AWS Plans Another 2 Million Nvidia GPUs — And Plans Trainium NVLink Expansion
Amazon and Nvidia said AWS will add two million more Nvidia GPUs in 2027–2028 on top of the GTC 2026 million-plus commitment, bring Vera CPUs to AWS, and put 100,000 GPUs on IL6+ AWS for U.S. federal AI factories. No dollar terms. The interesting clause is Trainium on NVLink Fusion.
Hyperscaler GPU math is now millions per announcement, not tens of thousands. The interesting clause is Trainium on NVLink Fusion: AWS is buying Rubin and wiring its own ASIC into Nvidia’s rack fabric.
Amazon Web Services and NVIDIA said AWS will add two million more NVIDIA GPUs in 2027–2028 — Blackwell Ultra, Rubin, and Rubin Ultra — on top of the GTC 2026 pledge of >1 million starting 2026. Combined disclosed NVIDIA GPU commitment: >3 million. Vera CPUs come to AWS. Trainium will connect via NVLink Fusion and NVHBM. 100,000 GPUs land on IL6+ AWS for U.S. federal AI factories. RTX PRO 4500 Blackwell on EC2 G7: NVIDIA cites 4.6× inference and 2.1× graphics vs G6. Huang: demand is “running ahead of every forecast.” No dollar terms.
What was announced#
| Item | AWS + NVIDIA Aug 26 |
|---|---|
| Incremental GPUs | +2,000,000 in 2027–28 |
| Prior GTC | >1,000,000 starting 2026 |
| CPU | Vera on AWS |
| Gov | 100,000 GPUs, IL6+ |
| Custom | Trainium + NVLink Fusion + NVHBM |
The two-million-GPU commitment is incremental to the million-plus order announced at GTC 2026.
Limits#
- No dollar terms in the joint announcement.
- Whether the 100k government tranche is new IL6 capacity or a re-badge of existing US gov clusters is not specified.
- Nvidia’s 4.6× / 2.1× G7 claims are vendor figures.
Correction (October 14, 2026): Clarified that the additional GPUs and Trainium connectivity are planned deployments, not completed installations.