Building Infrastructure for the Agent Era | AWS CEO Matt Garman
AWS CEO Matt Garman argues that AWS is rebuilding its infrastructure for agents, backing startups and a diversified customer base through a massive capacity buildout, while enterprises are held back by trust and workflow design.
Watch on YouTube ↗Charlie's summary
- ▶ 2:33
Startups are the lifeblood of AWS because they become tomorrow's enterprises and push the platform first
Matt interned at AWS in 2005 on an analysis of who the service would suit most, and the answer was startups. He estimates 30 to 40% of AWS revenue comes from companies that were startups during AWS's lifetime. They also teach AWS: they push services harder than banks or governments, so AWS can be ready when those customers want the same capability “5 years down the road.”
- ▶ 6:14
Agents need a cloud built for them: stripped-down signup, fast databases, and a context layer
Matt says agents want a well-defined API surface and care about tail latencies that people ignore, such as P999 S3 latency, which is why he thinks agentic workflows perform better on AWS. AWS now lets you create an account in under 30 seconds with a Gmail address, no credit card, and no VPC or IAM setup. It is a real account, so there is no migration later. A preview service, AWS Context, gives agents a layer for finding data across Aurora, S3 and other stores.
- ▶ 12:43
Agents need brand-new building blocks: sandboxes, gateways, and short-lived, fine-grained permissions
Matt says some agent needs are new primitives, not old ones used differently. You don't want to hand an agent your own permissions. You want time-boxed, narrowly scoped ones, and sometimes no access to a whole tool. Sandboxes matter too, and Firecracker microVMs, invented about 10 years ago, suit them because they spin up fast and have a good security boundary. There is also a tension: a throwaway agent database doesn't need “59s of durability,” yet AWS doesn't want a non-durable option, so it is trying to build databases that can be disposable or grow into production.
- ▶ 16:59
AWS deliberately reserves GPU capacity for startups instead of selling everything to the frontier labs
Matt says AWS could sell every GPU and accelerator to the big labs and call it a day, but chooses not to, because a broader ecosystem is healthier and diversifies risk. It allocates capacity for startups and says yes in some form to about 60% of requests, sometimes later, in another region or in a different configuration. It also announced plans to buy 2 million NVIDIA GPUs over the next couple of years.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.