Amazon triples its Nvidia GPU order and folds Nvidia's models and robots into AWS

On Wednesday, Amazon and Nvidia announced an expanded partnership that adds another 2 million Nvidia GPUs to Amazon Web Services data centers, just five months after Amazon committed to deploying more than a million Nvidia GPUs starting this year. The chips span the Blackwell Ultra, Rubin, and Rubin Ultra lines, and they head to AWS in 2027 and 2028. The companies cited "surging demand" from startups, enterprises, AI labs, and even governments, and Nvidia added that demand had "exceeded those expectations."

Neither company shared financial terms. At GPU unit pricing, the extra 2 million chips puts the deal in the tens of billions of dollars. The announcement came during Nvidia's quarterly earnings call.

The order is the visible part; the integration is the point

The expansion goes beyond buying chips. Nvidia's networking hardware, which connects thousands of GPUs into a single system, along with its open models, CPUs, data processing software, and robotics platform, will be integrated across AWS.

On the model side, AWS will serve Nvidia's Nemotron family of open models on Bedrock and SageMaker. On the robotics side, Amazon plans to adopt Nvidia's full physical AI stack to run its warehouse robot fleet, per Nvidia CFO Colette Kress. The stack covers Omniverse, Cosmos, Isaac, and Jetson. Nvidia also introduced a new version of Jetson this week, aimed at entry-level edge AI.

Kress said Nvidia will also send an unspecified number of Vera CPUs to Amazon, some integrated with Rubin and others standalone. Separately, she said Nvidia expects Vera to be deployed by "every major hyperscaler, neocloud, AI lab, and system OEM," with shipments already underway to lead partners that include Oracle and SpaceXAI. Nvidia CEO Jensen Huang claimed in May that Vera opens a "brand-new $200 billion TAM."

Amazon keeps building its own chips anyway

The deal deepens at the same time Amazon is ramping up competing silicon. AWS is in talks to sell its Trainium chips to other companies, per Amazon's AI chief Peter DeSantis. Trainium is a direct alternative to Nvidia's H100 and Blackwell for deep learning workloads. Graviton, its Arm-built CPU, targets the traditional server chips from Intel and AMD.

Amazon has said its custom chip business crossed a $25 billion annualized revenue run rate, driven by $225 billion in total commitments from AI labs like Anthropic and OpenAI, according to the company. That growth did not stop it from taking another 2 million Nvidia GPUs.

The quarter behind the announcement

Nvidia reported $96.2 billion in second-quarter revenue, beating analyst estimates. Data center revenue accounted for $89 billion, up 117% year over year. The company guided third-quarter revenue to $108 billion, some of it from its next-gen Rubin GPUs, with production shipments beginning this quarter.

It has committed $279 billion to secure supply and manufacturing capacity for current and future data center projects, up substantially from $119 billion the previous quarter. That includes $92 billion in projected spending for the rest of the fiscal year and $87 billion in fiscal 2028.

"The thing that matters for the industry is that AI is now doing productive and useful work," Huang said during the call. "If we had more compute, we could generate more profitable tokens." Investors will be watching whether that extra compute converts into profit as quickly as the money is flowing into infrastructure.