DeepSeek's 2K-GPU model may beat Google's best, raising questions about GPU efficiency and Google's strategy

teortaxesTex · x · 2026-08-06

Three years ago, Dylan Patel introduced the notion of 'GPU-Poors', illustrated by Google Gemini 'eating the world'. Now, DeepSeek's latest model, trained on about 2K GPUs, is likely better than Google's best, and Google DeepMind seems to be falling apart. This raises questions about compute efficiency and model capability.

Original post →

More from Companies & People

Companies & People channel →