Amazon's years-long effort to build a credible alternative to Nvidia's dominant AI chips is beginning to gain meaningful traction, according to reports. Anthropic and OpenAI — both of which have struck multibillion-dollar investment and infrastructure deals with Amazon — have already committed to renting large amounts of current and future Amazon custom silicon capacity, signalling a potential shift in the AI compute landscape.

The commitment taking shape

Amazon Web Services (AWS) has spent years developing custom AI accelerators, including its Inferentia chip for inference workloads and Trainium chip for training workloads, as part of a broader push to reduce the AI industry's near-total dependence on Nvidia hardware. The reported commitments from Anthropic and OpenAI represent a significant vote of confidence in that programme. Both companies operate at the frontier of large language model development and require massive, sustained compute capacity to train and serve their models.

Why it matters

For years, Nvidia's CUDA software ecosystem and GPU hardware have been the default infrastructure choice for AI training and inference at scale. Any meaningful adoption of Amazon's custom chips by top-tier AI labs would mark the first credible crack in that dominance. Hyperscale cloud providers including Amazon, Google, and Microsoft have all pursued custom silicon programmes precisely to gain cost control and supply-chain independence from Nvidia, whose chips have faced persistent shortages amid surging demand.

The competitive backdrop

Amazon announced a multibillion-dollar strategic investment in Anthropic in 2023, tied to expanded use of AWS infrastructure — a deal that reportedly included commitments to use Amazon's custom AI chips alongside standard cloud services. Anthropic, founded in 2021 by former OpenAI researchers including Dario and Daniela Amodei, develops the Claude series of large language models and has positioned AWS as a primary cloud partner. OpenAI, meanwhile, has maintained its primary infrastructure relationship with Microsoft Azure but has separately established cloud commitments with Amazon as well.

What's next

The broader industry will be watching whether Amazon's chip programme can scale to meet the demands of frontier AI workloads — not just in inference, where custom silicon has historically been more competitive, but in the computationally intensive training runs that define the capabilities of next-generation models. If Anthropic and OpenAI expand their reliance on Amazon's custom silicon, other AI developers and cloud customers could follow, accelerating the diversification of the AI chip supply chain. Nvidia's stock and long-term pricing power remain the most exposed variable to watch as these commitments mature.