AWS and Nvidia to deploy 2 million more GPUs by 2028

AWS and Nvidia said Wednesday they will deploy 2 million additional Nvidia GPUs across AWS's global infrastructure in 2027 and 2028, expanding an AI computing partnership that both companies say has been overtaken by demand faster than either expected.

The new commitment covers Nvidia's Blackwell Ultra, Rubin and Rubin Ultra chips. It builds on a pledge of more than 1 million GPUs for 2026 that AWS made only five months ago, at Nvidia's GTC developer conference; that earlier allocation, the companies said, was consumed faster than projected. With the new addition, AWS's publicly disclosed Nvidia GPU commitment now totals more than 3 million units across roughly three years.

Part of the expansion is earmarked for government work: 100,000 of the GPUs are designated for US federal and national-security workloads on AWS infrastructure classified at Impact Level 6 and above, the tier used for the most sensitive government data. The companies did not name a specific agency.

The deal also brings new Nvidia hardware to AWS beyond raw GPU count. Nvidia's next-generation Vera CPUs are coming to AWS infrastructure, alongside NVLink Fusion with custom high-bandwidth memory support and Nvidia Spectrum networking optimized for large-scale AI training. AWS said its new G7 instances deliver 4.6 times the AI inference performance and 2.1 times the graphics performance of the prior G6 generation, with GPU-accelerated data processing running 3.7 times faster at 30 percent better price-performance, and vector indexing running 9 times faster at a quarter of the cost.

"NVIDIA and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast," Nvidia CEO Jensen Huang said in a statement. AWS CEO Matt Garman said customers "want the freedom to choose the best tools for their AI workloads, and they want confidence that everything works seamlessly together." The companies describe their partnership, which spans 16 years, as extending beyond GPUs into custom silicon, networking, open models, data processing and robotics.

No financial terms were disclosed for the new capacity.