Getty Images/iStockphoto

Snowflake signals AI commitment with Mistral AI partnership

Days after appointing a new CEO, the data cloud vendor showed a heightened focus on emerging technologies by aligning closely with a rising large language model developer.

Snowflake on Tuesday unveiled a new multiyear partnership with generative AI specialist Mistral AI that includes a technological integration and financial investment from Snowflake Ventures.

The integration, which enables Snowflake customers to access Mistral AI's large language models (LLMs), is now in public preview.

It comes five days after Sridhar Ramaswamy was named Snowflake's new CEO after Frank Slootman, who guided the vendor through a record-setting IPO in 2020 and helped Snowflake's revenues quintuple, resigned after five years.

Slootman was hired following the departure of Bob Muglia to lead Snowflake's economic growth. Ramaswamy, who came to Snowflake as part of the vendor's May 2023 acquisition of Neeva, a search engine vendor whose platform was fueled by generative AI, is now expected to lead Snowflake's technological growth.

Coming so soon after Ramaswamy took over as Snowflake's CEO, the vendor's new relationship with Mistral AI is perhaps a signal that Snowflake plans to make AI more of a priority going forward, according to Doug Henschen, an analyst at Constellation Research.

In addition, given that Wall Street reacted poorly to the CEO change -- with Snowflake's stock losing about 20% of its value since Ramaswamy took over -- it's possible that the introduction of the partnership was sped up to show investors that Snowflake is moving quickly to add AI capabilities.

The transition is about accelerating the company's AI strategy and delivering AI capabilities as quickly as possible. It's obvious the Mistral deal was already in the works, but I'm guessing it may have been accelerated to this week in part to show progress.
Doug HenschenAnalyst, Constellation Research

"As Slootman and Snowflake stated last week, the transition is about accelerating the company's AI strategy and delivering AI capabilities as quickly as possible," Henschen said. "It's obvious the Mistral deal was already in the works, but I'm guessing it may have been accelerated to this week in part to show progress and change the narrative as quickly as possible from last week's executive shake-up."

Based in Bozeman, Mont., but without a central headquarters, Snowflake is a data cloud vendor whose platform enables customers to store, query and analyze data.

Competitors include data lakehouse pioneer Databricks as well as tech giants such as AWS, Google Cloud and Microsoft. And like those competitors, Snowflake has made AI -- and generative AI, in particular -- a focal point of its product development over the past year.

Mistral AI, based in Paris, is an open source AI vendor. In February, the 2023 startup -- a rival to more established generative AI developers such as OpenAI and Anthropic -- formed a multiyear partnership with Microsoft that includes a $16.2 million investment.

Snowflake declined to specify the amount of Snowflake Ventures' financial investment in Mistral AI.

Snowflake and AI

Generative AI has been the dominant analytics and data management trend over the past year.

LLMs have the potential to expand analytics use to a broad audience of nontechnical users. In addition, part of their promise is to make trained data experts more efficient.

Data management and analytics platforms are complex, requiring data literacy training to comprehend and analyze data, as well as coding knowledge to create pipelines and query data. As a result, analytics use has hovered around a quarter of all employees within organizations for about two decades.

LLMs, however, enable true natural language interactions.

Therefore, when LLMs are integrated with data management and analytics tools, the LLMs eliminate one of the main hindrances to widespread analytics use. In addition, LLMs can be trained to generate code and automate certain repeatable tasks, which makes data scientists and other experts more productive by freeing them from time-consuming responsibilities.

To help customers develop AI models, including generative AI, Snowflake unveiled Cortex in November.

Cortex is a fully managed service still in private preview that provides access to LLMs -- including Document AI, Llama 2 and LLMs being developed by Snowflake -- so that users can integrate generative AI with their own data and train models to understand their business. Cortex also includes vector search capabilities to help customers build AI pipelines, universal search across an enterprise's different data environments and an AI copilot.

However, while Snowflake has unveiled AI capabilities in private preview over the past few months -- including generative AI -- other data platform vendors were earlier to prioritize AI and are ahead of Snowflake in moving such capabilities to public preview and general availability.

For example, Databricks developed its own LLM in March 2023. In addition, it acquired MosaicML for $1.3 billion in June to add AI development capabilities, unveiled added support for generative AI models in October, and introduced a new suite of tools in December to help customers reduce the frequency of AI hallucinations and improve generative AI outcomes.

"It was well apparent during last year's back-to-back [Snowflake and Databricks user conferences] in June that Databricks is much more mature, capable and ready where all things AI are concerned, including GenAI," Henschen said. "Databricks is less proven and polished, compared to Snowflake, as a data warehousing platform, but all eyes are on AI at the moment."

Mike Leone, an analyst at TechTarget's Enterprise Strategy Group, similarly noted that Snowflake has appeared to embrace AI, including generative AI, slower than its rivals. Now, however, with the appointment of Ramaswamy as CEO and its partnership with Mistral AI, Snowflake is demonstrating a greater commitment to AI.

"I think we're going to see Snowflake really accelerate their messaging and partnerships on GenAI under the new leadership," Leone said. "They're perceived to be behind their competition in this area because of their modest approach to LLMs to this point."

The integration

Through the public preview of the integration between Snowflake and Mistral AI, Snowflake customers will be able to access and deploy Mistral AI's large language models, including Mistral Large, in Snowflake's secure and governed data environment.

By using proprietary data in conjunction with Mistral's LLMs, enterprises will be able to develop models that address their specific business needs such as sentiment analysis.

As a result, the integration provides significant new capabilities for Snowflake users, according to Leone.

"This announcement will empower customers to explore a wider set of use cases with a highly performant LLM in Mistral that rivals -- and in some cases surpasses -- industry-leading LLMs like GPT-4 and Gemini Pro," he said, referring to popular LLMs from OpenAI and Google.

In addition, by enabling new applications for AI, the integration provides Snowflake users with some degree of choice when deciding which LLM to use when building generative AI models, Leone continued.

"I think it's critical to provide customers with flexible options to incorporate GenAI and LLMs on their terms," he said. "And while Snowflake has made several GenAI announcements throughout last year, there was a sense of limited support for a broad set of use cases. They've taken a modest approach to LLMs, especially compared to some of their competitors who have come out of the gates hot."

Henschen likewise noted that Snowflake needs to be more aggressive in its embrace of AI and provide customers with a choice of LLMs.

"It's important for Snowflake to show that it will be part of the AI conversation as soon and as frequently as possible," he said. "The Mistral announcement is a start, but they'll need to provide customers with many more choices. Model choice is what will really matter for customers, [and] Snowflake customers will want and need more choices."

More choices should be available once Cortex moves to public preview and general availability, he continued.

With respect to why Snowflake chose to partner with Mistral AI rather than another LLM vendor, the quality and speed of Mistral AI's innovation were factors, according to Baris Gultekin, Snowflake's head of product for AI.

Mistral AI was founded in late April 2023 and has already raised $548 million in funding. In addition, the vendor's LLMs measure up to those from rival vendors Anthropic and OpenAI in benchmark studies, according to Gultekin.

Mistral AI's approach to open source development was also a factor for Snowflake, he continued.

"Mistral's commitment to releasing the strongest open models in parallel to developing their commercial offerings provides Snowflake customers with the option of choice so they can harness the best LLM for their specific use case," Gultekin said.

Future plans

Snowflake, like many data management and analytics vendors, has unveiled AI capabilities in preview, but taken a relatively long time to get them ready for public consumption.

Among data platform vendors, Google Cloud went from August 2023 to February 2024 before moving more than a lone generative AI tool out of preview, while Microsoft has slowly rolled out its Copilots.

Databricks, however, has moved more quickly, Henschen noted. And Google Cloud, even though it unveiled many AI features over the past six months that are still in preview, has made some generally available and hinted that more will be generally available in April when the tech giant hosts its Google Cloud Next '24 user conference.

As a result, it is imperative that Snowflake accelerate its AI development cycle, according to Henschen.

"Snowflake has to get its AI capabilities out of private preview and into production, and show that the Snowflake Data Cloud can be a cost-effective platform for developing and delivering AI," Henschen said.

Leone, meanwhile, said that while the partnership with Mistral AI is a start, he'd like Snowflake to add more partnerships and integrations with LLM providers to give customers more choices among LLMs when developing AI models.

Just as ecosystems are vital to analytics, they are vital to AI, he noted. Partnering with more providers will enable customers to build those ecosystems.

"With openness being so critical right now in the GenAI ecosystem, more partnerships, more support for open source models, more guidance on a broader set of use cases and -- most importantly -- more public customer stories will go a long way in [helping] Snowflake catch up with where the industry is headed," Leone said.

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.

Dig Deeper on Business intelligence technology

Data Management
Content Management