Amazon has secured $225 billion in commitments for its custom silicon business, primarily linked to Trainium chips.
During the Q1 2026 earnings call on April 29, CEO Andy Jassy revealed that the annual revenue from this segment could reach $20 billion, with potential growth to $50 billion if it operated independently. The demand is bolstered by major clients including Anthropic, which has a substantial multi-year agreement tied to over $100 billion in AWS expenditure. OpenAI has also committed to significant Trainium capacity, while Uber and Meta are adopting both Trainium and Graviton technologies.
Trainium2 is nearly sold out, and Trainium3, which began shipping earlier this year, is almost fully subscribed. Substantial pre-reservations for Trainium4 indicate strong future demand projected for broad availability in about 18 months. The competitive edge lies in Trainium's price-performance, which exceeds that of rival GPUs by around 30% with Trainium2 and 30-40% with Trainium3.
Amazon is reportedly considering direct sales of Trainium chips and server racks to external data centers, marking a strategic shift from internal infrastructure to external market sales. This move could pose a significant challenge to Nvidia, which currently leads the AI compute sector with its GPUs.
Amazon's success in custom silicon reflects a shift among major tech companies, with competitors like Google and Microsoft also developing their own chip solutions, as seen with their respective TPU and Maia AI accelerator projects. This trend indicates a growing preference for hyperscalers to create in-house solutions rather than solely rely on third-party chip manufacturers.
This material is informational and not financial advice.
