An Indian space startup launched what it calls the country's first orbital computing satellite on October 1, 2026 — and it's carrying NVIDIA chips.
TakeMe2Space's MOI-1A spacecraft, weighing less than 50 kilograms, rode to orbit aboard SpaceX's Transporter-18 rideshare mission. Instead of beaming huge volumes of raw satellite imagery back to Earth, the satellite runs AI models directly in orbit and sends down only the answers — processed analysis, not the full underlying dataset.
The pitch is simple and economic. Earth-observation customers today pay to download massive raw datasets and then run the compute themselves. TakeMe2Space flips that: customers upload their own containerized AI models to the satellite, it crunches the imagery as it passes over a specified area, and transmits only the results. "The customer who sees value in our first-generation satellites is the one paying to download a lot of raw data and then do the compute," founder and CEO Ronak Samantray told Reuters. "The reduction in the cost of not downloading it offsets the premium they pay for compute in orbit."
The first customer list is already real: 23 customers, including geographic information system companies and educational institutions. Commercial use spans agriculture, mining, supply-chain monitoring, and insurance. US space data analytics firm Little Place Labs is among the customers.
The spacecraft itself is modest — an onboard edge-computing satellite running on about 150 watts of power, a far cry from a terrestrial data center. That's the point: it does AI inference in orbit rather than pretending to be a server farm. A technology-demonstration mission in late 2024 already validated that TakeMe2Space can upload applications, perform AI inference in orbit, and download results. The earlier MOI-1 satellite was lost after a launch-vehicle third-stage failure; MOI-1A is the comeback flight.
The other satellite on the same rocket
Here's the fun part: SpaceX's Transporter-18 is also carrying Google's Project Suncatcher prototype today — a fridge-sized satellite with four Google TPUs. Google's mission is an experiment to answer one question: can AI hardware survive and operate in space? TakeMe2Space is asking a different question: will customers pay for compute in orbit? One is a moonshot test; the other is a product with paying customers. The orbital AI race is splitting into two tracks already.
Why it matters
Downlink bandwidth is one of the biggest bottlenecks in Earth observation — satellites generate far more data than they can cheaply send home. Processing at the edge shrinks that bottleneck, and the economics finally work. If MOI-1A performs, TakeMe2Space plans bigger, networked satellites capable of serving multiple customers starting in 2027.
The honest take: this is one of those rare space stories that's already a real business. Twenty-three customers signed before the rocket even left the pad. The AI industry keeps building bigger data centers on Earth — while a 50-kilogram box in orbit quietly started selling compute by the answer.
Comments
Post a Comment