AI Workloads in Data Centers: New Aspects of UPS System Design
- Alpha Grissin
- Mar 2
- 4 min read
The implementation of artificial intelligence (AI) algorithms is significantly changing the energy consumption profile of data centers. Unlike traditional computing processes, AI workloads are characterized by high-amplitude transients (instantaneous changes in electrical current): power demand can fluctuate from 10% to 130% within milliseconds. Such pulsed loads pose risks to grid frequency stability and the thermal behavior of power electronics.
Based on Vertiv’s latest developments, we highlight three key mechanisms that help UPS systems adapt under these conditions.
1. Battery Shield: Battery Life Protection
In many modern UPS systems, any sudden change in output load causes the system to immediately draw on the batteries to compensate for an energy deficit. Because AI workloads can change thousands of times per hour, the battery is repeatedly exposed to micro-discharges. This accelerates chemical degradation and can overheat the electrodes.
Vertiv 's Battery Shield technology addresses this differently:
Energy buffering on the DC bus: the UPS uses a bank of high-capacitance capacitors to maintain stable voltage between the rectifier and the inverter. Capacitors can absorb and deliver energy almost instantly, unlike batteries, where energy exchange depends on slower electrochemical processes.
Dynamic voltage window: the system allows the DC-bus voltage to fluctuate within defined limits. When a load surge occurs, energy is supplied from the capacitors’ electric field.
Filtering effect: high-frequency AI transients are absorbed by the capacitor bank before they reach the battery terminals.
As a result, the battery remains effectively decoupled from these rapid fluctuations (or operates in a “rest” mode), which is critical for preserving the service life of both lead-acid and lithium battery systems.
2. Input Power Smoothing (IPS): Input Power Stabilization
Input Power Smoothing (IPS) is an algorithm that turns the UPS energy storage system into an active power buffer. Its main purpose is to ensure that the upstream utility (or generator) sees a smoother load profile and does not experience sharp power swings caused by AI equipment.
According to Vertiv’s technical documentation, the process is implemented as follows:
Baseline calculation: the system tracks the facility’s average load. Any deviation above or below this “normal” range becomes the subject of control.
Peak shaving mode: when the load suddenly increases (for example, the 10% to 130% transient), the UPS does not draw the entire incremental demand from the grid. Instead, the inverter supplies the deficit energy from the UPS’s internal storage (capacitors and, when required, batteries). This keeps the input current stable.
Valley filling mode: after the peak passes and consumption drops, the UPS gradually recharges its internal energy storage. Importantly, this recharge is controlled rather than instantaneous, avoiding additional disturbances on the input side.
FR% (Frequency Range) parameter: this setting defines a tolerance window. IPS can be configured to activate only when power fluctuations threaten network frequency stability.
This capability is especially important for sites with limited grid connection capacity or where the supply network has low inertia.
3. Input Power Ramp: Safe Operation with Generators
Input Power Ramp is critical for stable operation when a data center switches to on-site generation (diesel or gas gensets).
The key technical challenge is the dI/dt parameter, the rate of change of current. Most generators have mechanical inertia and cannot instantly respond to sudden load surges generated by AI workloads.
According to Vertiv’s technical documentation, Input Power Ramp addresses this by:
Load ramp-rate control: instead of applying 100% load to the generator in a fraction of a second, the UPS limits the rate at which input current increases. This creates a controlled ramp rather than a vertical step.
Using batteries as a temporary bridge: while the generator ramps up, the UPS supplies the shortfall from the batteries. The generator load increases smoothly, while the AI equipment receives the required power immediately.
Protection against “dips”: this approach prevents abrupt drops in generator output frequency and voltage that could trigger protective systems and cause a full shutdown.
In practice, the algorithm gives the generator time to spin up without subjecting it to a shock load.
The rise of AI workloads is changing the design paradigm for uninterruptible power supply systems. The critical factor is no longer only providing the required power rating and runtime, but also the ability to handle highly dynamic changes in demand.
High-amplitude, millisecond-scale transients increase requirements for:
power electronics response speed;
allowable battery charge and discharge currents;
input power smoothing algorithms;
component thermal stability;
controlled load transfer to generator sets (gensets).
Technologies such as DC-bus energy buffering, Input Power Smoothing, and Input Power Ramp illustrate a shift from passive backup toward active management of a facility’s energy dynamics.
Therefore, infrastructure readiness for AI is determined not by the UPS nameplate ratings alone, but by its ability to control transients, limit dI/dt, and maintain stable power parameters under rapidly changing load profiles.
* Based on an analysis of the Vertiv technical document "Advanced UPS controls for AI workload management," an article was prepared for Alpha Grissin Infotech Ukraine.
Source: alphagrissin.ua
Alpha Grissin – Your Guide to Vertiv Technologies
As an authorized Vertiv project distributor, Alpha Grissin continues to strengthen the brand's presence in Uzbekistan, Kazakhstan, and Turkmenistan, providing local customers with expertise, technical support, equipment supply, and implementation of comprehensive solutions for critical digital infrastructure.
Our mission is to make Vertiv's global technologies accessible at the regional level and ensure high standards of project implementation, regardless of scale.
Contact us for a consultation and a demo of Vertiv™ materials for your project.
