Why AI loads place mechanical stress on power plants, why Qnetic has created a product category around that challenge—and what evidence is still missing

An AI data center can jump from around 40% to 100% of its contracted power capacity in a fraction of a second—and then back again. Texas grid operator ERCOT describes this as a load pattern unlike that of any other type of consumer. In June 2026, U.S. storage manufacturer Qnetic defined a new product category in response: AI-Grade Energy Storage. This article examines whether the problem is real, what the category offers, and what measurement data are still needed.

What Is AI-Grade Energy Storage?

AI-Grade Energy Storage is a storage category defined by Qnetic for AI data centers. It requires six characteristics at once: millisecond response, unlimited daily cycling without degradation, high burst power, multi-hour capacity, a service life measured in decades, and inherent safety with no risk of thermal runaway.

The term comes from the white paper AI-Grade Energy Storage: Why AI Data Centers Need a New Category of Energy Storage, published by Qnetic in June 2026. The case for a new category is that the industry’s two established classes—battery energy storage systems (BESS) and long-duration energy storage (LDES)—were not designed for AI loads. BESS addresses power and daily arbitrage; LDES addresses duration. The combination of extreme dynamics and multi-hour capacity falls between the two.

The key word in the definition is “at once.” Each of the six characteristics is available on the market individually. A supercapacitor responds faster than any battery, but cannot provide hours of storage. A lithium-ion system can deliver power for hours, but degrades with every cycle. The category describes the overlap—and until now, that space had been unoccupied.

The Problem: Why AI Loads Affect the Grid Differently

AI data centers can swing between roughly 40% and 100% of their contracted power capacity in a fraction of a second. If the disturbance frequency matches a power plant’s oscillation mode, the generator can amplify the oscillation by as much as 50%—even on a strong grid.

The clearest account of the problem comes not from a storage manufacturer, but from the grid operator. In an analysis of the impact of AI loads presented in August 2025, ERCOT described the behavior of large data centers this way: their load shifts between roughly 40% of contracted capacity at idle and 100% under full load, within fractions of a second. No other type of consumer had previously exhibited this load pattern.

The consequences affect generation, not storage. Abrupt load changes force generators to handle high subtransient currents and torque. If the disturbance falls within the local oscillation mode—typically 0.5 to 2 hertz—the generator can amplify the oscillation by as much as 50%. Between 5 and 55 hertz, torsional resonance can threaten the shaft and turbine. Below 0.5 hertz, or when events occur several times a day, mechanical and thermal fatigue can result. On a weak grid, the same power plant can experience two to three times the fluctuation.

Qnetic puts ramp rates for gigawatt-scale facilities at more than 1,000 megawatts per second. That is a manufacturer’s estimate, but it is within the range documented by ERCOT. The industry’s concern is clear from the mitigation measure discussed there: power electronics installed directly at the data center. One example is an E-STATCOM rated at ±75 megawatts of active power, with a response time under five milliseconds, intended to turn millisecond ramps into minute-scale ramps.

Why Looking at Energy Use Alone Misses the Point

Public debate measures the AI boom in terawatt-hours. For grid stability, however, the critical factor is not the amount of energy consumed, but how quickly the load changes.

In its special report Energy and AI, published April 10, 2025, the International Energy Agency projects that data center electricity consumption will reach around 945 terawatt-hours by 2030—more than double today’s level and slightly more than Japan’s total electricity consumption today. In the United States, data centers are expected to account for nearly half of all demand growth through 2030; across advanced economies, they will account for more than 20%.

“AI changes the problem. The question is no longer how much energy we can store.”

—Michael Pratt, CEO of Qnetic

This shift in focus is analytically sound, even though it comes from a company with a stake in the outcome. The amount of energy is an infrastructure expansion problem, and therefore one that can be solved. The nature of the load is a grid-control problem—and adding another power plant does not solve it.

Why Flywheel Storage Meets the Six Criteria

The category describes a requirement, not a technology. Qnetic came up with it because mechanical storage can meet nearly all the criteria—a fact the company does not hide, and that readers should keep in mind.

  1. Millisecond response: Qnetic specifies a response time of under 50 milliseconds.
  2. Unlimited cycling: A rotor does not experience electrochemical aging. Cycling uses up fatigue life, but does not reduce power or capacity; and capacity depends solely on rotational speed.
  3. High power: The Q500 is designed for 125 kilowatts and up to 500 kilowatt-hours.
  4. Multi-hour capacity: Four hours at full power and up to 12 hours at reduced power— both points at which design departs from conventional flywheels.
  5. Decades of service: A 30-year operating life, with 22,000 full-equivalent cycles as the design target.
  6. Inherent safety: No electrochemistry and no thermal runaway. The system is designed for underground installation so the surrounding earth can absorb energy in the event of a failure.

The system uses a carbon-fiber rotor that spins without contact on active and passive magnetic bearings inside a vacuum chamber. Its design speed is 13,000 rpm; the approved operating point is 11,000 rpm, with an upper limit of 11,100 rpm. The safety system includes continuous vibration and harmonic monitoring with controlled shutdown, two mechanical backup bearings, and an enclosure that remains sealed during operation.

A Hybrid, Not a Replacement

Qnetic positions the system as a hybrid architecture: the flywheel absorbs high-frequency load spikes, while the lithium-ion battery handles the steadier baseline load—and therefore degrades more slowly.

This positioning is the category’s most commercially interesting feature. It does not ask operators to turn away from battery investments they have already made; it offers to protect them. For data centers that have already built their UPS and buffer architecture around lithium-ion batteries, this substantially lowers the barrier to adoption.

Flywheels themselves are nothing new in data centers. Other suppliers have offered flywheel UPS systems for years to bridge the seconds until backup diesel generators start—high power for a short time. The difference is in the design: Qnetic is targeting storage and load shifting over hours, not short-term ride-through. It is the same underlying physics, but a different product class—and that shift is what the new category is meant to capture.

What Qnetic Has Measured—and What Is Still Missing

So far, Qnetic has reported 100 trouble-free charge and discharge cycles on its Pulsar prototype, along with 130 kilowatts of power and 97.3% motor efficiency at more than 11,000 rpm. The measured round-trip efficiency, the carbon-fiber rotor’s performance at high speed, and safety certification are still pending.

The full-scale Pulsar prototype is designed for 200 kilowatt-hours. Mechanical assembly was completed in June 2026. Since September 2026, the company has been running low-speed tests to validate the control and safety systems before ramping up.

These results carry more weight than any marketing claim—and they are narrower in scope than they might first appear. The 97.3% figure refers to the motor’s energy-conversion efficiency, not the round-trip efficiency of the entire system; Qnetic gives around 85% as the system’s design target. And the test used a steel commissioning rotor, not the carbon-fiber rotor intended to provide the system’s actual capacity.

Qnetic has identified the remaining milestones itself: the carbon-fiber rotor, high-speed testing, a measured round-trip efficiency, safety validation by a U.S. certification body, and beta units in a utility pilot. The company plans to begin low volume production between late 2027 and 2028, explicitly subject to test results and available capital.

This disclosure matters more to the category than any individual performance figure. Since August 2026, Qnetic has participated in the Electric Power Research Institute’s (EPRI) deRISKED program. Test results from the in-house testing and pilots will feed into a standardized evaluation process and be shared with member utilities. A company that defines a category must also let its product be measured against independent standards; otherwise, the category remains a matter of terminology.

Conclusion

AI-Grade Energy Storage is both a diagnosis and a positioning strategy—and it would be a mistake to see only one side. The underlying problem is independently documented: AI data centers create load patterns that place mechanical stress on power plants, and that conventional battery storage and conventional long-duration storage were not designed to address. It is unsurprising that Qnetic has shaped the problem into a category that describes its own product. That does not invalidate the diagnosis, especially as competitors are now entering the same niche.

The answer is still unknown. Qnetic has a 200-kilowatt-hour prototype undergoing low-speed testing, 100 trouble-free cycles, a measured motor efficiency of 97.3%, and a timeline that puts volume production no earlier than late 2027. The decisive data—high-speed operation of the carbon-fiber rotor, measured round-trip efficiency, and safety certification—are not yet available. The fact that these results will be measured against independent standards through the EPRI program is the most solid part of the announcement.

Data center operators should still look into the issue now. Not because the products are available, but because commitments for grid connections are being made today, and grid operators are increasingly asking how a site will manage its own volatility. The six criteria offer a useful framework for evaluating any supplier, including Qnetic.