a16z (Andreessen Horowitz) · Oct 23, 2026 · 60 min

Amazon CEO Andy Jassy on the $220 Billion AI Buildout

Amazon CEO Andy Jassy argues the AI buildout is justified by demand that still exceeds supply, with capacity sold out through 2026, 2027 and much of 2028 despite $220B of 2026 capex. He describes AI demand as a barbell of labs and enterprise cost-avoidance, with the large middle of enterprise production workloads yet to come, and says cheaper inference via custom chips is key to unlocking it. He also expects fewer humans doing today's work, agent-native rewrites of code, and near-term cybersecurity risk from AI-found vulnerabilities.

Watch on YouTube ↗More from a16z (Andreessen Horowitz)

Where they stand

Very bullish

Demand for AI compute far exceeds supply, enterprise adoption is still early, and the buildout is a sound use of capital even at larger scale.

High conviction · years

Markets

Bullish

Jassy sees strong secular growth for AWS and Amazon's businesses, while acknowledging near-term free cash flow headwinds from heavy capex.

Medium conviction · years

Key takeaways

Amazon expects about $220B of capex in 2026 and says it is still leaving a meaningful amount of demand on the table. 13:18
Capacity is sold out for 2026, largely sold out for 2027 and a good chunk of 2028; supply constraints (memory, power, chips) will last several years, not just 2027. 24:48
AWS AI business is over $25B run rate after 3 years of the generative AI wave, several hundred times faster than early AWS; AWS grew 36.7% last quarter on a $169B run rate. 12:45
AI demand is a barbell: labs and a few runaway apps at one end, enterprise cost-avoidance at the other, with the largest future segment being enterprise production workloads in the middle. 21:27
Inference is still an order of magnitude too expensive for pervasive use; Trainium and Graviton price-performance gains are central to lowering it. 32:36
Capex risk is managed because data centers are built about 2 years ahead but server and networking gear is bought only months ahead with good demand visibility; free cash flow faces headwinds in early years. 14:24
Jassy expects fewer humans doing today's work, smaller cross-functional teams, and companies having to rewrite almost all code to be agent-capable. 39:52
The most immediate AI risk is cybersecurity: models find many more vulnerabilities, so patching must happen in minutes via agents. 43:06

Calls and positions

Supply constraints in memory and other components will persist for several years, not end in 2027.
“I think we'll see them in different components for several years.”
BullishPrediction: supply constraints last several years (several years)Andy Jassy, Oct 23, 2026Watch at 27:33
Within five to seven years, most enterprise applications will use inference pervasively.
“quite a few, if not most, of the enterprise applications that we use and that exist today are going to be using inference”
Very bullishPrediction: most enterprise apps use inference (5 to 7 years)Andy Jassy, Oct 23, 2026Watch at 22:34
Companies will have fewer humans doing the work done today, with smaller teams.
“we're going to have fewer humans doing the work that they're doing today”
BullishPrediction: fewer humans doing current workAndy Jassy, Oct 23, 2026Watch at 39:52 #AI agents
Companies will have to rewrite almost all their code to be agentically capable.
“I predict that companies will have to rewrite um almost all of their code to be agentically capable”
BullishPrediction: rewrite almost all codeAndy Jassy, Oct 23, 2026Watch at 42:01 #AI agents

Views expressed

Even at $220B of capex in 2026, Amazon is leaving a meaningful amount of demand unmet.
“even spending $220 billion of capex this year, we're leaving a meaningful amount of demand on the table”
Very bullishAndy Jassy, Oct 23, 2026Watch at 13:51
Amazon's AI capacity is sold out for 2026, largely for 2027, and a good chunk of 2028.
“26 is obviously this point sold out. 27 is largely sold out.”
Very bullishAndy Jassy, Oct 23, 2026Watch at 24:48
Amazon could spend a lot more than $220B and it would not be a bad use of capital, though it still wouldn't satisfy all demand.
“we could spend a lot more and it wouldn't be a bad use of of capital”
Very bullishAndy Jassy, Oct 23, 2026Watch at 27:33
Inference cost is still an order of magnitude too high for pervasive enterprise use.
“it's still an order of magnitude too expensive in my opinion for it to be used as pervasively as possible”
NeutralAndy Jassy, Oct 23, 2026Watch at 32:36
Anthropic and OpenAI are expected to be among the most successful AI companies over the long term, justifying Amazon's investments.
“we believe these are going to be two of the most successful companies in the space over a long period of time”
Very bullishAndy Jassy, Oct 23, 2026Watch at 34:50 #Anthropic #OpenAI
Large incumbents that move slowly on becoming AI-native will be more vulnerable to startups.
NeutralAndy Jassy, Oct 23, 2026Watch at 41:30 #AI agents
Near-term AI risk is mainly cybersecurity, requiring patch deployment in minutes rather than weeks.
“I don't think that there is enough discussion about the more immediate term risks, which are really around cybersecurity.”
BearishAndy Jassy, Oct 23, 2026Watch at 43:06 #AI and cybersecurity
The US cannot afford not to build data centers or China will set leadership and norms.
“cannot afford not to build data centers, or we will find that another country like China will will establish all the leadership”
Very bullishAndy Jassy, Oct 23, 2026Watch at 17:39 #AI capex

Topics covered

More from a16z (Andreessen Horowitz)

Summary generated by AI from the video's captionsand checked against it; quotes are verbatim. Processed Oct 25, 2026.

Read on Risk summarises what other people said in public videos and podcasts. Summaries are generated by AI and checked by code against the transcript, but can still be wrong; follow the source links before relying on anything. This is not investment advice and nothing here is a recommendation to buy or sell. Disclaimer ·How this works