Features including a multimodal data ingestion and integration service that enables access to unstructured data demonstrate the vendor's move toward enabling app development.
SAN FRANCISCO -- After a slow reaction to surging interest in AI and a hurried attempt to play catch-up, Snowflake unveiled a slew of new capabilities this week that impressed customers.
Among the many new features introduced Tuesday during Snowflake Summit, the vendor's annual user conference, were an agentic AI-powered natural language interface for querying and analyzing data and an agent that enables data scientists to automate routine model development tasks using natural language.
In addition, they include a feature that lets users share and access semantic models to feed AI-ready data into applications; a multimodal data ingestion service called Openflow that enables users to combine structured and unstructured data; Cortex AISQL to automatically apply AI to data as it's ingested so users can run AI-powered SQL queries on all data types without having to restructure data; and performance optimization capabilities in Snowflake's Standard Warehouse.
All, however, are in preview with general availability months away. Therefore, how close the new capabilities bring Snowflake's AI development environment and the AI tools it's providing within its own platform to those offered by rival Databricks and other competitors remains to be seen.
I'm genuinely surprised and happy with what I'm seeing. Last year, we heard a lot of, 'This might be coming,' and there were a lot of ideas. This year, a foundation has been set, and it will be possible to hit the keyboards next week and do more in the coming months than we could just a few months ago.
Sam BiggsDirector of AI and automation engineering, CHG Healthcare
What is clear, however, is that Snowflake has a vision for AI in a way it did not as recently as the vendor's 2024 user conference, according to Sam Biggs, director of AI and automation engineering at CHG Healthcare Services.
"I was very excited [by the new capabilities]," he said. "I'm genuinely surprised and happy with what I'm seeing. Last year, we heard a lot of, 'This might be coming,' and there were a lot of ideas. This year, a foundation has been set, and it will be possible to hit the keyboards next week and do more in the coming months than we could just a few months ago."
Joji Phillip, director of data science at Swedish telecom giant Ericsson, was likewise positive about the breadth of Snowflake's product development plans.
"I think there were a lot of very good announcements -- more than what I was expecting, honestly, from an AI perspective," he said. "This year, it seems like 75% of the feature releases were very much AI-focused, a lot of which is of high interest to us -- like, for example, the Cortex AISQL part, the semantics part and Openflow."
By then, it had been 14 months since OpenAI's November 2022 launch of ChatGPT marked a significant improvement in generative AI (GenAI) technology and sparked the explosion of interest in AI. GenAI, unlike traditional AI, enables true natural language processing, has reasoning capabilities and can be trained to automate tasks, all of which can make workers better informed and more efficient.
Microsoft almost immediately invested in OpenAI and developed a partnership with the AI vendor. Databricks quickly developed its own large language model (LLM). And other vendors, including tech giants AWS and Google Cloud, developed integrations with LLMs that formed the foundations of GenAI development environments.
Snowflake, meanwhile, continued pumping out industry-specific versions of its data cloud and didn't make its first foray into GenAI until its May 2023 acquisition of Neeva. Even then, beyond an integration with Nvidia in June 2023 that enabled users to access that vendor's AI development capabilities, Snowflake's own AI development capabilities were nascent.
Finally, when Ramaswamy, who joined Snowflake as part of the Neeva acquisition, became CEO, AI became a priority.
Over the past 15 months, Snowflake has formed partnerships with LLM providers, including Mistral AI and Anthropic. It also turned Cortex, which was initially introduced in 2023 as an idea for an AI development environment, into a full-fledged platform for creating and managing AI applications.
Recent additions have included Cortex Agents, a managed service for developing agentic AI applications, and the general availability of Cortex Search and Cortex Analyst, which are tools for developers to discover relevant data for AI applications.
"Data is going very quickly from a situation in which it was used for reporting, or post-fact measurement, to being a central part of how change gets driven," Ramaswamy said during a virtual media conference before the conference. "We think we are set up to help drive change within organizations [by using AI] to provide faster, easier access to data."
The plans Snowflake unveiled during the conference provide evidence of that commitment to transformation through AI, according to Thevany Narayanamoorthy, manager of product management at retailer TJX.
"Right now, it seems that every competitor out there is working really hard to bring in technology that puts them at the top," she said. "There's a lot that you can do with Snowflake. They're continuing to innovate and try to stay on top of it, which is what you need. It's going to be interesting over the next couple of years because everything is evolving so fast."
Highlight features
While the new capabilities collectively advance Snowflake's AI development capabilities, perhaps most significant are Snowflake Intelligence and its access to semantic models, according to Mike Oldroyd, a data architect at technology consultant firm InterWorks.
Snowflake Intelligence is an agentic AI-powered interface that unifies all of an enterprise's data -- including unstructured data -- and enables users to query, analyze and model data using natural language rather than code. Sharing of semantic models, meanwhile, provides users access to semantic models created both internally and by third parties to easily integrate structured data into applications developed in Cortex.
"The big thing I didn't expect was Snowflake Intelligence, and then the semantic views that power that," Oldroyd said. "It looks like Snowflake is really starting to nail that semantic layer, which has been a point of contention [across the industry]. It's not felt like there's been a great benefit to it, and now Snowflake has come along with semantic views and put Snowflake Intelligence on top, so there's a practical use."
In addition, he highlighted OpenFlow, which saves Snowflake customers from having to use a third-party data ingestion and integration platform.
Like Oldroyd, Biggs said Openflow stood out as a potentially significant addition.
To date, CHG Healthcare has been using Snowflake only to operationalize its structured data. Potentially valuable information documents, PDFs, contracts, photos of receipts and other unstructured data formats had largely gone unused.
"It's always been difficult to bridge the gap between all the different ways we can consume information compared to structured data," Biggs said. "In reality, most businesses thrive off of and use a lot of unstructured data."
Openflow is designed to ingest and integrate unstructured data with structured data, enabling Snowflake customers to effectively make use of unstructured data.
"I'm very excited about OpenFlow," Biggs said. "It does seem to give us a way to start thinking about capturing unstructured data and bringing it into one unified platform so we can combine it with structured data."
As a retailer, sentiment analysis is important to TJX and something Narayanamoorthy said she wished were possible with Snowflake. Sentiment, however, is layered in unstructured data rather than structured data. Openflow, therefore, stood out to Narayanamoorthy as well.
"I was curious about … being able to do some of the analysis of images and files, sentiment analysis of that," she said. "There are so many technologies and different ways of doing it, so seeing how Snowflake [could make] it easy and accessible was the real question, and the announcements are great."
Competitive standing
Though not yet generally available, Snowflake's new capabilities bring its AI development environment and the AI tools it's providing customers more in line with those of its competitors, according to Biggs.
Before joining CHG Healthcare in March, Biggs was the director of R&D and data science at Pura and a senior analyst at Pluralsight. In those roles, he worked with other data management and AI platforms.
When he joined CHG Healthcare and started using Snowflake, he was able to compare Snowflake with the systems he previously used and found it lacking in some ways and strong in others. Now, pending the general availability of what it is building, Snowflake is providing a complete data and AI platform, according to Biggs.
"From the perspective of creating a unified platform that does everything, they're getting to a point where, if not the leaders, they're at least the leaders in terms of the ways they're thinking about problems," he said. "They're giving us tools that we may not be able to use right now, but we can see a path forward, and maybe in six months or a year from now it will give us a clear way of doing things."
InterWorks' Oldroyd noted that directly comparing vendors such as Snowflake and Databricks, or either one with tech giants, isn't a fair exercise, given that each has different strategies and target audiences.
For example, Databricks, though often the closest comparison for Snowflake, has historically targeted data scientists and geared its platform for machine learning workloads. Snowflake, meanwhile, is aimed more at business analysts with ease of use as a primary focal point.
"I think they were on a level playing field before," Oldroyd said. "Databricks still has an edge [in machine learning]. What Snowflake has now is Snowflake Intelligence and the extension to Cortex -- AISQL -- so they're going a different way. Rather than go for data scientists, they're going after regular data analysts who understand SQL and can now do text summarization."
As for a direct comparison of AI development capabilities, irrespective of target audience, Snowflake hasn't quite fully recovered from its slow start, according to David Menninger, an analyst at ISG Software Research.
Each year, the firm assesses all AI platform providers and publishes a guide. With many of Snowflake's capabilities -- even those introduced before this week -- not yet generally available, there's work to be done to match what others are doing.
"Snowflake and Databricks capabilities are rated closely in our analysis, but we give a slight nod to Databricks," Menninger said. "Some key features are still in preview mode, such as multimodal capabilities and Cortex Agents. There is also work to do on their low-code/no-code environment to be able to compete with some of the specialist AI/ML vendors."
Snowflake CEO Sridhar Ramaswamy speaks during the opening keynote address of Snowflake Summit, the vendor's user conference in San Francisco.
Wish list
Snowflake largely addressed the wishes of customers with the features unveiled Tuesday.
Being able to use unstructured data to inform analytics and AI tools was something Snowflake users wanted and will presumably get when Openflow is made generally available. The vendor also addressed some more specialized wants.
InterWorks' Oldroyd said he wanted an integration with Git to help track changes in source code throughout the development process. With the introduction of Workspaces, an editor for creating and managing code across file types, Snowflake now provides an integration with Git.
"For a while, I was grumbling about them not having proper Git integration with their workbooks," Oldroyd said. "They've now added Workspaces. It's making me rethink some of the architectures we have."
Ericsson's Phillip, meanwhile, said he hoped that the performance improvements in Standard Warehouse, which includes Adaptive Compute to automatically scale resources as needed, were also available for virtual private clouds.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.