The explosion of cloud data workloads has brought a parallel spike in costs—especially for platforms like Snowflake. But what if those costs could be cut dramatically, without rewriting a single query?

That’s the core idea behind Espresso AI’s newest release, dubbed “Kubernetes for Snowflake.” According to CEO and Co-Founder Ben Lerner, the company uses real-time warehouse intelligence and large language models (LLMs) to optimize workloads, making cloud data usage far more efficient.

The Google DNA Behind Espresso AI
Before founding Espresso AI, Lerner worked at Google on machine learning and systems performance. “The idea was to take how Google runs its infrastructure and bring that power to every company,” Lerner said. Espresso AI emerged from research with DeepMind and practical insights from managing one of the world’s most advanced infrastructures.

Smarter Scheduling for Warehouses
At the heart of Espresso AI’s approach is scheduling—much like Kubernetes does for containers. “We sit as a proxy between your tooling and your Snowflake instance,” Lerner explained. The system dynamically decides which warehouse to route queries to, based on real-time analysis of compute demand and available capacity.

This allows for substantial increases in warehouse utilization—from 40–60% up to 90% in many cases—translating into cost savings as high as 70%. “It’s just like packing a moving truck efficiently—you don’t want to leave half-empty bins and then rent a second truck.”

One of Espresso AI’s key advantages is how unobtrusive it is. “You don’t need to change how you write queries or structure data. Setup can be done in under an hour, and once it’s live, you just see your bill go down,” Lerner said.

Most customers use it without ongoing intervention. “We often only check in quarterly. The feedback is usually: ‘We didn’t notice anything—except the bill dropped.’”

Not Just for Snowflake
Though the Kubernetes for Snowflake feature is the headline today, Espresso AI is already expanding. A beta for Databricks SQL is underway, and support for BigQuery and Redshift is on the roadmap.

The underlying insight is that LLMs now understand code and workloads in ways that traditional static optimizers cannot. “Any system where a good data engineer could make cost-saving tweaks—LLMs can now do that, and they do it at scale,” Lerner noted.

Lerner says Espresso AI is especially useful for fast-growing companies with small, overburdened data teams. “You’ve just crossed half a million dollars a year in Snowflake spend, you don’t get more headcount, and now finance wants answers,” he said. “We act like an expert data engineer—one that scales instantly and never takes a break.”

A Sign of What’s to Come
With data warehouse costs rising fast, Espresso AI represents a shift toward invisible infrastructure intelligence—one that may become the norm as data ops mature. As Lerner put it, “We’re just using the hardware you already pay for, but smarter.”