Amazon Web Services has announced plans to deploy approximately one million Nvidia GPUs as part of a broad expansion of its AI infrastructure. An Nvidia executive confirmed to Reuters on Thursday that the rollout is expected to run through the end of 2027. The deployment will begin this year across AWS’s global cloud regions and will be accompanied by expanded collaboration on networking and related infrastructure.
AWS described the initiative as part of its work on agentic AI systems — platforms designed to reason, plan, and act autonomously across complex workflows. The scale of the commitment reflects a broader shift in how cloud providers are building out their underlying compute layers. The deal also points to changes in where demand is concentrating within the AI technology stack.
Analysts note that inference — the process of running trained AI models in real time rather than retraining them — now accounts for roughly two-thirds of AI compute demand, up from about one-third in 2023. Dermot McGrath, co-founder of strategy and growth studio ZenGen Labs, told Decrypt that the chips involved in the deal are oriented toward running AI models at scale with a focus on reducing operational costs. He cited Deloitte estimates projecting the market for inference-focused chips will exceed $50 billion by 2026.
McGrath characterized Nvidia’s evolving role as something larger than a conventional chip supplier. “Nvidia is becoming the infrastructure layer underneath the cloud providers, not just a chip vendor to them,” he said. He added that Nvidia is now extending its influence through networking and rack architecture, mirroring how software platforms have historically built competitive advantages through programming models. AWS’s ability to combine Nvidia hardware with its own chips in the same systems, he noted, gives customers more flexibility than some competing platforms.
Pichapen Prateepavanich, policy strategist and founder of infrastructure firm Gather Beyond, told Decrypt that demand for inference is driving long-term compute commitments and deepening ties between cloud providers and chipmakers. “Cloud providers want independence over the long term, but in the near term they need Nvidia to remain competitive,” she said. She described the resulting dynamic as one where cooperation and competition occur simultaneously.
Berna Misa, deal partner at AI-led investment fund Boardy Ventures, called the development an “infrastructure flip,” arguing that Nvidia is embedding its full stack — spanning compute, networking, and inference — inside AWS data centers that previously ran proprietary equipment. Although AWS continues to develop its own AI chips for both training and inference, Misa said this does not fundamentally alter the current dependency, since inference relies on multiple components across the stack and Nvidia supplies most of them. “When you’re that deep in your customer’s stack, switching cost and the context layer that comes out of it becomes the moat,” she said.
The announcement arrives as the deal’s broader context includes legal and geopolitical pressures on Nvidia. U.S. prosecutors are pursuing a case alleging that Nvidia chips were smuggled to China, placing the company’s global supply controls under renewed scrutiny. Since 2022, Nvidia’s most advanced chips have been subject to tight export restrictions as part of a U.S. strategy to limit China’s progress in advanced computing and AI. Observers suggest the AWS agreement could further widen the technological gap between U.S. and Chinese AI capabilities.
Originally reported by Decrypt.
