Loading page…
The immediate infrastructure challenge is increasingly response speed, not simply total megawatts: batteries can cushion abrupt computing demand, but only if they are sized and controlled alongside generation assets.
Listen to this story
Emerson executive Phil Jones is calling for a different power architecture for AI factories: batteries and coordinated controls must absorb rapid computing-driven fluctuations before they reach turbines, engines, or other slower equipment. The analysis uses a potential 100-MW change within seconds, versus roughly 20 MW per minute for the fastest cited generation equipment.
Jones’s scenario is a potential design case, not a measured event at a named operating data center.
Rapid swings could cause torque damage, accelerated wear, or failure in turbines, engines, and other rotating equipment.
Battery effectiveness depends on capacity, load modeling, asset integration, and real-time supervisory controls.
AI data centers may need to be designed around sudden changes in demand, not just their enormous appetite for electricity. A new analysis by Phil Jones, Emerson’s director of AI data centers and microgrid sales, argues that potential load swings of 100 MW in seconds or less require batteries and coordinated controls to protect slower mechanical generation assets.
The distinction is consequential. Traditional power demand is generally spread across many users whose consumption changes at different times. AI facilities can instead bring thousands or tens of thousands of processors up or down together, creating synchronized demand changes within one site.
Jones’s analysis says the fastest original-equipment generation may be designed for swings of about 20 MW per minute. Rapid AI fluctuations could impose mechanical stress, torque damage, shorter asset life and potential failure on turbines, engines and other rotating equipment. The scenario is presented as an industry analysis by an executive at Emerson, which sells automation systems for the sector.
The proposed buffer is a battery energy storage system, or BESS: batteries positioned between volatile computing demand and generation that cannot change output as quickly. The systems can absorb rapid spikes and smooth oscillations, helping maintain frequency and voltage stability while protecting rotating equipment.
That does not make storage a replacement for generation. Its role is to make generation usable under abrupt conditions. Its effectiveness depends on battery size, load modeling, integration with the site’s power assets and a control system that can coordinate them.
The harder task is coordinating a mixed fleet. Advanced energy-control systems can use real-time parallel processing to manage batteries, gas turbines, engines and grid connections together. That can apply whether a facility uses grid power, on-site islanded generation, or both.
This requirement is intensified by the way AI projects are being assembled. The analysis contrasts traditional industrial timelines of three to four years with AI-factory projects executed in months. When procurement happens before designs are final, developers may combine equipment from several manufacturers with different control systems and ramp rates.
Connecting those disparate systems also broadens the cyberattack surface, particularly for behind-the-meter power systems that may not have dedicated security teams. The unresolved commercial question is whether developers can assemble storage, generation and control systems quickly enough while delivering the stability and security that high-volatility loads demand.
Editorial analysis
The important shift is from buying enough electricity to managing how fast a facility asks for it. The 100-MW-in-seconds scenario is a potential load profile in an Emerson executive’s analysis, not a published operating measurement. Still, it points to a practical test for the next wave of campuses: whether batteries, generation and controls work as one system under real workloads. That question connects to earlier efforts to make computing responsive to grid conditions and to the wider race for deployable power. The next useful evidence will be measured performance from operating facilities, including how much load storage can smooth and what it costs to maintain reliability.
Citation desk / original work
The important shift is from buying enough electricity to managing how fast a facility asks for it.
/posts/ai-data-centers-may-need-batteries-and-fast-controls-for-100-mw-load-swings#finding-1
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.