Espresso AI, a new optimization platform founded by ex-Googlers, has launched a large language model (LLM)-driven solution designed to transform Databricks into what it calls an “agentic lakehouse.” The company claims its technology can reduce data warehouse costs by up to 50% while boosting utilization and performance—all without manual tuning.
“Databricks is seeing explosive growth with their Data Lakehouse product,” said Ben Lerner, CEO and co-founder of Espresso AI. “But if they want to catch up with Snowflake adoption, they’ll need to be as optimized and cost efficient as possible. By leveraging Espresso AI, Databricks customers can cut their bill in half and see their efficiency skyrocket without any manual effort.”
It’s a timely pitch: Databricks, now valued at over $100 billion after its August 2025 funding round, is growing revenue at a 50% year-over-year pace, with AI products alone crossing a $1 billion run rate. But as enterprise workloads explode, cost control remains one of the few pain points in the data infrastructure gold rush.
From DeepMind to Deep Optimization
Espresso AI was founded by Ben Lerner, Alex Kouzemtchenko, and Juri Ganitkevitch, all former Googlers who worked on DeepMind, Google Cloud, and Google Search—where they honed their expertise in large-scale systems, performance optimization, and machine learning.
Backed by $11 million in seed funding from FirstMark Capital, Nat Friedman, and Daniel Gross, the company’s mission is to make enterprise data platforms as intelligent and self-optimizing as the AI models they serve.
In a six-month beta program with hundreds of enterprises, including Booz Allen Hamilton and Comcast, Espresso AI proved it could deliver immediate, measurable savings.
“Espresso AI cut our bill in half with no lift from our side,” said Nataliia Mykytento, Head of Engineering at Minerva. “They were instrumental in reducing costs that were growing too fast for comfort.”
How It Works: Three AI Agents, One Smarter Lakehouse
Espresso AI for Databricks is built around three core “agents” that act autonomously to optimize compute, scheduling, and query execution:
- Autoscaling Agent:
Trained on a customer’s own metadata logs, it predicts usage spikes and fluctuations, scaling compute resources up or down automatically while balancing cost and performance. - Scheduling Agent:
Tackles one of Databricks’ most persistent inefficiencies: idle utilization. With typical usage hovering around 40–60%, Espresso AI routes queries to underused machines, ensuring compute power doesn’t sit idle. - Query Agent:
Fine-tunes every SQL query before it even touches the data lakehouse—improving performance and lowering costs at the query level.
Together, these agents act as a self-optimizing control layer over Databricks’ infrastructure—what Espresso calls an “agentic lakehouse.” The result: cloud-class performance, data-driven efficiency, and zero manual babysitting.
The Bigger Picture: Agentic AI Comes for Data Infrastructure
While Snowflake and Databricks continue to duel for dominance in the enterprise data arena, Espresso AI’s launch underscores a new arms race—agentic AI infrastructure. Instead of humans writing rules or setting quotas, AI agents now dynamically manage compute, workloads, and cost policies in real time.
That could be a seismic shift for data teams: less time tuning clusters, more time deriving insights. And if Espresso’s numbers hold true—cutting costs in half without code changes—it may have just brewed the shot Databricks users didn’t know they needed.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
