Pixxel, Agnikul explore orbital data centers

Pixxel, Agnikul Cosmos and TakeMe2Space are all now publicly pitching variations of the same idea: move some AI compute into orbit instead of keeping it on the ground. The companies have disclosed separate projects in 2026 that use satellites or rocket upper stages as computing platforms for inference, remote-sensing analysis and other space-native workloads. Their case is that solar power is continuous in orbit for long stretches, cooling conditions are different from those on Earth, and data collected in space can be processed closer to where it is generated. The projects are still early. Pixxel has announced a pathfinder satellite with Sarvam for launch as early as the fourth quarter of 2026, Agnikul and NeevCloud have outlined a proof-of-concept mission before the end of 2026 with commercial operations targeted in 2027, and TakeMe2Space said it is scaling from a single spacecraft toward a six-satellite network. (pixxel.space) ### What exactly are these companies proposing to put in orbit? Pixxel said on May 4 that it will design, build, launch and operate a spacecraft called Pathfinder, a 200 kg-class satellite that will carry datacentre-class GPUs and Pixxel’s hyperspectral imaging camera. Sarvam will provide the AI stack and run both training and inference directly on board, according to Pixxel’s announcement. The company said the satellite could reach orbit as early as Q4 2026. (pixxel.space) Agnikul Cosmos said its system will look different. Srinath Ravichandran, the company’s co-founder, told PTI in March that Agnikul’s prototype would host an AI data centre on the extendable upper stage of its rockets rather than on a standalone satellite, with the initial use case focused on inference rather than model training. Commercial service is targeted for 2027, he said. (pixxel.space) TakeMe2Space has already been marketing orbital compute access through OrbitLab. The company said in January that customers can upload and run AI models on orbital infrastructure, and that it plans to expand to six spacecraft with roughly 5 kilowatts of in-orbit compute in the next phase. (economictimes.indiatimes.com) ### Why do they say orbit helps with AI workloads? Awais Ahmed, Pixxel’s chief executive, said ground-based data centres face constraints around “energy, land, regulation, and scale,” and argued that orbital systems can use abundant solar energy and sit closer to space-based data sources. Pixxel’s pitch is tied directly to Earth-observation workloads: process hyperspectral imagery in orbit, then send down insights instead of raw data. (economictimes.indiatimes.com) Srinath Ravichandran made a similar case for Agnikul’s project. In his PTI interview, he said space offers “unlimited solar energy” and more efficient cooling because the hardware is exposed to very low temperatures, while also offering physical separation from terrestrial infrastructure. He said the first Agnikul-NeevCloud system is meant for AI inference, which needs less power than training. (pixxel.space) TakeMe2Space has framed the argument in operational terms. The company said in January that processing data in orbit rather than downlinking large datasets can reduce latency and costs for Earth-observation applications, and its website says users pay by utilization time to run models directly in space. (economictimes.indiatimes.com) ### Are these full-scale data centres or smaller demonstrations? The numbers disclosed so far point to demonstrations and early platforms, not hyperscale facilities. Pixxel’s Pathfinder is a single 200 kg-class satellite. Agnikul’s first mission is described as a prototype and proof-of-concept. TakeMe2Space’s near-term target is a six-spacecraft network with about 5 kilowatts of compute. (economictimes.indiatimes.com) That does not mean the companies are using the term loosely. In each case, they are describing orbital hardware built to perform computing tasks that would otherwise be done in terrestrial data centres, especially AI inference and image processing. The scale, however, is much smaller than conventional cloud campuses on Earth, based on the specifications they have released so far. That comparison is an inference from the companies’ disclosed payload sizes and power targets. (pixxel.space) ### What are the practical constraints? Latency remains one limit. Agnikul said its system would rely on intra-satellite and satellite-to-ground communications, which means any service still depends on links between orbit and Earth. That makes the model more suited, based on the companies’ own descriptions, to processing data already generated in space than to replacing terrestrial cloud computing for everyday internet applications. (pixxel.space) Launch economics are another constraint. None of the three companies has published a full cost model for delivering compute to orbit at scale, and the current plans all depend on getting hardware into low Earth orbit reliably. Social media discussion around the concept has cited larger reusable launch systems such as SpaceX’s Starship as a possible enabler, but the companies’ disclosed 2026 plans do not rely on Starship. Pixxel has not named a launch vehicle for Pathfinder in its announcement, Agnikul is using its own rocket architecture, and TakeMe2Space has described orbital access through its own satellite program. (economictimes.indiatimes.com) ### What happens next? Q4 2026 is the next named milestone from Pixxel, which said Pathfinder could reach orbit by then. End-2026 is the target Agnikul gave for its prototype AI data-centre launch, with 2027 set for commercial operations. TakeMe2Space said in January that its next phase is a six-spacecraft constellation delivering about 5 kilowatts of compute through OrbitLab. (pixxel.space)