Amazon cloud AI revenue run rate tops $15b, CEO says

Amazon said its cloud unit’s AI revenue run rate was more than US$15 billion in Q1 2026, marking the company’s first disclosure of direct financial returns from its AI business.

CEO Andy Jassy said the numbers are “ascending rapidly” and its cloud business as a whole would be growing faster without the capacity constraints affecting the wider tech industry.

Jassy also said Amazon’s chips business, which includes Graviton and Trainium processors, now has an annual revenue run rate above US$20 billion, up from US$10 billion earlier this year.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Behind Amazon’s new numbers lies a US$200 billion spending defense

Amazon’s disclosures answer investor worries about a plan to spend about US$200 billion on capital expenditures in 2026, up nearly 60% from the prior year 1.
CEO Andy Jassy framed the AI and chip revenue figures as proof that demand supports the investment, saying it is “not on a hunch” 2.
Jassy said Amazon’s AI revenue run rate is nearly 260 times the Amazon Web Services (AWS) cloud division’s revenue run rate three years after its commercial launch 2.
The US$20 billion custom silicon business spans Graviton general-purpose processors, Trainium AI chips, and Nitro, Amazon’s cloud infrastructure technology for networking, storage, and security. Revenue comes through AWS services, not standalone hardware sales 3.

Amazon’s long game is to vertically integrate and expand alternatives to Nvidia in AI compute

Amazon aims to take on Nvidia by building a more vertically integrated option that covers chips, systems, and cloud services, with Jassy saying “a new shift has started” in the AI chip market 4.
One analysis puts Trainium at 30% to 50% cheaper for cloud customers to run than comparable Nvidia instances, and it adds that Trainium’s manufacturing cost may be materially lower than Nvidia’s 5.
The same analysis cites anecdotal comparisons that place Trainium at about one-sixth to one-seventh the hourly price of an Nvidia H100 in some cases, depending on discounts and commitments 5.
Amazon plans for Trainium4 to support Nvidia’s NVLink Fusion interconnect, a technology that connects different chips at high speed. That support lets customers run Trainium4 systems with Nvidia graphics processing units (GPUs) 6.