Emerald AI, a member of Nvidia’s Inception program for startups that has some ideas–and now field test proof–about how to flexibly manage power consumption by AI workloads during peak power grid events, celebrated its official launch this week with a $24.5 million funding round.
That seed funding was led by Radical Ventures, with participation from Nvidia venture capital unit NVentures, AMPLO, CRV, and Neotribe. The Washington, D.C., company’s notable individual investors include Google Chief Scientist Jeff Dean, former US Secretary of State John Kerry, former Australian Prime Minister Malcolm Turnbull, AI pioneer Professor Fei-Fei Li, and Kleiner Perkins Chairman John Doerr. At least a couple of those names testify to the far-reaching orbit around Emerald AI founder and CEO Dr. Varun Sivaram, who was an advisor to Kerry when he was U.S. Special Presidential Envoy for Climate during the Biden administration. Sivaram also has served as a senior executive with at least two international alternative energy firms.
With that experience in mind, Emerald AI’s AI software addresses a key growing concern in the data center AI evolution: How can electric power grids keep up the rapidly escalating demand for energy coming from increasingly large AI data centers? Nvidia CEO Jensen Huang has talked often of how his company’s ever-more-powerful GPUs are helping to build the next generation of such “AI factories,” and how Nvidia is addressing the need for greater power efficiency.
Emerald AI said in a statement that new US data centers will seek 50-100 Gigawatts of additional power by 2030. Yet, “the deep deficit in available power grid interconnections will stunt AI innovation and U.S. competitiveness,” the company said.
The Emerald AI Conductor software platform aims to help resolve this challenge by orchestrating the power use of AI workloads in real-time, “enabling data centers to dynamically adjust their energy consumption and support grid stability while assuring acceptable AI compute performance,” the company said. “By intelligently managing energy consumption through software, AI data centers can bypass grid interconnection queues that can stretch 7-10 years or more.”
And now, Emerald AI is armed with more than just a notion about how to accomplish this, as the company this week also unveiled results of a field test in Phoenix, where the company, along with partners including Oracle Cloud Infrastructure, Nvidia, the Electrical Power Research Institute, and regional power utility Salt River Project, showed how a cluster of Nvidia GPUs in a commercial data center can reduce power consumption by 25% for three hours as a precise response to critical periods of grid stress, an example of which would be a summer system peak load event. The kicker, however, is that even with power consumption reduced, the platform helped acceptable customer performance for AI workloads, Emerald AI said.
An Nvidia blog post offered a further glimpse at how this was achieved, stating, “Emerald AI achieved this by orchestrating the host of different workloads that AI factories run. Some jobs can be paused or slowed, like the training or fine-tuning of a large language model for academic research. Others, like inference queries for an AI service used by thousands or even millions of people, can’t be rescheduled, but could be redirected to another data center where the local power grid is less stressed.”
Sivaram added in Emerald AI’s own statement, "We're at a critical inflection point as exponential growth of AI computing pressures our electrical infrastructure. To unshackle AI technology progress from power constraints, Emerald AI transforms data centers from grid liabilities into flexible assets, enabling grid operators to swiftly interconnect AI, bolster reliability and energy security, and more efficiently harness the massive spare capacity on today's grids."
Emerald AI said it is preparing for even larger-scale demonstrations in Phoenix and around the country.
"The next six to12 months will determine the gold standard for grid-responsive computing, and we're confident Emerald AI is uniquely positioned to set that standard," stated Rob Toews, Radical Ventures Partner.