By Tom Whitfield

Amazon plans to sell its own AI chips, taking a more direct aim at Nvidia's dominant position in the market for accelerators. The move, reported by TechCrunch, signals a shift from using its silicon strictly inside Amazon Web Services to offering it as a product in its own right.

The decision matters because Nvidia currently controls the merchant market for AI training and inference chips. Cloud providers buy large volumes of Nvidia GPUs to rent out to customers. If Amazon sells its own accelerators, it gives buyers an alternative and puts pricing pressure on the incumbent.

What changes commercially?

Amazon has already designed custom silicon for its data centers. Until now, the commercial logic was internal: build chips that lower Amazon's own cost per unit of AI compute. Selling those chips outside changes the equation.

The buyer pool widens from one company to many. That spreads design and manufacturing costs across more volume, and it turns a cost-saving tool into a revenue line.

Why take on Nvidia directly?

Nvidia's grip on AI accelerators has left cloud customers with few alternatives and high bills. Every hyperscaler has responded with custom silicon of some kind. Amazon's step toward outside sales is the most aggressive form of that response.

The mechanism is straightforward. A chip designed in-house avoids the margin Nvidia charges, and AWS can price access to it below GPU-based instances while protecting its own gross margin. Direct sales would extend that leverage beyond Amazon's walls.

The open question is software. Nvidia's CUDA ecosystem keeps customers locked in, and any challenger must prove that its toolchain can run the same workloads without painful migration.

The takeaway: Amazon is moving its chip effort from a private cost play to a public contest for the AI accelerator market.