Dive Brief:
- Snowflake is banking on enterprise-grade AI solutions that are affordable to build to drive its data cloud business, CEO Sridhar Ramaswamy said Wednesday, during the company’s Q1 2025 earnings call. The company saw product revenue grow 34% year over year to nearly $800 million for the three-month period ending April 30.
- Arctic, the company’s open-source large language model, cost just $2 million to train and less than three months to develop. “Companies are talking about spending billions of dollars,” Ramaswamy said. “I don’t think we need to be there.”
- OpenAI’s GPT-4 and Google’s Gemini Ultra cost $78 million and $191 million to train, respectively, according to Stanford University’s Institute for Human-Centered Artificial Intelligence.
Dive Insight:
Ramaswamy touted Snowflake’s relatively modest AI development budget as an advantage.
“Being creative in how we develop these models is something that the team comes to naturally expect,” he said. “That kind of discipline and scarcity, to be honest, produces a lot of innovation.”
As large language models proliferate, technology vendors are aiming to ease enterprise adoption by embedding generative AI capabilities in existing products, from ERP and CRM platforms to data management solutions like Snowflake.
“It’s almost Darwinian,” Ramaswamy said, referring to the dynamic driving consumption of data services. “Tasks that required software engineering before just become a little pipeline that runs in Snowflake every hour or every two hours, acting on all of the data that is coming into Snowflake anyway.”
In addition to its Arctic model, Snowflake deployed Meta’s Llama 3 and Reka’s Core multimodal LLM family in its Cortex AI managed service, which reached general availability in early May. The company also integrated its Iceberg data storage solution with Microsoft’s Fabric analytics platform through an expanded partnership announced Wednesday.
To help shore up its AI capabilities, Snowflake acquired model performance observability platform TruEra, the companies announced Wednesday.
CFO Mike Scarpelli scaled back fiscal-year profit margin guidance due to AI-related costs after a quarter that saw Snowflake’s spending on purchases and property increase 137% year over year, to $16.5 million.
“We may be spending a little bit more on GPUs, but it’s also people that we’re hiring, specifically in AI,” Scarpelli said.
The TruEra acquisition will add 35 “key employees” to the payroll, Scarpelli noted.
Ramaswamy emphasized the company is looking to develop enterprise use cases like its Document AI unstructured data extraction tool, a data migration copilot and database querying capabilities.
“What we gain as Snowflake is the ability to fast follow on a number of fronts… to optimize against metrics that we care about, not [on] producing the latest, greatest biggest model for image generation,” he said.
Currently, the company is testing a Snowflake schema natural language tool that is not yet ready for public preview.
“We aren’t quite there yet, but I’d like to give Mike Scarpelli an app that knows about finance information that he’s able to query and actually trust the information that is coming out of it,” Ramaswamy said.
