|
Welcome back! For all the talk about the rise of low-cost or open-source AI models, enterprises are still gravitating to brand-name models from OpenAI, Google and Anthropic. That’s prompting database providers like Snowflake, Databricks and Oracle to strike deals allowing database customers to use the brand-name AI to build chatbots and search tools that can analyze each customers’ proprietary data. In the latest example, Oracle and Google announced an agreement this morning that lets the database pioneer’s customers rent access to Gemini models through its cloud services and applications. While the database firms typically tout these agreements as wins, there is a defensive element to them. For Snowflake and Databricks, which offer customers access to Anthropic models, these deals reduce the risk that customers will switch to a cloud provider like Amazon Web Services or Google Cloud, both of which also host Anthropic models and sell competing database products. Snowflake, which also offers its customers access to OpenAI models through Microsoft’s cloud service, is seeing some early returns from these offerings. At the end of July, halfway through its fiscal year, Snowflake’s AI products were generating nearly $51 million in annual recurring revenue, according to an employee who has seen the number. The biggest contributor to that figure is Cortex AISQL, which lets customers use Anthropic and OpenAI models to query and analyze their data, the person said. Back in April, we reported that Snowflake executives had privately set a goal to reach $100 million in ARR from AI products by the close of the company’s fiscal year ending in January 2026, so they seem poised to meet that goal. The profitability of Snowflake’s AI business couldn’t be learned, but it probably isn’t as lucrative as its core database businesses because Anthropic and Microsoft get a cut whenever Cortex is used. It’s similar to the financially tricky situation AI coding applications are facing, as their surging revenue growth is tempered by the costs of using state-of-the-art models to power the apps. Still, Snowflake executives feel it’s an acceptable price to pay to keep customers from moving their data to another provider, the employee said. And it’s further evidence that chief information officers still want the best (and most expensive) AI they can find.
| 
