My thoughts on the best announcements from Matt Garman's AWS re:Invent 2024 keynote

As usual with re:Invent, I have high expectations. This year, I must say: wow.

Here’s what stood out to me from Matt Garman 's keynote.

Amazon S3 Tables: This one’s huge. S3 Tables, built on Apache Iceberg, are a game-changer for anyone dealing with data lakes. By automating metadata and optimising tabular storage, they make real-time analytics faster and so much easier. If you’re juggling massive datasets (who isn’t?), this simplifies everything. I love seeing AWS make analytics more accessible right on S3.

SageMaker Lakehouse: SageMaker got a massive upgrade. It’s now positioned as “the center for all your data, analytics, and AI.” Translation: SageMaker isn’t just for ML engineers and data scientists anymore. Teams across the board can now use it to gain insights and build smarter solutions. The Lakehouse architecture is my favourite part—seamlessly connecting data lakes and AI tools.

AWS Trainium 2 in GA, and Trainium 3 on the way: AWS is doubling down on AI hardware. Trainium 2 offers up to 40% better price-performance than leading GPUs, and seeing Apple and Anthropic using it for their AI workloads speaks volumes. Trainium 3 is on the horizon, promising even more performance and efficiency. It’s exciting to help push the boundaries of AI innovation.

Amazon Nova Models: AWS is back in the foundation model game in a big way—with not 1 or 3, but SIX new frontier models for everything from text to image and video generation. Matt showed benchmarks that stack up against other industry leaders and can save customers up to 75%. And the best part to me? They’re available on Bedrock, with all the tools customers love—fine-tuning, knowledge bases for RAG, guardrails, and more. I can’t wait to get hands-on with them.

Speaking of Bedrock: The new Automated Reasoning feature sounds like a game-changer. It helps reduce hallucinations in GenAI, which is exactly what many organisations need to move from experimentation to production. This feels like a big step toward real AI maturity.

Amazon Aurora DSQL: One of my favourites! It’s the fastest distributed SQL database out there, with 99.999% multi-region availability, virtually unlimited scalability, and no infrastructure to manage. Nothing else comes close—well, maybe Spanner—but Aurora DSQL delivers four times the speed. This makes high-performance global relational databases a reality.

Amazon Q Developer: After paving the way with Java, Amazon Q Developer now supports .NET, VMware, and mainframe transformations. This makes it easier for customers to tackle tough challenges like moving off Windows, VMware, and mainframes. A big win for organisations looking to modernise.

What a time!