Amazon says custom chips surpass $25 billion annual run rate as Trainium and Graviton power AWS AI

Amazon said its custom-silicon business has surpassed a $25 billion annual revenue run rate, a company-reported milestone that shows how a decade-old in-house chip effort has become a major piece of Amazon Web Services’ AI and cloud infrastructure business.

The benchmark came in Amazon’s Q2 2026 earnings release, published July 30 for the quarter ended June 30, and in a companion AboutAmazon post. Amazon did not disclose a separate audited annual revenue figure for chips or break out a chips line item in its financial statements. Instead, the company described a run rate — an annualized pace based on current business levels, not the same as reported full-year revenue. In the same release, Amazon said AWS sales rose 37% from a year earlier to $42.2 billion in the quarter, putting the cloud unit at a $169 billion annualized revenue run rate. CEO Andy Jassy said, “AWS is booming, growing 36.7% year‑over‑year in Q2—our fastest growth in 18 quarters—and our AI and Chips businesses each eclipsed run rates of more than $25 billion.”

As Amazon describes it, the chips business includes its Graviton server processors, Trainium AI chips and Nitro infrastructure technology. The company said it delivered more than 2.1 million AI chips over the past 12 months, with more than half of them Trainium, and said Trainium capacity is heavily committed by customers. AboutAmazon also said the custom-silicon business was growing at triple-digit percentages year over year. Jassy, as quoted by AboutAmazon from the earnings call, said: “Our Chips revenue run rate is now over $25 billion. We’re unusually well‑positioned for this AI inflection given our leading price‑performance chips in both AI (with Trainium) and CPUs (with Graviton). …”

The milestone caps roughly a decade of investment in designing chips for AWS. Amazon’s push began with its 2015 acquisition of Annapurna Labs, an Israeli chip-design company that became the foundation of its in-house silicon program. Over the years, that effort expanded into chip families such as Graviton for general cloud computing and Trainium for AI workloads, alongside Nitro, which underpins parts of AWS infrastructure.

The significance of the new run-rate figure is less about branding than about how large customers are planning capacity. Two of Amazon’s biggest AI partners have made multi-year, multi-gigawatt commitments tied directly to Trainium and Graviton. Anthropic, the AI startup behind the Claude chatbot, said on April 20 that it would secure up to 5 gigawatts of Trainium and Graviton capacity and spend more than $100 billion on AWS over the next 10 years. Anthropic CEO Dario Amodei said, “Our users tell us Claude is increasingly essential to how they work, and we need to build the infrastructure to keep pace with rapidly growing demand. Our collaboration with Amazon will allow us to continue advancing AI research while delivering Claude to our customers, including the more than 100,000 building on AWS.”

OpenAI said on Feb. 27 that it would use about 2 gigawatts of Trainium capacity through AWS, and that AWS would be the exclusive third-party cloud distribution provider for OpenAI Frontier. Those commitments suggest Amazon’s chips are no longer a side project inside AWS but part of core infrastructure planning for major AI customers.

Amazon also pointed to broad adoption of Graviton across its existing cloud base, saying 98% of the top 1,000 EC2 customers use the processor family. Taken together with the company’s reported $25 billion chips run rate, the delivery of more than 2.1 million AI chips and the large external capacity deals, Amazon is making the case that custom silicon is now central to its AI infrastructure strategy.