„Our AWS Trainium chips, designed in-house for AI workloads, now power more than 50% of Amazon Bedrock token usage.
written by Stefan Christoph
„Our AWS Trainium chips, designed in-house for AI workloads, now power more than 50% of Amazon Bedrock token usage. In practice, that means most of the generative AI responses you see from models on Bedrock are running on Trainium. That’s a big deal for customers: faster results, lower costs, and more efficient scaling as demands grows.“ - Amazon’s own framing of its Q3 2025 results — Andy Jassy made the same token-share point on the October 30 earnings call — and it is impressive (strictly, a majority of tokens need not mean a majority of responses — longer generations weigh heavier — but the scale stands out either way). Having the flexibility to - in this case chose between different compute options - is great, but the payoff — here: price performance at Bedrock’s scale — comes from actually exercising that choice, the way Amazon does with its own silicon ..
📝 Last updated: August 14, 2026 — Technical corrections from a quality audit; repaired a title that the LinkedIn import had truncated mid-word