Amazon and Nvidia Corp. have expanded their partnership with a deal to add another 2 million Nvidia Corp. GPU chips to Amazon Web Services data centers, tripling the original commitment made five months earlier2.
The announcement was made during Nvidia Corp.'s quarterly earnings call on Wednesday. The original agreement, struck roughly five months prior, called for more than 1 million Nvidia Corp. GPUs to be deployed across AWS infrastructure starting this year. Nvidia Corp. said in a statement that since then, "demand has exceeded those expectations".
The new chips include Nvidia Corp. Blackwell Ultra, Rubin, and Rubin Ultra GPUs, with deliveries to AWS data centers scheduled for 2027 and 2028. Neither company shared financial terms, though TechCrunch reported that, considering GPU unit costs, the deal is worth tens of billions of dollars.
The expanded partnership goes beyond chip procurement. Nvidia Corp. said its networking hardware, which connects thousands of GPUs into a single system, along with its open models, CPUs, data processing software, and robotics platform, will also be integrated across AWS.
The companies cited "surging demand" from startups, enterprises, AI labs, and governments as the driver behind the deeper collaboration. The deal is proceeding even as Amazon invests in its own potentially competing AI chips.
ANALYSIS The speed of the escalation, from 1 million GPUs to a tripled order in five months, quantifies the pace at which hyperscaler compute demand is outrunning initial procurement plans. The breadth of the integration, spanning networking, CPUs, software, and robotics alongside GPUs, ties AWS more deeply into Nvidia Corp.'s full-stack ecosystem rather than treating the relationship as a pure chip supply arrangement. That Amazon is simultaneously scaling Nvidia Corp. purchases and developing its own AI silicon underscores the gap between in-house chip timelines and the immediate compute pressure from cloud customers.
Vector Wire previously covered Nvidia Corp.'s AI Dynamo project releasing v1.4.0 with Qwen 3.8-2.4T serving support on GB300 and GB200 hardware ctx, part of the broader Blackwell-generation infrastructure that the Amazon deal will extend.