ARTIFICIAL INTELLIGENCE
Advanced energy storage technology provides a “shock-absorber” solution for AI datacenter rack power.
By using a proprietary tech stack combining chemistry, battery technology and power electronics, Nyobolt solves the primary performance limitations that previously forced AI datacenters to use wasteful ‘artifi cial workloads’
U
nlike traditional cloud computing, which maintains a relatively steady power baseline, the new AI models require immense power instantly during inference or training spikes, followed by dramatic drops in between. To prevent a sudden power surge, many datacenters currently run artifi cial AI workloads (secondary workloads or traffi c) to keep power feeds stable and cooling systems circulating and to prevent “cold starts” that can delay infrastructure, impact utility grid stability, or damage sensitive components. These “artifi cial workloads” draw substantial power while delivering no revenue-generating compute or useful output. This ‘idling high’ strategy is a signifi cant drain on both profi tability and environmental sustainability.
An alternative solution utilises advanced battery technology to perform the equivalent of a fast-charging battery energy storage system (BESS) to provide ride- through combined with an ultracapacitor/ supercapacitor to deliver short-term energy with a completely new class of energy storage, the battery capacitor. Such a Dynamic Response System (DRS) can deliver instant power when demand spikes, providing peak shaving and GPU stabilisation during millisecond-scale events in AI racks. The system intelligently combines high power density and very high charge/ discharge capability with meaningful energy storage, giving supercapacitorlike instantaneous power response with higher energy density. By providing this energy buffer, a DRS-enabled system allows hardware to safely enter low-power states when idle, knowing the system can handle a near-instantaneous surge without the need for ‘fake’ loads.
Ultra-fast charging technology acts as the “electrical shock absorber” that makes a rack-level DRS physically possible and economical. Ultrafast charging moves the battery from a ‘backup-only’ component to an active power management tool. By
managing power fl uctuations at the rack level, it eliminates the need for ‘secondary workloads,’ proving that keeping systems artifi cially active is an ineffi cient and outdated practice.
A DRS system like this is only as good as its reaction time. If a battery takes even a second to kick in, an AI server might crash before it gets the power it needs. For a DRS to be effective, it must respond in milliseconds. This incredible speed acts
18 OCTOBER 2026 | ELECTRONICS FOR ENGINEERS
as a ‘bridge,’ instantly covering the power gap between a sudden AI task and the power feed’s slower response, a.k.a. – load smoothing. Because the battery charges/ discharges so fast, there is no longer any need to waste energy on secondary workloads to keep the grid stable. Handling ‘bursty’ AI workloads requires a battery to charge and discharge potentially hundreds of times a day as AI models spike and dip. Typical batteries are designed for ‘one big, static discharge’ during a power failure. If you used them for daily DRS load-balancing, they would wear out in months; however, advanced battery technology has been proven to support above and beyond 20,000 fast-charge cycles without signifi cant degradation. This means the DRS can actively use the battery to ‘shave the peaks’ of AI demand many thousands of times a year, effectively removing the need for secondary workloads as a stabilisation tool. As a bonus, performing peak shaving and load smoothing direct at the rack can enable signifi cant reduction of large, expensive, utility-class BESS that must otherwise be greatly oversized to perform smoothing and overcome the signifi cant lifetime degradation highlighted above.
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42
