Snowflake CEO Sridhar Ramaswamy on AI-accelerated migrations, 75% growth in customer use cases, and why it's a bad time for 3-year plans

Snowflake CEO Sridhar Ramaswamy on AI-accelerated migrations and why three-year plans are obsolete

Sridhar Ramaswamy took over as Snowflake CEO in February 2024, inheriting a company built on Frank Slootman's aggressive go-to-market machine. His pitch is that AI is now doing two things simultaneously for Snowflake: collapsing the time and cost of getting data into the platform, and opening up entirely new ways for enterprises to extract value from it once it's there.

Migration timelines

The most concrete number Ramaswamy offers is on legacy migrations. Work that used to run three years now finishes in one or two quarters. He says the migration team, operating on top of agentic harnesses Snowflake has built, expects the remaining technical problems to be solved by end of 2025. What won't disappear is change management — validating that all the applications running on the old system still work on the new one, and managing business continuity through the switch.

“The number of use cases — these are new basically within customers — that our account executives won, went up by some 75%, year on year. That's the power of sort of using AI to get work done faster... This is a bad time to be making two and three year plans. Any time things are getting better or worse by 30% every month, all the human ability to predict these things, which are exponential in nature, just doesn't really work.”

Growth signal

The metric he uses to capture AI-driven demand is customer use cases won by account executives, which grew 75% year-on-year. Ramaswamy ties this directly to product: sales reps now have Snowflake Cowork on their phones and are expected to demo it live in front of customers. Solution engineers, who previously rotated through six or so standard demos, can now build custom demos tailored to a specific customer's vertical on the fly.

That's partly a product story and partly a deliberate cultural shift. Ramaswamy says Snowflake has internalized that the chasm between web-era click-through products and agentic workflows is real, and the companies that treat AI as a different mode of product delivery — not just a feature layer — are the ones that will compound.

AWS partnership

Snowflake buys significant GPU capacity from AWS and runs inference workloads on it. The more commercially meaningful layer is joint go-to-market: when a customer asks for AWS plus Snowflake, AWS leans in. Ramaswamy describes the relationship as running from the account level up to CEO-to-CEO, which keeps practical problems from festering.

The planning horizon problem

Ramaswamy is direct that two and three-year plans don't work right now. When capabilities are improving or shifting by 30% a month, human forecasting of exponential curves just breaks down. His working heuristic is to treat coding agents the way a driver treats Google Maps — consult them reflexively before doing anything, assume they'll know more than you do, and stay open to what a new model release makes possible that wasn't possible a month ago. The practical corollary for Snowflake's own engineering is structuring work so that when a model gets better, prior investment carries forward rather than gets stranded.

Strategic positioning

Ramaswamy frames Snowflake's bet around two positions in what he calls the "token path" — the value chain that runs from model inference through to customer outcomes. The first is the coding side, where Snowflake's agent Coco is meant to be the tool developers use to build data pipelines and artifacts. The second is the consumption side, where Cowork is meant to put enterprise data access in front of every employee, pulling from Snowflake and from other applications via MCP integrations.

He wants CROs to buy Cowork based on what Snowflake has already done to its own sales organization — using its internal transformation as the proof of concept. The strategic logic is that simply being embedded in an enterprise workflow is no longer a durable moat, because AI makes switching cheaper too. The companies that will hold their position are the ones generating genuine value each time an agent touches their platform.