The first Malaysian individual to send a satellite.
Not a national space agency, not a corporation. KENSAT is the first satellite from a Malaysian individual — proving that orbit is now within reach.
Built and launched by the first Malaysian individual to send a satellite to space.
KENSAT is a 2U CubeSat carrying Gemma 3, a real open-weights language model, into low earth orbit. It thinks in space, and beams the answer back to anyone listening.
─ sun-synchronous orbit · 510 km · one orbit ≈ 95 min
□ launch site · drag to rotate
Not a national space agency, not a corporation. KENSAT is the first satellite from a Malaysian individual — proving that orbit is now within reach.
Almost every AI in orbit today belongs to a government, a defense program, or a billion-dollar company. KENSAT carries its own — a real model running onboard.
A computer that no one else can access or switch off. Sovereign compute is the new sovereign money — and it shouldn't be reserved for states and corporations.
Running real inference on hardware in orbit, far from any data center, proves that capable AI can live at the edge — even the highest edge there is.
The radio system's tape-measure antennas — KENSAT's voice and ears.
The moment of release — antennas spring open the instant KENSAT leaves its deployer.
The brain of the spacecraft — where the AI actually runs.
Electrical power: storing and distributing every watt onboard.
Power protection & management — keeping the bus safe and stable.
The backbone everything plugs into — the CubeSat bus.
Gemma 3, Google's open language model, runs onboard KENSAT. Send a message up, and the answer is computed 510 km above you, then beamed back down.
Pre-launch: replies come from an identical ground copy of the onboard Gemma 3 model. After launch, inference moves to orbit.
GROUND>kensat, where are you right now?
KENSAT>On the ground at Vandenberg, waiting for my ride to 510 km. After launch, every answer gets computed on the onboard GPU and beamed down with radio — ask me anything while we wait.
Imagine reaching out to OpenAI, Microsoft, Google, or Anthropic's Claude to run their models on a data center in space. That's the launch.
Gemma 3, Google's open-weights language model, runs onboard the NVIDIA Jetson Orin Nano — inference performed in orbit, not on the ground.
KENSAT is a testbed for letting an onboard model help run the spacecraft itself.
Test the effects of radiation in sun-synchronous orbit — the specific orbit and altitude we care about — on the compute hardware.
Characterize how the payload behaves thermally through the heat and cold of every orbit.