re:Invent 2024 keynote
Matt Garman · Dec 3, 2024 · 2h 46m · Las Vegas, Nevada
Summary
Garman's first re:Invent as CEO framed AI as a set of building blocks inside every application. The keynote spanned custom silicon (Trainium2 instances, Trainium3), new data primitives (S3 Tables, S3 Metadata, Aurora DSQL), Bedrock features against hallucinations and for multi-agent work, the Amazon Nova models introduced by Andy Jassy, and a next-generation SageMaker.
Key themes: AI infrastructure, Data platforms, Models · AI-written summary from the published transcript
- 1EC2 P6 instances with NVIDIA Blackwell chips were announced20:34
- 2Trainium2 instances offer 30 to 40% better price performance than current GPU instances, and UltraServers link 64 chips23:26
- 3Project Rainier: a Trainium2 UltraServer cluster with Anthropic, five times the compute used for its current models26:41
- 4Trainium3, AWS's first 3-nanometer chip, will bring twice the compute of Trainium234:40
- 5S3 Table Buckets, a new bucket type for Apache Iceberg tables, promise 3x faster queries and 10x more transactions per second42:33
- 6Amazon Aurora DSQL, a distributed SQL database with strong consistency across regions, was announced59:31
- 7Andy Jassy introduced the Amazon Nova family of frontier models, including Canvas and Reel for images and video1:48:00
Announced
2 available the same day| Announcement | Status |
|---|---|
| EC2 P6 instancesNVIDIA Blackwell GPUs | Coming |
| Trainium2 UltraServers64 Trainium2 chips over NeuronLink | Preview |
| AWS Trainium3First 3nm AWS chip, 2x Trainium2 compute | Coming |
| Amazon S3 TablesTable buckets optimized for Apache Iceberg | GA |
| Amazon S3 MetadataQueryable metadata for S3 objects | Preview |
| Amazon Aurora DSQLServerless distributed SQL, multi-region strong consistency | Preview |
| Amazon Bedrock Model DistillationDistilled models up to 500% faster and 75% cheaper | Preview |
| Amazon Bedrock Automated Reasoning checksCatch factual errors with automated reasoning | Preview |
| Bedrock multi-agent collaborationAgents that coordinate on complex workflows | Preview |
| Amazon Nova modelsMicro, Lite, Pro, plus Canvas and Reel | GA |
| Amazon Q Developer transformation agents.NET, VMware and mainframe transformation | Preview |
| Next generation of Amazon SageMakerSageMaker Unified Studio in preview, Lakehouse GA | Announced |
Related keynotes
| Date | Keynote | |||
|---|---|---|---|---|
| Dec 3, 2025 | AWS re:Invent 2025Swami Sivasubramanian · 1h 41m · 7 announcedAgents, Models, Open models | Agents, Models, Open models | Swami opened by declaring agentic AI as a fundamentally new way of using AI. | 7 |
| Dec 2, 2025 | AWS re:Invent 2025Matt Garman · 2h 8m · 7 announcedAI infrastructure, Agents, Pricing | AI infrastructure, Agents, Pricing | Garman highlighted how AWS customers like Pfizer chose AWS to transform their business. | 7 |
| Nov 28, 2023 | AWS re:Invent 2023Adam Selipsky · 2h 15m · 7 announcedAI infrastructure, Models, Assistants | AI infrastructure, Models, Assistants | Amazon S3 Express One Zone launched: single-digit millisecond object storage, up to 10x faster than S3 Standard | 7 |
About this keynote
Keynote page: www.youtube.com/watch?v=LY7m5LQliAo
- Matt GarmanCEO, Amazon Web Services
Companies named: Anthropic, Apple, GitLab, JPMorgan Chase, Nvidia, PagerDuty, AWS
Last changed 2026-09-24.
Sources: official keynote videos, their captions and the hosts' recaps, summarized by AI; watch the source for exact wording. Logos via logo.dev; trademarks belong to their owners.