AWS bought 1M Nvidia GPUs
Amazon confirmed a purchase of 1 million Nvidia GPUs through 2027 as hyperscalers lock into Blackwell-era infrastructure — a sign of heavy cloud demand for accelerated training and inference. That scale reshapes available capacity and will influence where startups and labs schedule large GPU jobs. (reuters.com)
Starting in 2026 AWS said it will add “more than 1 million” NVIDIA GPUs across its global cloud regions, explicitly naming both the Blackwell and Rubin GPU architectures. (aws.amazon.com) NVIDIA vice president Ian Buck told Reuters that shipments for the agreement will begin in 2026 and continue through 2027, providing the first public timing for the multi‑year supply. (finance.yahoo.com) AWS announced new EC2 options accelerated by NVIDIA’s RTX PRO 4500 Blackwell Server Edition and said those instances will be built on the AWS Nitro System to maximize host resource availability. (aws.amazon.com) NVIDIA’s product page for the RTX PRO 4500 Blackwell Server Edition lists the card as a single‑slot, 32GB accelerator aimed at data center and cloud deployments, matching AWS’s instance announcements. (nvidia.com) NVIDIA and AWS did not disclose financial terms for the supply arrangement, and the timing coincides with CEO Jensen Huang’s projection of a roughly $1 trillion revenue opportunity for the Rubin and Blackwell families through 2027. (money.usnews.com) (bloomberg.com) Datacenter analysis outlets reported AWS framed the expansion as part of deeper NVIDIA integrations across interconnect and model tooling, and separate trade coverage noted the plan to deploy the GPUs at hyperscaler scale within the coming 12 months. (datacenterdynamics.com)
