AWS to Deploy 2 Million Additional NVIDIA GPUs by 2028
Amazon AWS has announced plans to deploy an additional 2 million NVIDIA GPUs between 2027 and 2028, increasing its total planned deployment to 3 million units. The cloud provider will also integrate its proprietary chips with NVIDIA's NVLink Fusion and custom NVHBM technologies. This massive expansion highlights the escalating demand for AI hardware and AWS's commitment to maintaining its lead in cloud infrastructure. By integrating custom chips with NVIDIA's ecosystem, AWS aims to optimize performance and energy efficiency for next-generation AI workloads. The announcement includes integration with NVIDIA's new NVHBM (custom high-bandwidth memory) and NVLink Fusion, which allows third-party silicon to connect to NVIDIA Vera CPUs. However, the original report contains technical inconsistencies, such as misattributing Groq's LPU to NVIDIA and referencing a future "GTC 2026" event.
## BACKGROUND
NVLink Fusion is NVIDIA's platform that opens its proprietary interconnect fabric to third-party accelerators and custom silicon. NVHBM is NVIDIA's custom high-bandwidth memory technology designed to offer 30% higher bandwidth and 15% lower power consumption compared to standard HBM4e. Groq is an independent AI chip startup known for its Language Processing Unit (LPU) architecture, which is separate from NVIDIA.