This A.I. Start-Up Aims to Reverse the Backlash Against Data Centers

Silicon Valley Power is nowhere near the size of the California utility giant PG&E. But with roughly 55 data centers plugged into its grid across just 20 square miles, the municipal power company serves one of the most energy-intensive areas in the state.

That means it’s getting squeezed by the artificial intelligence build-out. Although Silicon Valley Power is working on doubling its capacity, the utility can’t keep up with rising demand from its customers, such as Nvidia and Intel, that want more immediately.

“All the spare capacity we may have had in the past has all been spoken for,” said Nicolas Procos, the utility director at Silicon Valley Power. “So now you’re talking about building out the system, or coming up with those creative solutions.”

As one of those solutions, Silicon Valley Power recently enlisted a two-year-old start-up called Emerald AI. Using its proprietary software, Emerald helps data centers be more flexible in their power consumption based on input from utilities.

Building and running A.I. models require immense amounts of computing power — and energy from the grid to power that workload. That strain on resources has helped spark a backlash against the data center build-out across the country that threatens to impede the industry’s growth.

Emerald is one start-up looking to mitigate that problem. On Tuesday, the company announced a $150 million funding haul that gave it a $1.05 billion valuation. The venture firms DCVC and Energize Capital are together leading the round. Nvidia, Samsung Ventures, GE Vernova and Salesforce Ventures, among others, are also joining the round. (Early investors include former Secretary of State John Kerry and Jeff Dean, a Google veteran who recently launched his own start-up.)

The stake from Nvidia is another example of how the chip giant is financing the growth of the A.I. economy.

“You can’t deliver compute without energy,” said John Tough, a managing partner at Energize Capital.

But the power that A.I. systems require is variable, with compute-intensive tasks potentially overloading the grid at peak times. Using Emerald’s software, data centers can reduce their energy consumption when utilities request it, and use excess capacity when utilities have it.

Emerald was founded by Varun Sivaram, 37, a renewable energy industry veteran who serves as its chief executive and also held climate policy roles in the Biden administration. The start-up’s chief scientist is Ayse Coskun, a computer science professor at Boston University who pioneered research into data center flexibility.

Emerald generates revenue in a couple of ways. By optimizing efficiency, it can allow utilities to serve more power to data centers, and it takes a cut of the extra sales. The start-up also helps data center operators work with their own customers — such as A.I. labs or hyperscalers — to figure out which A.I. workloads can be reduced and where they can utilize excess computing capacity. Emerald takes a cut from whatever extra revenue is generated.

Emerald’s software is powered by A.I. agents, which autonomously make decisions on how the data center operator and its customers should best reduce their power load. The company has tested its product at five data centers, working with partners including Nvidia and Oracle. It now has two contracts in California and Virginia.

Mr. Sivaram said his goal was to make data centers more grid-flexible and, in the process, “transform them from villains into heroes.”