Podcast Episode
"We're not investing approximately $200 billion in capex in 2026 on a hunch," Jassy wrote, pointing to committed customer demand including OpenAI's commitment of more than $100 billion to AWS.
Jassy estimated that if the chip division operated as a standalone company, its revenue would approach $50 billion annually. Demand has been so intense that capacity for Trainium3 is nearly sold out, and Trainium4 - still 18 months from availability - is close to being fully subscribed.
Amazon Bets $200 Billion on AI While Wrestling Internal 'AI Sprawl'
April 18, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Amazon CEO Andy Jassy has revealed AWS is generating over $15 billion in annualised AI revenue, while defending the company's record-breaking $200 billion capital expenditure plan for 2026. But behind the bold spending, employees report growing 'AI sprawl' - duplicate tools, fragmented data, and AI-assisted coding errors that have already caused major outages.
Amazon Goes All-In on AI with $200 Billion Bet
Amazon CEO Andy Jassy used his annual shareholder letter to mount a forceful defence of the company's plan to spend approximately $200 billion in capital expenditure in 2026 - the largest such outlay in corporate history and a nearly 60% increase on the prior year's $131.8 billion."We're not investing approximately $200 billion in capex in 2026 on a hunch," Jassy wrote, pointing to committed customer demand including OpenAI's commitment of more than $100 billion to AWS.
Record AI Revenue and Custom Chips
For the first time, Jassy disclosed that AWS's cloud-computing division is generating more than $15 billion in annualised AI revenue. The company's custom silicon lines - Graviton CPUs, Trainium AI accelerators, and Nitro networking cards - have collectively reached a $20 billion annualised revenue run rate with triple-digit year-on-year growth.Jassy estimated that if the chip division operated as a standalone company, its revenue would approach $50 billion annually. Demand has been so intense that capacity for Trainium3 is nearly sold out, and Trainium4 - still 18 months from availability - is close to being fully subscribed.
AI Reshaping Amazon From Within
Jassy outlined plans to rebuild Amazon's retail shopping experience from scratch using AI, citing a Bedrock team of six engineers who used Amazon's Kiro AI coding assistant to deliver a new inference engine in 76 days - work that traditionally would have required 40 people over a full year.The Growing Pains of AI Sprawl
The rapid scaling has not been seamless. Employees describe an internal phenomenon of 'AI sprawl' - a proliferation of duplicate tools and fragmented data as teams independently race to build AI-powered solutions. Earlier in 2026, AI-assisted coding errors caused multiple outages on Amazon's main e-commerce site, including one incident that resulted in approximately 6.3 million lost orders, prompting mandatory internal reviews and new safeguards for generative AI use.Published April 18, 2026 at 9:06am