Amazon is adding two million Nvidia Blackwell Ultra, Rubin, and Rubin Ultra GPUs to AWS data centre infrastructure across 2027 and 2028 — tripling a commitment made five months ago to deploy over one million Nvidia GPUs starting in 2026. Nvidia's chief financial officer confirmed the expanded deal is valued in "tens of billions of dollars," with Vera CPU deployments beginning in the third quarter. The scale of the commitment is notable for what it reveals about hyperscaler demand projections: Amazon has both the financial incentive and the technical capability to reduce Nvidia dependency through its own Trainium chips, yet it is simultaneously placing orders at a rate that materially exceeds its own silicon development timeline. The implication is that AI workload growth on AWS is outpacing every compute source available to the company — Nvidia, Trainium, and custom silicon combined. Jensen Huang's description of the dynamic — "AI is generating profitable tokens; if we had more compute, we could generate more profitable tokens" — applies, evidently, to Amazon's customers as much as to any other class of AI operator.
LLMs
Google Releases Gemini 4 Argon — Tops Frontier Benchmarks and Targets Autonomous Cybersecurity
30 September: Google's Gemini 4 Argon leads the Vals AI benchmark index over OpenAI GPT-6 Astra and …