This month Google, which pioneered the “transformer” algorithms underpinning the current AI wave, launched Gemini 3, a cutting-edge model that outperforms those of its biggest rivals, including OpenAI, on most benchmarks. Crucially, Gemini 3 was trained entirely on Google’s own chips, called tensor-processing units (TPUs), which it has begun peddling to others as a far cheaper alternative to Nvidia’s graphics-processing units (GPUs).
