Cambridge Built a Planetary-Scale AI Model, Skipping Nvidia
A Cambridge team has built an AI foundation model for the whole planet. It turns years of satellite data into a reusable "fingerprint" for every 10-meter patch of land on Earth, cutting the data needed for environmental monitoring by up to 30 times. And it was trained and run entirely on AMD chips, not Nvidia.
July 23, 2026 – 4:04 pm
Image by: Pete Linforth
The AI industry runs on Nvidia. A team at the University of Cambridge has just shown it does not have to. They have built a planetary-scale foundation model for the Earth, and every chip that trained and ran it came from AMD.
The Model: TESSERA
This model, called TESSERA, borrows its concept from large language models. Where an LLM learns from text, TESSERA learns from space. It consumes years of imagery from the European Space Agency’s Sentinel satellites, both radar and optical.
It then condenses each 10-meter square of the planet’s land into a compact 128-number "fingerprint" or embedding.
This sounds abstract. The payoff is tangible:
- Normally, mapping crops, forests, or floods from satellite data requires building a bespoke model and hand-labeling thousands of examples.
- TESSERA’s fingerprints enable researchers to build these tools with significantly less data, often on an ordinary CPU. The team claims it needs about 30 times less labelled data than starting from raw imagery.
A Base Layer for the Planet
The result is a kind of base layer for the planet, published free for anyone to use. The more unusual aspect? It was run on AMD hardware.
Training and Processing
Cambridge’s Energy and Environment group trained TESSERA on 16 AMD Instinct MI300X GPUs, using approximately:
- 6,200 GPU-hours
- AMD’s open ROCm software
Vultr, an independent cloud firm that sponsored the compute, provided six bare-metal servers, each packing eight AMD Instinct MI325X GPUs. Together, they produce roughly three terabytes of compressed data daily—equivalent to covering the landmass of Italy daily.
A single year’s global coverage takes months of continuous processing.
Open Access and Impact
The team is releasing the embeddings free under a Creative Commons license, along with the full training recipe. Any government, researcher, or startup can build upon them.
Professor Anil Madhavapeddy, who leads the work, emphasizes their goal: "democratize access to planetary-scale environmental monitoring."
Uses are practical and far-reaching:
- Farmers: Track crop health and forecast yields at 10-meter resolution, benefiting smallholders in developing regions.
- Conservationists: Monitor habitat shifts, from tropical forests to UK hedgerows that shelter hedgehogs.
- Energy planners: Map solar and wind sites to guide the transition.
All of it runs off the same shared fingerprints.
TESSERA is a small story with a significant message: A public university built a frontier-scale AI model, made it free, and did it on hardware everyone says you cannot use.