any single US or Chinese company.
Five years is a long timeline for a fast-moving industry, and the in-orbit store raises as many questions as it answers about pricing, model choice and latency. But the direction is clear: the next frontier in AI infrastructure may be literally up in the air, and the LLM store is coming with it.
Comments