AWS’s plan for two million additional NVIDIA GPUs raises the facility delivery question
AWS plans to deploy two million additional NVIDIA GPUs during 2027 and 2028 across its global infrastructure, including AI factories.
Infraday Coverage · AI Infrastructure
AWS plans to deploy two million additional NVIDIA GPUs during 2027 and 2028 across its global infrastructure, including AI factories.
The announced expansion includes Blackwell Ultra, Rubin and Rubin Ultra systems. AWS describes it as additional to an earlier plan to begin adding more than one million NVIDIA GPUs in 2026. These are deployment plans, rather than evidence that the full additional fleet is installed.
A hardware target needs a facility schedule
A chip count is useful for understanding the scale of a computing commitment, but it cannot independently establish the electrical capacity or construction scope required. Those depend on the installed system, utilization and facility design.
Infraday perspective: Owners should connect hardware orders to site readiness, electrical delivery, cooling acceptance and network testing. A single integrated schedule can expose gaps that separate equipment and building calendars conceal.
For utilities and suppliers, phased deployment information is especially valuable. A multi-year commitment may translate into several local projects with different startup dates. Regional infrastructure planning needs that geographic and temporal detail before treating a global announcement as a local load forecast.
What to watch
Follow site-level deployment disclosures and commissioning milestones. Avoid converting GPU totals into megawatts without a disclosed configuration and a defensible design basis.
Source: AWS and NVIDIA, August 2026. Reviewed October 1, 2026.
AI-generated editorial illustration; not a photograph or rendering of the specific project.