AWS breaks the mold: SageMaker becomes unified solution

Digesting our thoughts on the flight back from Amazon Web Services (AWS) Reinvent24, one of our first impressions was that Matt Garman's keynote did not start with all AI all the time, but began with the infrastructure and data that must support it. Yes, AWS is upping its game in genAI foundation models with Nova, which Garman said was such a major step beyond Titan that it merited a new identity. Or that Bedrock is now opening a foundation model marketplace to give Google Model Garden a run for the money. And as AWS still needs to be BFF with NVIDIA, it's putting huge skin in the game in making Trainium 2 a viable alternative. Not surprising, given supply & demand for GPUs, the urgency of opening a second source (Google & Microsoft Azure are active there as well).

Bob O'Donnell has given a great wrap up on infrastructure & AI.
https://lnkd.in/e3j8ENm3

For us, the biggest impression of the week was about three related data & AI announcements:

1) Sagemaker is being repositioned, not just as the workspace for AI developers, but as the place where data + AI come together. Sound familiar? That's been Databricks MO over the past few years, and their success has prodded AWS to do what we've been asking them to all these years: weave their bespoke databases, IDEs, and lifecycle management/governance toolchains into a coherent, unified solution. SageMaker Unified Studio makes data a common resource from which the various query and analytics engines, from Redshift to Athena, Spark, and EMR feed. And that you can write SQL queries or use Q code assistant to generate them, or use Python through Jupyter notebooks to access, wrangle, and generate visualizations, run predictive models, and develop into AI applications from a common engine where you choose the "skin" and query engine of choice. And the projects are all commonly governed.

2) SageMaker Data Lakehouse, as a full implementation of ApacheIceberg, becomes the default data store on S3, but of course you can still push down query to data where it lives, including Redshift's native managed storage.

3) S3 Table Buckets, that enable SageMaker Data Lakehouse perform almost like a data warehouse.

Frankly, with Matt Wood's departure pre-conference, our expectations were pretty modest. For data, announcements in recent years were pretty incremental and not terribly exciting. This year it all changed. Building such a broad-based solution is a break in the mold for AWS. To put it simply, there's a lot for us to digest.

Watch this space, we'll be out with our full analysis in a couple weeks.