Amazon and Nvidia have widened a supply and integration deal so Amazon Web Services will add 2 million GPUs to its data‑centre fleet and fold more of Nvidia’s software and robotics technology into AWS.
The chips Amazon will receive include Blackwell Ultra, Rubin and Rubin Ultra GPUs, and they are due to arrive in 2027 and 2028. Nvidia said parts of its stack, from networking hardware that links many GPUs together to model tooling and robotics platforms such as Omniverse, Isaac and Jetson, will be integrated across AWS, while AWS will serve Nvidia’s Nemotron family of open models on Bedrock and SageMaker, giving customers both more raw compute and direct access to Nvidia software and models on AWS. For Amazon this buys immediate scale to meet what the companies called surging demand; for Nvidia it extends distribution of next‑generation accelerators and its broader AI stack into the largest cloud provider.
The enlarged order follows an earlier commitment that saw Amazon agree to deploy more than 1 million Nvidia GPUs beginning this year, taking the total committed to well over three million by 2027-28. Neither company shared financial terms, though the report noted the volume implies a deal worth at least tens of billions of dollars given GPU unit costs, and Nvidia reported sales of $96.2 billion for the quarter with data‑centre revenue of $89 billion. Nvidia has also set aside committed $279 billion to secure supply and manufacturing capacity; Amazon, meanwhile, said its custom chip business has grown, it has crossed a $25 billion annualized revenue run rate, backed by $225 billion in commitments from AI labs, even as it continues to develop its Trainium and Graviton processors.
Nvidia added that it will ship Vera CPUs, some paired with Rubin accelerators and some standalone, to hyperscalers and system OEMs, with shipments already underway to lead partners such as Oracle and SpaceXAI, signalling more hardware integrations beyond AWS are already in motion. The next milestones to watch are the start of large‑scale AWS deployments in 2027 and 2028 and the rollout of Nvidia’s software and robotics platforms inside Amazon’s operations and customer services. Jensen Huang summed the seller’s case succinctly: “AI is now doing productive and useful work,” he said, framing the rush for more compute as demand for profitable AI tasks.
For customers this means more Nvidia‑powered capacity and Nvidia software available through AWS in the coming years; for Amazon it buys time and scale while its own silicon business grows.
