Snowflake Looks to an AI Deal to Jump-Start Growth
Offering Cohere's AI models could boost revenue at Snowflake, where growth has slowed recently
The Takeaway
Snowflake is discussing a partnership with AI model developer Cohere that would allow Snowflake customers to use Cohere’s models in AI-powered applications
Snowflake had been riding high before the artificial intelligence boom. But questions about the usefulness of its data management products for AI developers—a growing source of revenue for rivals like Databricks—and a broader pullback in business software spending have sent its stock plummeting 42% so far this year.
Now, in an effort to jump-start slowing revenue growth, Snowflake is discussing a partnership with Canadian AI model developer Cohere that would allow Snowflake customers to use Cohere’s models in AI-powered applications, such as customer service chatbots or coding assistants, according to a person directly involved in the deal.
Such a deal could help Snowflake sell its products to data scientists and AI researchers, customers who have traditionally favored those of rival Databricks. That’s because Databricks’ software is specialized for machine-learning applications, while Snowflake has historically targeted business analysts at big companies who use its database products to make business recommendations.
Snowflake’s revenue growth has slowed recently. Revenue for the 12 months ended April 30 grew 33% from the same period a year earlier, down from 60% growth in the prior 12 months.
Offering access to Cohere, which specializes in building conversational AI for large businesses, could appeal to Snowflake’s customers. For instance, Cohere offers special models called embedding models that can convert business data into a format AI models can understand. With this software, Snowflake customers could more easily transform private data they store and manage in Snowflake for use in AI models.
The partnership would add to a string of initiatives by Sridhar Ramaswamy, who became Snowflake’s CEO in February, to boost the company’s AI offerings. The former Google ad tech executive previously ran Neeva, an AI-powered search engine startup that Snowflake acquired in May 2023.
This May, Snowflake bought TruEra, a startup that helps developers evaluate and monitor AI models and apps powered by large language models. And Snowflake has begun offering its customers access to open-source LLMs developed by Meta Platforms, Mistral AI and Google, as well as its in-house Arctic LLM.
This model-hosting business is nascent, though, and likely isn’t a large contributor to revenue, said Gil Luria, a senior analyst at D.A. Davidson who covers Snowflake. Given that, it’s unclear how much a potential Cohere-Snowflake partnership could boost Snowflake’s sales to AI developers.
Cohere’s models would be among the few closed-source LLMs Snowflake offers, in addition to those from Reka AI, which specializes in AI that can ingest text, images, audio and video.
Snowflake faces mounting competition from privately held Databricks, which has gained an edge by selling database management software that utilizes advances in generative AI.
Snowflake’s market capitalization shot past $120 billion in late 2021, thanks to soaring revenue from sales of software that helped businesses access and analyze data on cloud computing servers. But its core product, best suited for data that can fit neatly in data tables, proved less useful when generative AI started to transform software businesses.
In contrast, Databricks has expertise in storing and manipulating unstructured data such as images, video and audio, which are useful in training generative AI models. Many developers also use Databricks’ software to clean and format their data through a process called fine-tuning, they say. Databricks has also acquired smaller AI startups such as MosaicML and Lilac.
A Snowflake spokesperson said in a statement that the company “also offers support for unstructured data.” Cohere declined to comment.
A partnership would knit Cohere closer to Snowflake at a time when the founders of several startups engaged in the costly business of developing AI models have joined larger tech companies.
Last week, the founders of Character.AI agreed to rejoin Google in return for a licensing fee that pays back investors and employees at a price implying a $2.5 billion valuation. The deal followed two similar deals from the founders of chatbot maker Inflection AI and Adept AI.
Similar to those three startups, Cohere has a founding team that includes former Google AI researchers: CEO Aidan Gomez, who co-authored a foundational generative AI research paper, and Nick Frosst. The five-year-old company has raised $970 million from investors including Index Ventures and Tiger Global Management, and it was last valued at $5.5 billion in July.
The Toronto-based startup’s revenue has been dwarfed by that of rival model makers. In March, Cohere was generating revenue at $35 million annualized rate, while Anthropic was generating $100 million in ARR last fall and OpenAI surpassed $3.4 billion in ARR in June. (Annualized revenue typically reflects the prior month’s revenue multiplied by 12.)