Agnikul to launch orbital AI data centre prototype by year-end, targeting 2027 commercialisation

Agnikul Cosmos, a Chennai-based space company, is building something that sounds like science fiction but is grounded in a very real problem: the world is running out of room for data centres. By the end of this year, the company plans to launch a prototype of an artificial intelligence data centre into orbit, with the goal of running it commercially by 2027. The venture is a collaboration with NeevCloud, a Bengaluru-based AI platform, and it represents a fundamentally different approach to where and how we process information at scale.

The scale of the challenge is staggering. Global demand for data centre capacity is expected to grow between 19 and 22 percent annually through 2030, reaching somewhere between 171 and 219 gigawatts of power consumption per year, according to McKinsey. Traditional data centres are massive, power-hungry installations that require enormous amounts of water for cooling and vast tracts of land. Companies like SpaceX, Google, and Axiom have all begun exploring whether space might be the answer. Agnikul's co-founder Srinath Ravichandran sees it as inevitable. "In space, there is an availability of unlimited solar energy, and cooling is much more efficient because you are exposed to temperatures close to absolute zero," he said in an interview. Beyond the physics, there is also a security argument: a data centre in orbit is simply harder to access, which means the data stored there is harder to compromise.

The system Agnikul is building will not be a single satellite. Instead, it will be a constellation of upper stages from rockets, each one hosting different modules of the data centre, all orbiting in low Earth orbit and communicating with each other and with ground stations. The company is not attempting to build a facility for training AI models, which would require enormous computational resources. Instead, it is focusing on inference—the process of taking a model that has already been trained and using it to analyze new data and make predictions. This is less energy-intensive than training, which makes it a more realistic first step for an orbital system.

The engineering challenges are significant but not unsolved. The constellation will need to manage power carefully, since satellites pass through Earth's shadow roughly half the time. During these eclipse phases, when sunlight is blocked, the system will shift into low-power mode, handling tasks like data transmission. When the satellites face the sun again, they will switch to high-power mode for the heavy computational work. This switching framework already exists in imaging satellites, so Agnikul is not inventing the concept from scratch. The company has also invested heavily in radiative cooling—a process by which heat is shed directly into space through thermal radiation—which is how the system will maintain safe operating temperatures without the water cooling that terrestrial data centres depend on.

Ravichandran is clear that the prototype launching at year's end is not the beginning of commercial service. It is a proof of concept, a demonstration that Agnikul can actually build and operate an AI data centre in space. The real business begins in 2027, when the company moves into what Ravichandran calls "commercialisation mode." That is when the system will open up to paying customers and when Agnikul's addressable market will expand significantly. For now, the company is focused on showing that the idea works—that you can take the upper stage of a rocket, outfit it with computing hardware and cooling systems, put it in orbit, and have it reliably process AI inference tasks while drawing power from the sun. If that demonstration succeeds, it will establish a new category of infrastructure and prove that some of the world's most demanding computational work does not have to happen on Earth.