From Icebergs to Arctics: Snowflake's Cool New Tools Heats Up the AI Race

Forget data warehouses. Snowflake, the company formerly known as the "Data Cloud," just dropped a bombshell at its 2024 summit: it's not just cloud-based anymore; it's AI-powered. That's why it is calling itself the "AI Data Cloud."

Sanjay Deshmukh, Snowflake's senior regional vice president for ASEAN and India, laid out the company's ambitious plans to transform how businesses leverage artificial intelligence.

"if you look at our aspirations, we believe that we can be the pioneers in enterprise AI," Deshmukh said. "We want to offer our customers an AI Data Cloud platform that is easy, efficient, and trusted."

Easy does it

The emphasis on ease of use has always been Snowflake's main advantage. Managing data across different data stores, lakes, etc. was mind-boggling for data engineers, with data scientists pointing fingers at them. The company helped them use the cloud as a data aggregator of sorts. Developers and data wranglers immediately saw the advantage.

Today, Snowflake sees a similar issue with building AI applications. It is a complex and fragmented process, akin to assembling a car from spare parts. Snowflake aims to change that by providing a unified platform that removes the underlying complexity.

"Our goal is to take that complexity onto us and offer you...an autonomous vehicle that is ready and can be taken everywhere," Deshmukh explained.

This means businesses of all sizes can now harness AI's power without specialized expertise. Snowflake's new no-code platform, Snowsight AI & ML Studio, encompasses this spirit. Now generally available, it enables users to build powerful AI applications within minutes simply by dragging and dropping components. Snowflake's own CoPilot, soon to be generally available, will allow business analysts to general SQL queries like a pro with written text.

The company also made a slew of new announcements around its AI offering. It launched Cortex Finetuning, which allows AI engineers and data scientists to train LLM models directly from the user interface. Cortex Analysts offers serverless LLM services for business users who are not amused with SQL text answers. Cortex Search combines vectors and keywords for lightning-fast queries.

Meanwhile, ModelOps engineers can manage ML features using the current user interface or Snowpark ML APIs. The Snowflake Model Registry, currently generally available, allows them to track and share AI/ML models and their metadata natively.

Trust is paramount

While consumers remain fascinated with generative AI, enterprises fret about its security and governance. They have much more to lose if the infant technology throws tantrums.

Snowflake addresses security and governance concerns head-on with its trusted approach. The company's design principle of "bringing the models to the data, not sending the data to the models" resonated with customers wary of exposing sensitive information.

Snowflake is also making it easier for companies to search for their data assets. It announced Universal Search, a generally available tool that leverages AI to classify and protect data assets. This, along with the Internal Marketplace, empowers businesses to create and share data products within their organization securely.

Snowflake's Trust Center, which will be generally available soon, helps to discover security risks and gives recommendations. The new Table Governance Views show points for users accessing each table. The company also previewed the Data Lineage Visualization interface, which unravels upstream and downstream dependencies for all tables.

To make AI accessible to a broader audience, Snowflake is providing customers with a choice of models, including its in-house model, Arctic, which the company claims is efficient in specific enterprise use cases. This allows the company to help companies that are still scratching their heads about use cases while feeling under pressure to drive LLM-based efficiency.

Snowflake's pay-as-you-go model ensures businesses only pay for the compute resources they use, making AI more cost-effective — a major focus for many enterprises.

Innovations across the board

Beyond the core pillars of ease, trust, and efficiency, Snowflake announced a slew of other groundbreaking innovations. These include:

• Apache Iceberg: This open-source table format lets you store structured and unstructured data in your own environment while still leveraging Snowflake's security and power. It's like having your cake and eating it, too — flexibility and control.
• Polaris Catalog: This open-source catalog allows you to choose your preferred query engine (Google, Microsoft, AWS) and run it on your Iceberg data. It's interoperability on steroids.
• Snowpark Container Services: Now generally available, Snowpark enables developers to build and deploy AI applications in a secure and managed environment.
• Document AI: This powerful tool, which will be generally available soon, allows businesses to extract insights from unstructured data like documents and images.
• Snowflake Container Services (SPCS), which is generally available, allows any workload to run in Snowflake using a container.

Developers get a lift

Snowflake has always been close to its developer and dataops communities. So, while the company is shifting focus to AI, it is also making data engineering and development easier.

They include Snowflake Trail, a new suite of observability features for developers to monitor and debut their data pipelines. Data Quality Monitoring offers metrics that allow users to easily define and measure tables.

Cost efficiency is another focus. Serverless Tasks Flex saves money by making it easier to decide when to run tasks, while Low Latency Tasks reduce task-scheduling intervals to 15 seconds.

Expanding the ecosystem: Native apps and the marketplace

At the Summit, Snowflake clarified that it isn't just building AI tools; they're creating a vibrant ecosystem of AI applications. The Snowflake Marketplace is a huge piece of this puzzle. It's likened to an App Store, where you can download AI-powered apps built by Snowflake partners.

Deshmukh highlighted two examples of these innovative apps:

• Landing AI: A Large Vision Model (LVM) player, which Snowflake invested in, showed how computer vision can be used to analyze images and videos, with applications in manufacturing, agriculture, and life sciences. Imagine using AI to identify robot assembly issues or discover new drugs.
• RelationalAI: The company has built an application on Snowpark Container Services to create knowledge graphs, revealing hidden semantic relationships in your data. Companies can use it to simplify decisions and create business rules minus the silos and security paralysis.

The future of AI today

"This is innovation," Deshmukh emphasized. "This was not being done before in the fashion that I explained." Snowflake's AI Data Cloud is not just about catching up with the AI hype; it's about unlocking new possibilities.

But what about customers not yet ready to dive headfirst into AI? Snowflake's strategy is to make the transition seamless. Their Snowpark developer framework and the growing number of "stable edges" (data-sharing partnerships) show that customers already use Snowflake for a wide range of use cases beyond traditional data warehousing.

As Deshmukh put it, "This is the most exciting part of AI...building something that can truly make an impact on the business."

Image credit: iStockphoto/Artur Didyk