Sources Podcast · Nov 10, 2026 · 53m · 43m saved
With Matt Garman, CEO of Amazon Web Services
Photo via aboutamazon.com
AWS is adding one GPU every 14 seconds and still expects to be short of computing capacity for the next couple of years. Matt Garman, CEO of Amazon Web Services, told Alex Heath on Sources Podcast that demand from AI is pulling workloads into the cloud that customers had refused to move for years.
Investors worry that the AI build-out is a debt-fueled bet that could break. Matt draws a line between AWS and the neoclouds, the newer GPU-rental providers, whose training clusters are bought by a few customers on borrowed money.
"I don't think it's systemic risk, actually. I do think that the risk is localized, because the bets that people are making are materially different." — Matt Garman
Matt runs a business at a $182 billion annual revenue run rate that Amazon CEO Andy Jassy has said could reach $1 trillion within five to ten years, and he decides which customers get its scarce GPUs.
The full interview is covered here so you can skip it. 53 minutes of audio, 10 minutes of reading.
Here are the 11 insights that matter.
Key Takeaways
AI agents are forcing mainframes, on-premises databases and VMware clusters into the cloud
AWS growth has accelerated for six straight quarters
AWS expects to be short of capacity for the next couple of years despite record capital spending
The neocloud risk is localized, not systemic, because AWS's AI revenue is spread across millions of customers
Every 50% cut in inference prices has brought an order of magnitude more usage, and revenue still rises each year
AWS will sell Trainium chips for use outside its own data centers
Nova's team was reset, not abandoned: AWS still plans a first-party frontier model
Automated reasoning could give "provably correct agents", with early deployments next year
1. The Road to $1T in Revenue
Andy Jassy's $1 trillion figure, set against the current $182 billion run rate, implies roughly 40% compound annual growth over five years, or growth in the high teens over ten. Matt pointed to six straight quarters of accelerating year-over-year growth.
Matt Garman says AI agents are pulling holdout workloads into the cloud
Customers who had resisted moving on-premises databases, mainframes and VMware clusters are now moving them, because agents cannot work without access to that data. A data source locked on a mainframe "draws all of it in" and forces the migration.
"When I have an agent go across all of my sources of data... it has to have all of the data sources accessible... I actually get very new insights and can build new businesses, unlock new sources of revenue." — Matt Garman
Companies that stay off the cloud risk falling behind
"There's a real risk that companies that don't make that move... they're getting passed by the ones that are in the cloud." — Matt Garman
