The Motley Fool published an analysis on May 8 with a blunt headline: "The Evidence Is Piling Up: Nvidia's AI Chip Dominance May Be About to Come to an End." The piece compiles disclosures that have been trickling out across the year and turns them into a single uncomfortable picture.
Amazon's chip business — built around the Trainium and Inferentia families — now runs at over $20 billion in annualized revenue, with 40% sequential growth in Q1 2026. Andy Jassy publicly disclosed $225 billion in Trainium revenue commitments earlier this quarter. The anchor customers driving that book are Anthropic, with up to 5 gigawatts of committed Trainium capacity, and OpenAI, with approximately 2 gigawatts. Apple is reportedly also a Trainium customer per TechCrunch's earlier lab tour.
The demand-side data is the part that should make Nvidia's roadmap planners uncomfortable. Trainium 2 is sold out. Trainium 3 is nearly fully subscribed despite only just starting to ship. And the customer-disclosed metric across the AWS Trainium 3 base is 50% lower training and inference cost compared to GPU alternatives. That's structural, not workload-specific.
Nvidia still holds approximately 90% of the AI accelerator market overall. Research firm Kearney projects that share dropping to 70% by 2030 as AMD, Intel, and the hyperscaler-designed chips grow into the gap. That is a 20-point market share decline projected over four years on a market growing 40% annually.
What to watch: whether AMD's Helios system, announced at CES 2026, lands large frontier-lab anchor customers in the next 60 days, and whether Google's TPU v8 disclosures match Trainium's 50% TCO claim. If two of the three custom-silicon stacks can match those economics, Nvidia's pricing power on the next chip generation gets re-rated by the market.