Amazon Web Services (AWS) and Nvidia have announced a major expansion of their strategic partnership, with plans to deploy an additional 2 million high-end Nvidia GPUs across AWS's global infrastructure between 2027 and 2028. Combined with the 1 million GPU procurement plan announced in March, AWS's total commitment to Nvidia GPUs now reaches 3 million units.

The new order spans three generations of Nvidia's flagship products, including the current high-performance Blackwell Ultra, the next-generation Rubin platform, and the premium Rubin Ultra. Based on a list price of tens of thousands of dollars per unit, the contract for the 2 million GPUs is valued at tens of billions of dollars. Among these, 100,000 units will be specifically allocated to "AI factories" built for the U.S. government.

The collaboration has evolved from a simple GPU procurement arrangement into a full-stack technology partnership. AWS will deploy Nvidia's Vera CPU, purpose-built for agentic AI, for the first time, while deepening integration of NVLink Fusion high-speed interconnect and NVHBM high-efficiency memory technologies. The scope of cooperation has also expanded to encompass AI factories, open-source models, data processing, and robotics. Nvidia's Nemotron series of open models is now available on Amazon Bedrock and SageMaker, while Amazon's robotics division will leverage Nvidia's Omniverse and Isaac platforms to develop next-generation intelligent systems.

Notably, while scaling up its investment in Nvidia GPUs, AWS continues to advance its proprietary Trainium AI chips in parallel. Amazon emphasizes that this approach is designed to offer customers flexible compute options, enabling different workloads to be matched with the optimal hardware. This "dual-track" strategy reflects both the reality that current AI compute demand still outpaces the capacity of in-house alternatives, and signals that cloud providers will continue to grapple with the long-term trade-off between chip self-sufficiency and performance optimization.

CONEVO IC Supplier

CONEVO is a global IC distributor with Shenzhen/Hong Kong cloud warehouses, delivering authentic FPGA, MCU, power ICs worldwide with verified quality and spot inventory for OEM and EMS partners. Daily Featured IC Recommendations:

● MPQ6411GS-AEC1 — MPS automotive-grade voltage supervisor in 8-SOIC, AEC-Q100 qualified with -40°C to +125°C range.

● ADT7475ARQZ-RL7 — Precision temperature monitor and fan controller from Analog Devices, enabling intelligent thermal management for servers, networking gear, and high-performance computing platforms.

● GD25LQ32DSIGR — GigaDevice 32Mb low-power SPI NOR Flash in SOP-8, operating at 1.65V–2V with 120 MHz Quad-SPI speed and industrial -40°C to +85°C tolerance.

Website: www.conevoelec.com

Email: info@conevoelec.com