Bezos: Amazon's custom Trainium and Graviton chips could become next major business pillar

Jeff Bezos says Amazon's custom silicon operation, spanning Trainium AI accelerators and Graviton processors, is poised to become the company's next major business pillar. Speaking to Fortune, Bezos compared its potential to Marketplace, Prime, and AWS, noting that these chips are central to Amazon's strategy to reduce reliance on Nvidia while controlling more of the AI boom's economics. AWS has positioned Trainium as a lower-cost alternative for training and inference, with Anthropic already using it to run Claude models and OpenAI committing to 2 gigawatts of capacity from 2027. Amazon CEO Andy Jassy reinforced the bet, calling chips "the key to the compute" and highlighting a $15B annual AI revenue run rate. With $200B in capital spending this year — part of over $700B across hyperscalers — Amazon is betting demand will outlast fears of a bubble.

Jassy also hinted that Amazon may eventually sell racks of its custom chips directly to third parties, following Google's TPU model with Broadcom. This would broaden AWS's chip business beyond internal use and potentially undercut Nvidia's pricing. The move reflects a broader push by cloud giants to own the full stack of AI infrastructure — from silicon to services. For enterprise customers, more competition could mean lower costs and greater flexibility in choosing compute for specific workloads. However, the enormous capital outlay has sparked debate about whether AI demand can sustain such spending. Bezos remains confident, arguing that AI will transform every customer experience and that Amazon's decade-long chip investment is just beginning to pay off.