Amazon in Talks to Sell Custom AI Chips to External Customers

Amazon expands AI chip strategy beyond internal use

Amazon is negotiating to sell its proprietary artificial intelligence chips to external customers for deployment in their own data centers, according to the company's AI chief Peter DeSantis. The move represents a significant strategic shift for the cloud computing leader, which has until now used its custom silicon exclusively within its Amazon Web Services infrastructure.

DeSantis confirmed that discussions are underway but declined to identify potential customers, Bloomberg reported. The chips in question include Amazon's Trainium line of processors, designed specifically for training AI models.

Why it matters

This expansion puts Amazon in direct competition with Nvidia in the broader semiconductor market, not just within cloud services. If successful, it could give enterprises an alternative to Nvidia's GPUs for AI workloads while allowing Amazon to monetize years of chip development investment. The move also signals that major cloud providers see custom silicon as a viable business beyond cost optimization for their own operations.

Challenge to Nvidia's market position

Nvidia currently dominates the AI accelerator market, with its GPUs powering the majority of large-scale machine learning training and inference workloads. Amazon's entry as a chip vendor—rather than just a chip designer for internal use—introduces a new competitive dynamic.

The company has invested heavily in developing its own silicon alternatives, including the Trainium family for training and Inferentia chips for running AI models in production. By offering these processors to external customers, Amazon can potentially recoup development costs while providing enterprises with options that may offer better price-performance ratios for specific workloads.

Broader implications for cloud and chip markets

Amazon's decision to commercialize its custom chips follows a pattern among hyperscale cloud providers developing specialized hardware. Google has offered its Tensor Processing Units through its cloud platform, though primarily as a service rather than standalone products.

The success of this initiative will depend on whether Amazon can convince customers that its chips deliver sufficient performance advantages or cost savings to justify switching from established Nvidia solutions. It also raises questions about potential conflicts of interest, as Amazon would be selling chips that could be used in competing cloud environments.

For enterprises evaluating AI infrastructure investments, Amazon's entry as a chip vendor adds another option to consider alongside traditional semiconductor companies and other cloud providers' custom solutions.

These details were first reported by Bloomberg's Mark Bergen on June 18, 2026.