Podcast Summary: Bloomberg Talks – AWS CEO Matt Garman Talks AI Race
Date: December 2, 2025
Host: Ed Ludlow, Bloomberg Tech
Guest: Matt Garman, CEO of AWS
Event Context: Live from AWS re:Invent, Las Vegas, focused on AWS's latest advances in cloud and AI infrastructure.
Episode Overview
This episode spotlights the rapid AI advances by Amazon Web Services under CEO Matt Garman, focusing on custom AI chip innovation, the practical AI race among cloud giants, capacity scaling, customer adoption, and AWS’s partnerships—particularly with Nvidia and Anthropic. Garman delivers a comprehensive look into AWS’s approach to AI infrastructure, addressing both technical and business aspects, and gives a candid outlook on industry challenges and AWS’s competitive edge.
Key Discussion Points & Insights
1. AWS’s New AI Hardware and Rapid Innovation Cycle
-
AWS announced the next-generation “Trainium 3” accelerator chip, unveiled at re:Invent.
-
Garman explains that AWS benefits by controlling the whole stack—silicon, data center, server design—which enables faster delivery to customers and significant performance gains.
Matt Garman (01:40): “We control the silicon development, we control the data centers... and the performance that we’re seeing out of it is quite incredible.”
-
AWS is appearing to commit to an annual release cadence for new chips and server generations.
Matt Garman (02:30): “The desire and the hunger out there for more power and more compute is almost insatiable... we’re going to be pushing that envelope as fast as we possibly can.”
2. Balancing Nvidia and In-House AI Accelerators
-
Despite AWS’s advances with its own Trainium chips, Garman reasserts their ongoing commitment to Nvidia, with AWS touted as the best place to run Nvidia GPUs.
Matt Garman (03:22): “We’ve been working for 15+ years with the Nvidia team... people will tell you AWS is the best place, you get the best performance, the most stable cluster.”
-
Emphasizes customer choice—different workloads benefit from different accelerators—AWS aims to provide both best-in-class Nvidia infrastructure and proprietary solutions.
3. Scaling Data Center Capacity
-
AWS plans to double its cloud capacity by end of 2027 (to ~8GW).
-
Asked about resource allocation between in-house silicon and Nvidia GPUs, Garman says customer demand dictates provisioning.
Matt Garman (04:34): “We’ll let customer demand drive us a little bit on what they’re looking for... and that’s what we’ll continue to listen to.”
4. Profitability and Customer Benefit from In-House Chips
-
Garman points to rapid growth in AWS Bedrock (AI services platform) as evidence of accruing financial benefits from custom silicon.
Matt Garman (05:18): “We announced... more than half of all tokens and inference done in Bedrock are done on Trainium 2 servers under the covers.”
-
AWS’s own Nova and Nova 2 model families are also being accelerated by custom chips.
5. AWS & Anthropic: Deepening Partnership and Compute Challenges
-
The AWS-Anthropic partnership is “incredibly strong,” and Anthropic’s core models launch first on AWS/Trainium infrastructure—but due to massive demand, Anthropic uses other clouds as needed.
Matt Garman (06:18): “We’re definitely their primary cloud provider and closest partner for sure.”
-
Addresses AI compute supply constraints: issues are complex and shift quickly, not isolated to a single vendor or hardware type.
Matt Garman (07:14): “Never before has the technology industry ramped at the rate that we are right now... there are always constraints. It’ll change every month.”
6. From AI Assistants to AI Coworkers: The Agentic Era
-
AWS is investing heavily in “agentic” (autonomous AI coworker) technology, which Garman believes will provide “90% of the value” in enterprise AI—but says widespread customer adoption will take time and mindset change.
Matt Garman (08:38): “It is going to take change. People are going to have to change how they think about work, change their process flows… but they see that’s the path.”
-
Draws analogy to cloud adoption: still only a fraction of workloads have moved to cloud after 20 years.
7. Is AWS Number One in AI?
-
Ludlow asks Garman to address the industry perception of AWS as the AI leader.
Matt Garman (09:44): “When we see our customers... when I want to move it to production, I want to run on AWS. And that's... what makes me think we’re actually in a great position.”
Notable Quotes & Memorable Moments
-
On controlling the full stack:
“We control the silicon development, we control the data centers that IT land in... we can land that in very large clusters for people to take advantage of that.”
— Matt Garman (01:40) -
On AI demand:
“The desire and the hunger out there for, for more power and more compute is, is almost insatiable.”
— Matt Garman (02:30) -
On AWS as Nvidia partner:
“When you’re running a large cluster of Nvidia GPUs, people will tell you AWS is the best place. You get the best performance, the most stable cluster.”
— Matt Garman (03:22) -
On customer-led infrastructure growth:
“We’ll let customer demand drive us a little bit on what they’re looking for and what they want.”
— Matt Garman (04:34) -
On Bedrock and custom silicon benefits:
“More than half of all tokens and inference done in Bedrock are done on Trainium 2 servers under the covers.”
— Matt Garman (05:18) -
On supply chain constraints:
“Never before has the technology industry ramped at the rate that we are right now... always there’s a constraint in that system and it’ll change every month.”
— Matt Garman (07:14) -
On the shift to AI coworkers:
“It is going to take change... almost everyone I talk to definitely sees that that’s the path.”
— Matt Garman (08:38) -
On AWS’s AI leadership:
“When I want to move it to production, I want to run on AWS. And that’s the thing that we hear over and over again, which makes me think we’re actually in a great position.”
— Matt Garman (09:44)
Timestamps for Key Segments
- 00:25 — Introduction to AWS re:Invent, latest news
- 00:55 — Trainium 3, new chip, real-world deployment
- 02:11 — Annual cadence of new chips, speed of iteration
- 03:05 — Balancing Trainium and Nvidia GPUs for customers
- 04:19 — Data center capacity and customer-driven resource allocation
- 05:18 — Profitability and Bedrock’s growth via Trainium
- 06:18 — AWS-Anthropic partnership and supply/demand in compute
- 07:14 — Global supply chain constraints in AI chip ramp
- 08:16 — From AI assistants to coworkers (agentic technologies)
- 09:29 — Is AWS number one in AI?
- 09:44 — Customer proof: “I want to run on AWS”
Tone
Throughout, Garman is pragmatic, bullish, and energetic about AWS’s innovations. He is frank about both the pace and the challenges of the AI/cloud race, repeatedly emphasizing AWS’s philosophy of customer choice and bottom-up innovation while projecting quiet confidence in AWS’s market position.
