NVIDIA and Amazon are expanding their long-term collaboration on artificial intelligence infrastructure. Under the new agreement, AWS will be able to deploy up to 3 million NVIDIA AI accelerators globally for agent- and entity-based AI, automation and other computing workloads.
Initially, the two companies agreed to supply 1 million GPUs, but will now add another 2 million. Amazon hopes to supplement its Trainium accelerator with NVIDIA solutions to provide AWS customers with more hardware options for AI computing. At the same time, companies are working to integrate Vera processors into Amazon’s cloud platform, which is especially important given the growing demand for central processing units in agent-based artificial intelligence systems.
NVIDIA acquires an artificial intelligence startup as a backup plan if it stops buying GPUs
Another area of collaboration will be the use of NVHBM technology. It will improve memory efficiency on systems with Trainium accelerators: NVHBM delivers 30% more bandwidth and 15% more energy efficiency than HBM4E, according to NVIDIA. As a result, the partnership no longer only covers GPUs, but also covers AI infrastructure components such as CPUs and memories.
In addition, Amazon has also allocated to the US public sector: 100,000 of the total GPUs are used in special “artificial intelligence factories” designed to handle government tasks affecting Level 6 security levels. NVIDIA CEO Jensen Huang noted that demand for the company’s computing platform in AWS is higher than initially forecast, so the collaboration is now being expanded across the technology stack.
