Capital signal

AWS expands NVIDIA GPU plan beyond 3M

AWS and NVIDIA confirmed that AWS will deploy 2 million additional GPUs in 2027 and 2028, taking its publicly announced NVIDIA GPU plan since 2026 to more than 3 million units.

By All We Need Editorial System

AWS and NVIDIA have expanded AWS's publicly disclosed NVIDIA GPU deployment plan beyond 3 million units. Following AWS's March commitment to add more than 1 million GPUs beginning in 2026, the companies now say AWS will deploy 2 million additional Blackwell Ultra, Rubin and Rubin Ultra GPUs in 2027 and 2028. This is a concrete multigenerational supply expansion, not a narrow instance update.

A plan growing from millions to more than three million

The AWS-NVIDIA joint announcement says AWS will deploy 2 million additional GPUs between 2027 and 2028, spanning Blackwell Ultra, Rubin and Rubin Ultra. In March, AWS said it planned to add more than 1 million NVIDIA GPUs in global cloud regions beginning in 2026, including Blackwell and Rubin systems. Taken together, the official disclosures put AWS's publicly announced NVIDIA GPU deployment path above 3 million units and extend it through 2028.

The commitment spans successive platforms

The significance is not only the number of GPUs but the inclusion of Rubin Ultra. By placing Blackwell Ultra, Rubin and Rubin Ultra in the same expansion plan, AWS is making infrastructure commitments across several product cycles rather than simply adding current-generation capacity. The companies tie the collaboration to agentic and physical AI, workloads that can require sustained inference throughput, low-latency networking and large-scale cluster scheduling. That makes long-term delivery capacity a more direct competitive variable for cloud providers.

Cloud supply competition moves to execution

For NVIDIA, the plan improves visibility into demand for future platforms. For AWS, multigeneration GPU capacity could enlarge the pool of compute it can sell as enterprise agent and physical-AI deployments grow. The strongest countercase is that the announcement describes planned deployments rather than installed capacity, with timing still dependent on data-center construction, power connections, server delivery and customer utilization. Even so, AWS has explicitly set the scale of infrastructure competition at a multiyear, multimillion-GPU level.

What to watch next

The next observable signals are AWS disclosures on regions, available instances, data-center commissioning dates and capital spending, along with NVIDIA commentary on Blackwell Ultra and Rubin demand from cloud customers. The case would strengthen if AWS begins reporting available capacity and customer deployments for agentic or physical AI. It would weaken if announced deployment milestones repeatedly move out.

Sources