Artificial intelligence workloads are pushing data center power demand into territory the industry has never had to engineer for. In a recent conversation, Bill Kleyman, CEO and Co-Founder of Apolo.us and Executive Chair of Data Center Programs for Informa, talked with Bob Yeager, President of Emerson’s Power and Water Solutions business, about what it actually takes to keep that power stable once a site goes live.
Why It Matters
Bill cited an industry forecast of 200 gigawatts of additional power demand by the end of 2030, in a country with roughly 1.1 to 1.2 million megawatts of installed capacity. For context, one gigawatt serves about a million homes, and all of California consumes about 50 gigawatts. Getting megawatts on the ground is only half the problem. Keeping them stable under AI load behavior is the other half.
Key Takeaways
- Data centers are powered either grid-connected or behind the meter, and each path creates different reliability challenges.
- AI training loads can swing between 500 megawatts and a gigawatt four times a minute, which no single asset can absorb alone.
- Coordinated control across turbines, batteries, and balance of plant makes a complex power block behave like one virtual power plant.
- The limiting factors now are supply chain and people, not ideas.
See the complete set of videos at the Power Meets Purpose Series YouTube playlist.
Two Ways to Power a Data Center
Bob framed the choice simply. A site is either grid-connected or behind the meter. Grid connection creates real strain on utilities, because these facilities need two, three, four, or five gigawatts, and most utilities cannot add that capacity fast enough. Behind-the-meter generation solves the timing problem but transfers the reliability burden to the site itself, which is where automation earns its place.
Making 30 Turbines Behave Like One Plant
Emerson is currently working on a site with 30 gas turbines from different original equipment manufacturers and 2.5 gigawatts of batteries. Because of how AI models train, the compute load can surge between 500 megawatts and a gigawatt four times a minute. A nuclear unit is about 1,000 megawatts, so picture switching that on and off repeatedly every minute.
Turbines cannot be whipsawed that way without risking gearbox or rotor damage, so the batteries act as a 2.5 gigawatt capacitor that absorbs the swings. That only works if state of charge is managed to the millisecond. The Ovation Distributed Control System (DCS) handles turbine control, balance of plant, and battery state of charge together, so operators in the control room see steady operation rather than constant transients.
Emerson’s Ovation Automation Platform is used in about 20% of global and 50% of North American power generation, with more than 10 gigawatts of data center power capacity under contract, and its unified, fault-tolerant approach to generation, storage, and microgrid supervisory control targets 99.99% reliability.
Supply Chain and People Are the Real Constraints
Hyperscalers and utilities rarely build these power blocks themselves. They contract an engineering, procurement, and construction firm, buy generation equipment from turbine manufacturers, and bring in automation partners. Bob noted that very capable computer science teams generally are not power plant builders, which is where that partnership matters.
Large turbine makers are struggling to keep pace, with gas turbines reported sold out into the 2030s, pushing some projects toward reciprocating engines. The cascading effect on hiring and supply chain is being felt across the industry.
Explore Data Center Energy Management: Microgrid Control and Grid Balancing on Emerson.com to see how integrated microgrid control system capabilities support always-on data center operations.
