Andy Jassy's Q1 2026 earnings reveal a custom silicon empire already running at $20B annually, with ambitions to challenge Nvidia on its own turf.
Amazon just quietly became one of the most important chip companies on the planet, and most people missed the memo.
During the company’s Q1 2026 earnings call on April 29, CEO Andy Jassy disclosed that Amazon’s custom AI silicon business has already secured over $225 billion in multi-year revenue commitments. That number, tied almost entirely to Trainium chips, is not a forecast or a wish list. It is contracted demand.
The numbers behind the backlog #
The custom silicon unit, which spans Trainium AI accelerators, Graviton CPUs, and Nitro networking chips, is currently running at an annualized internal rate of approximately $20 billion. Jassy suggested that figure could climb to roughly $50 billion if the business operated as a standalone merchant chip company.
The demand driving that backlog is not abstract. Anthropic, which has a multi-year, multi-gigawatt deal tied to over $100 billion in AWS spend, is a major anchor customer. OpenAI has committed to approximately two gigawatts of Trainium capacity. Uber is also adopting both Trainium and Graviton, and Meta has committed to tens of millions of Graviton cores.
Trainium2 is already largely sold out. Trainium3, which began shipping in early 2026, is nearly fully subscribed. Pre-reservations for Trainium4 are already substantial, with broad availability expected in roughly 18 months.
The price-performance angle is central to why customers are signing. Trainium2 offers around 30% better price-performance than competing GPUs. Trainium3 pushes that advantage another 30-40%.
Amazon goes merchant: a direct challenge to Nvidia #
Amazon is reportedly exploring direct external sales of Trainium chips and full server racks to outside data centers. That would represent a significant strategic pivot. Until now, Trainium was effectively internal infrastructure powering AWS cloud services sold to customers. Selling the chips themselves, the way Nvidia does, is a different business entirely.
Nvidia currently dominates AI compute with its H100 and B200 GPU lines. Its data center revenue has become the defining story of the AI supercycle. But Nvidia’s margins, and its pricing power, depend on there being no credible alternative at scale. Amazon, with $225 billion in committed demand and three chip generations in rapid succession, is building exactly that alternative.
What this means for the broader AI infrastructure trade #
Amazon’s custom silicon success signals a wider trend: hyperscalers are increasingly unwilling to remain pure customers of third-party chip vendors. Google has Tensor Processing Units. Microsoft is developing its Maia AI accelerator. Meta is building its own training silicon. Amazon is now the furthest along in commercializing that bet, with a $225 billion backlog to prove it.
For the crypto and decentralized compute sector, this matters at the infrastructure level. If Amazon successfully drives down AI compute costs by 30-40% through custom silicon, the value proposition for decentralized alternatives narrows unless those networks can match that price-performance curve. Projects like Render and Akash have argued that centralized providers are too expensive and too restrictive. A cheaper, more efficient AWS powered by Trainium chips complicates that argument. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our