Amazon has committed to deploying 2 million additional Nvidia GPUs across its AWS infrastructure by the end of 2028, the companies announced Wednesday. The news lifted Nvidia shares by about 4% in after-hours trading, while Amazon climbed roughly 0.7%.

Amazon.com, Inc. (AMZN)

The new deployment is scheduled to roll out between 2027 and 2028. It builds on Amazon’s March commitment to acquire 1 million Nvidia chips, which are currently being integrated into AWS facilities this year.

Wednesday’s agreement covers Nvidia’s Blackwell Ultra, Rubin, and Rubin Ultra GPU architectures. The chips are designed for AI model training and inference, making them central to the computing demand behind large language models and other AI workloads. Individual units are typically priced at more than tens of thousands of dollars, although large enterprise buyers generally receive volume discounts.

Neither company disclosed financial terms of the deal.

During Nvidia’s quarterly earnings call on Wednesday afternoon, CFO Colette Kress confirmed the expanded collaboration. She said Amazon’s cloud division “is deploying an additional two million GPUs starting this quarter through the second of FY29.”

Nvidia’s Physical AI Platform Coming to Amazon Warehouses

Kress also said Amazon plans to deploy Nvidia’s full physical AI platform across its warehouse robotics operations. The move broadens the partnership beyond conventional data center computing and into automation systems that support Amazon’s logistics network.

Nvidia CEO Jensen Huang addressed the agreement in a CNBC interview Wednesday night. “This is going to be a great partnership,” he said.

Amazon continues to invest heavily in Nvidia hardware even as it develops its own semiconductor products through Annapurna Labs. Those efforts include custom CPUs and the Trainium AI accelerator, which is intended to compete with Nvidia’s chips. Industry reports have also suggested that the next Trainium version will use Nvidia networking components.

Major Cloud Providers Still Depend on Nvidia

Microsoft and Alphabet’s Google are also developing in-house chip programs, but both companies still maintain significant Nvidia purchases. Amazon is taking a similar approach.

Large cloud infrastructure providers say proprietary chips can be optimized for specific workloads. Even so, Nvidia’s GPUs remain central to AI model training across the industry, which helps explain why the company continues to secure large orders even as customers build their own silicon.

With Wednesday’s announcement added to the March order, Amazon’s total Nvidia GPU commitment now stands at 3 million processors.

The announcement came during Nvidia’s quarterly earnings presentation, a timing that drew extra investor attention as both companies publicly highlighted the strengthened collaboration.