# Amazon Says Custom Chips Now Drive AWS Growth _Amazon said Graviton and Trainium commitments rose more than 20x as AWS grew 39.3%, with Graviton 5 pushing agentic AI workloads._ **Published:** 2026-10-30 **Source:** https://www.startuphub.ai/ai-news/technology/2026/amazon-says-custom-chips-now-drive-aws-growth --- [Amazon](https://www.aboutamazon.com/news/company-news/amazon-custom-chips-demand-q3-2026-earnings?utm_source=rss) said its custom silicon business kept its triple-digit growth streak on the third quarter call, as CEO Andy Jassy pointed to surging demand for Graviton and Trainium. Revenue commitments for each chip family increased more than 20x year over year, [Amazon](https://www.startuphub.ai/startups/amazon) said, and AWS grew 39.3% year over year for a sixth straight acceleration. Jassy, who has run Amazon since 2021 and built AWS before that, framed chips as a direct lever on that growth, not a side project. That last part matters. AI is now the tailwind for Graviton, not just Trainium. Jassy said post-training, reinforcement learning and agent tool use mostly run on CPUs rather than AI accelerators, and Graviton offers up to 30-40% better price-performance than alternatives, according to Amazon. Customers using the custom silicon saved more than $2 billion in Q3 alone, the company said. To reduce friction, Amazon just open-sourced TorchNeuron so teams can run existing PyTorch code on Trainium without a rewrite, a practical fix for the porting step where many migrations stall. The scale is showing up in capacity strain. After the launch of Graviton 5, backlog for prior-generation Graviton instances still more than doubled quarter over quarter, Amazon said, and short-term capacity for prior-generation Trainium instances is now selling at a premium to the first half of the year. Graviton 5 itself pushes the hardware further, with 192 cores, a 5x larger cache and up to 33% lower inter-core latency built for agentic workloads like real-time reasoning and multi-step orchestration, per [AWS](https://aws.amazon.com/ec2/graviton/). The related M9g instances built on Graviton deliver up to 25% better compute performance than M8g, per [AWS](https://aws.amazon.com/ec2/instance-types/m9g/). That positions Amazon more directly against [Nvidia](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/ai-ipo-race-heats-up-as-anthropic-files-nvidia-surges) for training and inference and against Intel and AMD for general-purpose compute, where price-performance decides placement. AWS said inference already accounts for well over half of its AI revenue and is growing, even as enterprises are still early in running inference at scale, and it described [AWS growth at 39.3% as a flywheel](/ai-news/public-companies/2026/aws-says-39-3-growth-is-flywheel-not-hype) rather than a one-off spike. For startups, the near-term constraint is availability. Trainium demand is so strong that customers have pushed Amazon to sell it outside AWS, but Jassy said the company intends to do that only late in 2027, so teams that want Trainium economics before then have to build inside AWS and live with the premium on older instances. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.