Cambridge built a planet-scale AI model, and skipped Nvidia to do it
What changed
A University of Cambridge team built a planet-scale AI foundation model called TESSERA without using Nvidia chips, relying entirely on AMD hardware. The model covers Earth-scale data for environmental observations and operates at a comparable scale to large language models. This breaks the near-monopoly Nvidia has held on AI training and inference infrastructure.
Why builders should care
Nvidia’s dominance in AI chip supply has shaped development choices, often forcing teams into its software ecosystem and hardware limitations. TESSERA’s success proves that building large-scale AI models without Nvidia silicon and CUDA frameworks is possible. This opens the door for more competition and flexibility in hardware sourcing, which could reduce costs and vendor lock-in for AI projects.
The practical takeaway
Shifting AI infrastructure away from Nvidia can lower expenses and introduce diverse optimization strategies. Organizations hesitant to rely solely on Nvidia chips due to supply chain risks, pricing, or proprietary software dependencies now have a concrete example showing AMD-based setups can handle massive AI workloads. Builders should consider hardware diversity in their scaling plans to avoid bottlenecks and strengthen supply chain resilience.
What to watch next
It will be important to see if AMD’s architecture gains broader AI adoption beyond specialized academic labs. Watch for ecosystem tools that simplify AMD GPU use at scale. Also, note how cloud providers adapt to this hardware mix and whether AMD chips start displacing Nvidia in AI infrastructure deals. The Cambridge model’s real-world performance and cost data could further pressure chip vendors to innovate and compete on price-performance.
AI Quick Briefs Editorial Desk
