On the same day NVIDIA reported record quarterly revenue of $96.2 billion, its largest customer Amazon Web Services provided the strongest validation for those numbers with what may be the largest GPU order in history. On August 26, AWS and NVIDIA jointly announced plans to deploy an additional 2 million NVIDIA GPUs across AWS’s global infrastructure in 2027–2028. This follows a commitment of 1 million GPUs made at GTC 2026 in March, bringing AWS’s total committed NVIDIA GPU deployment to 3 million.
What does 2 million chips mean? At a list price of tens of thousands of dollars per unit, the additional 200 million chips alone are valued conservatively in the tens of billions of dollars. NVIDIA CEO Jensen Huang stated: “Demand is running ahead of every forecast.”
Full-stack expansion: not just chips, but architecture
This is not just a GPU purchase but a full-stack technology lock-in. AWS will deploy NVIDIA Vera CPUs—high-performance processors purpose-built for agentic AI workloads—for the first time on its infrastructure. The companies also plan to build AI factories for the U.S. government, including 100,000 GPUs on secure AWS infrastructure rated for Impact Level 6 workloads.
Competition and collaboration: a dual-track strategy
This order comes as Amazon continues to invest heavily in its own Trainium AI chips, with its custom silicon business reportedly exceeding $25 billion in annualized revenue. Yet the 2-million-GPU order shows Trainium and NVIDIA GPUs are not an “either-or” choice. AWS’s strategy is to offer customers “freedom to choose”—NVIDIA for training, Trainium for inference, or a hybrid mix.
Increasing demand for its own chips while accelerating NVIDIA GPU purchases—AWS’s dual-track strategy reveals the true scale of AI compute demand: the market is large enough to accommodate both. When Huang says “demand is running ahead of every forecast,” he is referring to AI’s expansion from a training arms race to a full-spectrum, production-scale deployment. This deal also provides revenue visibility for NVIDIA through 2028—and NVIDIA has already locked in $279 billion in supply and capacity commitments.