What happened
Andy Jassy, Amazon's chief executive, said the company's chip business is tripling annually, according to a report Wednesday from Crypto Briefing. The comment framed Amazon's custom silicon effort, built around the Trainium training chip and the Inferentia inference line, as one of the fastest-scaling pieces inside AWS. Jassy didn't give a dollar figure in the excerpt, but a 3x year-over-year run rate puts Amazon's internal silicon on a growth curve that rivals the pace Nvidia has posted on its data-center segment through the current AI cycle.
The remarks were framed as a direct challenge to the incumbent chip supply chain, with Crypto Briefing noting the shift reshapes 'the semiconductor landscape' and tightens access to AI development capacity.
Why it matters
Crypto doesn't trade Amazon stock, but it trades the AI compute thesis. Tokens tied to GPU marketplaces, decentralized training, and inference networks have spent 2026 pricing off the same scarcity story: Nvidia controls the high end, hyperscalers hoard the supply, and anyone outside that stack pays a premium or waits in line. A tripling AWS silicon business is a crack in that story.
If Amazon can route a growing share of its own AI workloads onto Trainium and Inferentia, the marginal buyer of Nvidia H100s and B200s shrinks, and the compute spread that decentralized networks arbitrage narrows. That's a double-edged read. Bulls on the AI-crypto basket argue cheaper compute broadens the addressable market for on-chain inference.
Bears argue custom hyperscaler silicon kills the pricing wedge that made decentralized GPU rentals look cheap in the first place.