Loft Orbital and Marlan Space have announced a deal valued at $1 billion to deploy a constellation of 50 satellites capable of running artificial intelligence directly in orbit, with Mistral as the technology partner for the AI models. The first ten satellites are already part of the initial plan, with the first one expected in 2026, in a project aimed at processing images and space data before sending them to Earth.
Mistral’s AI in space in 30 seconds
- Loft Orbital and Marlan Space are planning a $1 billion, 50-satellite orbital infrastructure.
- Mistral will contribute models for reasoning and natural-language processing.
- The satellites will analyze data from optical sensors, radar, and other instruments in orbit.
- Early planned uses include wildfires, maritime surveillance, disasters, and critical infrastructure.
- Orbitworks will manufacture the satellites in Abu Dhabi, with deployment starting with ten units.
The project was unveiled in Paris during the International Space Summit and received public backing from French President Emmanuel Macron. The initiative brings together French and American technology from Loft Orbital, industrial capacity from the United Arab Emirates, and models from France’s Mistral.
Its backers describe it as the largest AI-dedicated satellite infrastructure announced so far. That’s a claim from the consortium itself, and it should be distinguished from other projects, still at various stages of development, that propose deploying large amounts of computing capacity in orbit.
The idea isn’t simply to put a conventional data center on a satellite, either. The immediate goal is to bring processing closer to the sensors that generate the data, avoiding the need for all that information to travel to ground stations first.
Processing images before sending them to Earth
The traditional way many observation satellites work involves capturing information, storing it temporarily, and waiting until a suitable connection to a ground station is available.
Only once the data is downloaded does processing begin.
That can introduce significant delays when what matters is detecting a wildfire, a ship, a change at a facility, or the aftermath of a natural disaster.
The architecture Loft Orbital is proposing tries to change that flow:
Sensors → in-orbit processing → AI model → result → transmission to Earth
Instead of necessarily downloading an entire large image to analyze it later, the satellite can run algorithms on the data and transmit information that’s already been processed.
The project’s leaders say certain results could be sent within seconds, compared with systems where the full process can take hours.
The benefit depends on the use case. Not all data can or should be discarded after processing, and transmitting information to Earth will still be necessary for many applications.
But in scenarios where responding quickly matters, deciding which information deserves priority before using the communications link can reduce both latency and the volume of data transmitted.
| Planned application | What AI can do in orbit |
|---|---|
| Wildfires | Detect signs in images and prioritize alerts |
| Maritime surveillance | Identify and track vessel activity |
| Natural disasters | Quickly analyze affected areas |
| Ports | Detect changes or relevant activity |
| Critical infrastructure | Identify certain events or anomalies |
| Earth observation | Classify information before transmitting it |
The constellation will combine radar, optical systems, and other instruments, allowing it to work with different types of information.
Mistral will provide the AI models
Mistral’s role will be to provide models capable of running inside this infrastructure.
According to the announcement, the French company will work on reasoning capabilities and natural-language interaction, in addition to powering services available to the constellation’s customers.
One of the concepts the partners want to develop is agentic AI in space.
In this scenario, a user could define a high-level goal, and the system would coordinate different sensors, models, and satellites to obtain the needed information.
For example, a maritime surveillance application might need to locate certain vessels, analyze images from different sensors, and decide which observations should be transmitted with priority.
How much autonomy is actually achieved will depend on the applications ultimately deployed. The announcement doesn’t specify which Mistral models will run, their size, the acceleration hardware used on each satellite, or how much compute capacity the constellation will have.
Therefore, 50 AI satellites don’t necessarily amount to 50 general-purpose orbital data centers.
At least initially, the project is closer to a distributed edge computing infrastructure, where the satellite itself is the edge of the network.
An app store for sharing the satellites
Another unusual element will be the creation of a sort of AI app store for the constellation.
Customers will be able to select models or applications and run them on a shared orbital infrastructure.
The concept resembles how a cloud platform works: different users share a common infrastructure without having to individually buy and operate a full satellite.
Loft Orbital brings experience specifically in this shared-satellite model.
The company says it has already launched more than 100 payloads and 20 AI-related missions in orbit. Its past work includes a demonstration with NASA’s Jet Propulsion Laboratory (JPL) in which a vision-and-language model ran aboard a satellite.
The new constellation aims to bring that concept to a larger commercial scale.
Its target customers include government agencies in France, other European countries, and the United Arab Emirates, as well as international private companies.
The satellites will be built in Abu Dhabi
Manufacturing will be handled by Orbitworks, a joint venture created by Marlan Space and Loft Orbital.
The company has production facilities in Abu Dhabi built for manufacturing satellites at scale.
This industrial piece carries strategic weight for the United Arab Emirates.
Hamdullah Mohib, CEO of Marlan Space, explained during the announcement that the goal is to develop in the country not just the capacity to use space services, but also engineering, intellectual property, and a supply chain tied to their manufacturing.
The 50 satellites are designed to operate in coordinated groups with different capabilities.
The program already included a first phase of ten satellites, whose deployment is set to begin in 2026. The project’s leaders haven’t published a full timeline for putting all 50 units into orbit.
For that reason, the $1 billion figure should be understood as the announced value of the deal and the planned program, not as investment already deployed or as a constellation currently in operation.
Space is starting to be seen as a new location for compute
The Loft Orbital, Marlan Space, and Mistral initiative coincides with growing interest in moving more computing capacity off Earth.
The logic makes sense for certain workloads.
Satellites generate growing amounts of information, and transmitting all that data has an energy cost and a physical bandwidth limit. Processing part of it directly in orbit can reduce the communications needed and speed up certain responses.
An architecture like this also comes with major challenges.
Available power is limited, cooling works very differently in a vacuum, hardware is exposed to radiation, and physically repairing a server that fails can be impossible. On top of that, electronics sent to space have to withstand conditions a terrestrial data center will never face.
AI models add another challenge: they change fast.
A GPU installed in a data center can be swapped out once it’s obsolete. Physically upgrading the processor on a satellite that’s already launched is a different matter, so the useful life and software upgradability take on considerable importance.
The announced project also shouldn’t be confused with the much more ambitious proposals for large-scale orbital data centers, like the ones Intel has proposed.
SpaceX has filed regulatory plans related to a massive orbital AI infrastructure, while Blue Origin has also floated space-based computing systems. Whether those projects reach the proposed scale will depend on economic, technical, and regulatory factors still to be worked out.
Loft Orbital and its partners are starting from a different approach: using AI to process, on the satellite itself, the data that satellite just generated.
Mistral’s involvement adds generative models and reasoning capabilities to that architecture. If the rollout proceeds as announced, the first major test will come with the initial ten satellites: proving that AI in orbit can move from technology demonstrations to delivering shared services on an ongoing basis.
via: Press release
Frequently asked questions
How much is the Loft Orbital-Marlan Space satellite deal worth?
The deal is valued at $1 billion, covering a planned 50-satellite constellation, announced during the International Space Summit in Paris.
What role will Mistral play in the project?
Mistral will provide AI models for reasoning and natural-language processing that run aboard the satellites and power services for the constellation’s customers.
Where will the satellites be built?
Orbitworks, a joint venture between Marlan Space and Loft Orbital, will manufacture them at its production facilities in Abu Dhabi.
When will the first satellites launch?
The first phase of ten satellites is set to begin deployment in 2026. The project’s leaders haven’t published a full timeline for putting all 50 units into orbit.
What will the satellites be used for?
Planned uses include wildfire detection, maritime surveillance, natural disaster response, and monitoring of ports and critical infrastructure.
