Jassy says even with roughly $220 billion of 2026 capex, 2026 capacity is sold out, 2027 is largely sold out, and a good chunk of 2028 is sold out. He expects component shortages to last several years, not just through 2027; Amazon is still leaving meaningful demand on the table.
Jassy predicts fewer humans will do today's work and teams will be smaller and more cross-functional. He says companies will need to rewrite almost all code so agents can update it, especially for security, because patching will need to happen in minutes. That favors agent-native startups over slow incumbents.
Jassy compares the AI infrastructure buildout to 1950s highways and bridges. If the U.S. doesn't build data centers, he says China could establish AI leadership and norms; he calls for public-private partnership and community benefits such as tax revenue. He says the country cannot afford to fall behind.
Jassy says companies treating AI as incremental improvement to successful products likely won't have winning experiences in three to five years. Amazon rewired Alexa's brains and has teams reimagining retail, Prime Video, Fire TV and music from scratch rather than just adding AI features. He expects every consumer experience to change.
Jassy tells young people to use AI heavily but not let it replace critical reasoning. AI can get everyone to an advanced starting point in case analysis, but human judgment, experimentation and synthesis still build strategy, products and customer experiences. He says the tool is unavoidable and should be practiced.
Jassy says Amazon did not buy Annapurna for AI chips. The team first offloaded virtualization, networking, security and storage from EC2 servers to a NIC, then built Graviton CPUs and Trainium accelerators. He calls it one of Amazon's best acquisitions and says 99% of top 1,000 EC2 customers use Graviton.
Jassy describes AI demand as a barbell: AI labs and a few runaway apps consume massive compute, while enterprises today mostly use AI for cost avoidance or productivity. The middle—current production enterprise apps—has barely adopted inference, but he expects it to become the largest absolute segment.
Jassy says inference costs have fallen but remain an order of magnitude too expensive. Enterprises considering rewriting CPU applications for AI accelerators face a high benefit bar when costs stay high. Lower inference costs would make pervasive adoption easier, and Amazon's Trainium chips target better price performance.
Jassy says public nervousness focuses on fantastical model risks like evil botnets, but cybersecurity is the nearer-term threat. Models find more vulnerabilities than humans did, so applications must be rewritten for agents to deploy patches in minutes rather than months. He calls this immediate risk more concrete than hypothetical robot takeovers.
Jassy pushes back on data center backlash with a water comparison: golf courses consume 200 times more water than all Amazon data centers, and use more in two days than Amazon does in a year. He says grid investment was needed regardless. He argues the industry can improve but is being unfairly villainized.
Andy Jassy says Amazon must build data centers about two years before it can install servers and monetize them, while server and networking gear are bought only months ahead. If demand disappoints, it can skip the server purchase; data centers last roughly 30 years. That cash-cycle gap limits downside.
With memory and components scarce, Jassy says Amazon could theoretically favor AI labs willing to pay more. Instead, he says AWS won't squeeze customers running their businesses on core services; it will help them scale while adding as much capacity as possible for AI demand.
The hosts cite a Bedrock team rebuilding its inference engine with six engineers in 76 days, work that would have taken 40 people a year. Jassy agrees the experience was eye-opening and says four or five people prototyped an agent idea over a weekend. It shows how small teams now build.
