Amazon and Nvidia Expand Partnership to Supply Two Million Artificial Intelligence Chips
Seattle, Thursday, 27 August 2026.
Amazon Web Services and Nvidia have expanded their partnership to supply two million next-generation processors by 2028, including 100,000 chips dedicated to secure national security government computing facilities.
Strategic Expansion of AI Infrastructure
Amazon Web Services (AWS), a subsidiary of Amazon.com, Inc. (NASDAQ: AMZN), and Nvidia Corporation (NASDAQ: NVDA) officially announced a significant expansion of their strategic collaboration on 26 August 2026 [1][2]. This agreement aims to deploy an additional 2 million next-generation GPUs across AWS global infrastructure to address accelerating demand for high-performance computing capabilities [1][7]. The initiative is designed to support agentic and physical artificial intelligence workloads, marking a substantial increase from previous deployment targets set earlier in the year [1][4].
Deployment Timeline and Scale
The deployment of these 2 million GPUs is scheduled to occur during the 2027-2028 fiscal period, reflecting a long-term infrastructure commitment [1][2]. This timeline follows an earlier announcement at NVIDIA GTC 2026 regarding the addition of over 1 million GPUs starting in 2026, a target that has since been exceeded by customer demand [2][7]. The expansion underscores the rapid pace at which enterprise and government entities are adopting AI technologies, necessitating a scaled response from cloud infrastructure providers [1][4].
Government AI Factories and Security
A critical component of this partnership involves the construction of AI factories specifically for the U.S. government, incorporating 100,000 GPUs on secure AWS infrastructure [1][2]. These facilities are designed to handle federal and national-security workloads, supporting classifications at Impact Level 6 (IL6) and above [2][7]. This secure infrastructure ensures that sensitive government data remains protected while leveraging the high-performance computing power required for advanced national security applications [1][7].
Technical Integration and Hardware
Technical integration efforts include bringing NVIDIA Vera CPU-based infrastructure to AWS and extending NVIDIA NVLink Fusion with custom high-bandwidth memory (NVHBM) [1][4]. The collaboration will also deploy NVIDIA Blackwell Ultra, Rubin, and Rubin Ultra GPUs, alongside NVIDIA RTX PRO™ 4500 Blackwell Server Edition GPUs for Amazon EC2 G7 instances [1][2]. These hardware upgrades are intended to optimize performance across networking, security, and deployment layers of the cloud infrastructure [1][7].
Performance Metrics and Efficiency
Recent benchmarking results indicate that Apache Spark workloads on Amazon EC2 G7 GPU instances perform up to 3.7 times faster than comparable CPU-based instances without requiring code modifications [3][4]. In TPC-DS 3 TB benchmark testing, GPU-accelerated clusters completed processing in 4.7 minutes, achieving a reduction in job run times by more than two-thirds compared to CPU baselines [3][4]. This performance gain is critical for enterprises managing large-scale data processing and machine learning feature engineering tasks [3][4].
Economic Implications and Cost Savings
Despite a higher hourly rate for GPU instances, the increased processing speed results in significant cost efficiencies for specific workloads [3][4]. GPU-accelerated clusters cost approximately $2.06 per run compared to $2.93–$3.18 for CPU clusters, representing a potential cost reduction calculable by 29.693 percent per run [3][4]. This economic advantage, combined with the performance uplift, drives the business case for migrating compute-heavy workloads to GPU-accelerated infrastructure [3][4].
Leadership Perspectives on Partnership
Matt Garman, CEO of AWS, stated that customers want confidence that tools work seamlessly together, justifying the deep investment in NVIDIA technologies [1][2]. Jensen Huang, founder and CEO of NVIDIA, noted that demand is running ahead of every forecast, necessitating this expansion across the full stack of GPUs, CPUs, networking, and software [1][7]. The partnership, now spanning 16 years, continues to evolve to meet the unprecedented scale required for agentic and physical AI [1][7].