Snowflake proves to be more than just a data specialist. You can now run containers on the Snowflake platform. The complexity and the management are taken care of by the company.

Snowflake introduces a managed service for containers. This allows you to run containers within the Snowflake platform. The entire service is essentially one managed KubernetesSolution that allows you to run code in Docker containers at will. Infrastructure management takes care of Snowflake. The service is named Snowflake Container Services.

As for the backend, we hear a familiar story. Don’t worry about the underlying complexity because Snowflake handles all the management hassles. Snowpark containers are ready for use. “Snowpark Container Services allow customers to run a Docker image in the data cloud while Snowflake takes care of everything that needs to happen underneath,” said Prasanna Krishnan, Director of Product Management for Data Sharing at Snowflake

Safe in Snowflake

So far from being unique, the managed aspect is where the containers run. Snowflake integrates them into its own platform, within its data governance system. The containers are introduced to the data, so to speak, so that the reverse does not have to happen. There they are subject to all the governance and policy rules that apply to the data in Snowflake.

Workloads must come on top of the data.

Frank Slootman, CEO of Snowflake

This is important. “As you move data across apps, you create silos,” said Frank Slootman, CEO of Snowflake. “You have to fight against that. Workloads have to come on top of the data.”

Tailored for AI

What workloads? Almost all of them, including those related to AI. Snowflake uses containers as a solution for external applications Large Language Models (LLMs) to Snowflake. These LLMs can come from Nvidia, which Snowflake has announced a partnership with, but also from third parties like AI21Labs and Reka.

The containers with the models do not have enough classic computing power to function optimally. “These days you also need GPU support,” agrees Christian Kleinerman, Snowflake’s senior vice president of product. Thanks to the partnership with Nvidia, this is now possible, still without leaving the Snowflake platform. “Training is 10 times faster and twice as cost-effective with the acceleration of Nvidia GPUs,” says Kleinerman.

native apps

The container service optionally works as part of the Snowflakes Native App Framework, which is now generally available to AWS customers. For example, users can containerize applications and publish them as a single package.

“You can publish a container application as an app,” says Krishnan. “But you can also publish another native app, which then calls a container service running at the user or provider. The container thus offers a function, for example an LLM.”

Snowflake Container Services is a very important announcement for Snowflake. The company insists that data must always be managed centrally, securely, and in one place, and that code must move into the data. The container service suddenly makes this possible on a very broad scale.