a16zAmazon CEO Andy Jassy on the $220 Billion AI Buildout
a16z
Amazon’s scale now 0:00
Even after spending $220 billion this year, demand is still ahead of supply. AWS is sold out for 2026, much of 2027 is sold out, and a good part of 2028 too. Amazon sees data centers as part of almost every modern industry, so the job is to build as much capacity as possible and use it to improve customer experiences.
Retail and AI growth 1:00
Since taking over in 2021, the business has grown sharply. Overall revenue is about 80% higher, AWS is nearly three times larger, and ads are about two and a half times bigger. He ties that growth to fixing retail costs by regionalizing the fulfillment network, then to the rise of generative AI, which has pushed AWS to a $169 billion run rate and made the opportunity much larger.
Data centers and politics 15:30
Data centers sit behind almost everything people do now, from banking and health care to shopping, education, and business. AI adds a new layer of demand on top of that. The buildout needed to support it is compared to the highways and bridges of the 1950s. The claim is that the United States cannot fall behind, or other countries such as China will set the terms and the leadership.
Demand and customer tradeoffs 20:31
AI demand is described as a barbell. One side is the big labs and a few fast-growing apps that use huge amounts of compute. The other side is enterprises using AI for cost savings and productivity, such as customer service, fraud, and process automation. He says the middle, where most enterprise applications will eventually use inference, is still coming. He also says supply is tight, with 2026 largely sold out and 2028 partly sold out.
Inference Costs Drive Adoption 31:30
Trainium is described as about 30 to 40 percent better on price performance than other comparable processors. That matters because the cost of inference still blocks wider use of AI in enterprise apps. If companies must rewrite CPU-based software for AI accelerators, the savings have to be large. Lowering inference cost makes that rewrite easier and lets you use AI more broadly.
Smaller Teams, Faster Builds 37:31
The Bedrock team rebuilt an inference engine with six engineers in 76 days, which used to take 40 people a year. That is why the company expects smaller, cross-functional teams. More work will be done by fewer people, and many teams will start from a much more advanced place. Larger companies will have to rewrite code for agentic systems, while startups already begin that way.
Reinventing customer experiences 47:31
AI will make organizations leaner and give small teams room to build things people have not seen before. But if existing companies treat AI as only an incremental upgrade, they will miss the winning experiences of the next few years. Andy Jassy says Alexa had to be rewired because generative AI changed what the product could be, and the same is true for retail, Prime Video, Fire TV, and Prime Music. He expects hard change management, because teams are attached to what they built. Young people should use AI, learn it deeply, and keep their own critical thinking.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
