Continuous Power for Data Centres

Data centres are among the most demanding electricity customers there are. Their load is high, it is almost flat across the day and the year, and it must never stop. Servers, storage and network equipment run around the clock, and cooling runs with them. The rapid growth of artificial intelligence has pushed new facilities from tens to hundreds of megawatts, and in many regions grid connections for new sites now take years. Operators therefore face two linked problems: securing enough power in the first place, and matching it to a load profile that does not follow the sun or the wind. Every hour in which supply falls short must be covered by storage or by backup generators, and every kilowatt-hour that does not come from a firm source has to be balanced somewhere else.

The Established Approaches

The energy use and supply of data centres are well documented.

Scale of demand. Global data centre electricity use was estimated at about 205 TWh in 2018, around one per cent of world electricity, with efficiency gains offsetting much of the growth in computing [1]. The International Energy Agency estimated about 415 TWh for 2024 and projected more than a doubling by 2030, driven largely by AI [2].

Efficiency metric. Power usage effectiveness (PUE) is the ratio of total facility energy to the energy used by IT equipment [3]. A PUE of 1.0 would mean no overhead at all; industry averages lie around 1.5, and the most efficient large facilities reach about 1.1.

Supply and backup. Most data centres take their power from the grid, with uninterruptible power supplies to bridge seconds to minutes and diesel generators to cover longer outages. Availability standards such as redundant (N+1 or 2N) power paths define how much backup is installed.

Renewable sourcing. Operators buy wind and solar power through long-term contracts and certificates. Because generation and consumption rarely coincide hour by hour, matching a flat load with variable sources in real time requires storage, firm backup or large contract portfolios.

The established approach works, but at a price: a flat load is served by a system built around variable supply, and the gap is filled with grid capacity, batteries and fuel.

The Open Question

What a data centre needs most is generation with the same shape as its load: continuous, predictable, on site. Wind and solar cannot provide that alone, and grid capacity is increasingly the bottleneck. Continuous conversion of ambient energy would match the load profile by nature, but it delivers power in small units. The open question is how such units add up to a meaningful share of a megawatt-scale load, how much space and material that requires, and how to state the result in a way that can be checked against conservation of energy before any investment is made.

A Generalized Approach

A generalized approach to this problem is represented by the Schubart Master Formula, which gives the output of each generation unit from its inputs:

P(t) = η · ∫_V Φ_eff(r, t) · σ_eff(E) dV

The derivation, notation and open questions are published at pt-schubart.com. The model is introduced in Mathematical Models for Multi-Source Energy Harvesting.

Definitions

P(t): instantaneous electrical output, the usable electrical power delivered at time t. Unit: watt (W). Energy over an interval is the time integral of P(t), in joules or watt-hours.
η: total conversion efficiency, dimensionless, between 0 and 1. It combines coupling, resonance, rectification and electrical losses.
Φ_eff(r, t): effective ambient flux density at position r and time t, the local strength of all ambient inputs acting on the material.
σ_eff(E): energy-dependent effective interaction cross-section, how strongly the material couples to an input of energy E.
∫_V: volume integral over the active material.

Together, Φ_eff · σ_eff is the coupled power per unit volume (W/m³), so the integral over V yields watts. How that product divides between the two factors depends on the input channel, so no separate units are assigned to them here.

Assumptions and Boundaries

The Schubart Master Formula is a proposed generalized framework, not an established equation of physics. Its assumptions and limits:

Each unit is an open, non-equilibrium system; its output is bounded by coupled input power: P_out ≤ ΣP_in.
Unit output must be verified by independent measurement before it is used to plan supply; the example below uses illustrative values only.
Continuous on-site generation works alongside the grid connection and backup systems. It does not replace them; availability requirements still apply.
Power electronics, distribution losses and cooling of the generation units themselves are outside the formula and must be added.

Worked Example

All values below are illustrative values, chosen to show the method. They are not measured results.

The load. A data centre has 10 MW of IT load and a PUE of 1.3, so the facility draws 13 MW, or 312 MWh per day, almost unchanged around the clock.

Continuous generation. Assume generation units with an illustrative continuous output of 4 kW each. To cover 10 per cent of the facility load, 1.3 MW, requires 1.3 MW ÷ 4 kW = 325 units. They would deliver 1.3 MW × 24 h = 31.2 MWh per day, with the same flat shape as the load.

What follows. That share needs no storage to match the load profile, because supply and demand have the same shape. Covering it with solar power alone would require roughly four to five times the peak capacity, because a solar plant produces at full output only for a few hours a day, plus storage for the night. The remaining 90 per cent stays with the grid and the existing backup design.

Where It Leads

At unit level, the output of each generator depends on the material and its active volume; see Mathematical Models for Multi-Source Energy Harvesting. How many small units combine into a firm supply is the subject of Distributed Generation: Mathematical Models for Many Small Sources, and the full scaling path is described in From Sensor to Grid.

References

E. Masanet, A. Shehabi, N. Lei, S. Smith and J. Koomey, “Recalibrating global data center energy-use estimates,” Science 367, 984–986 (2020).
International Energy Agency, Energy and AI, IEA, Paris (2025).
The Green Grid, “PUE: A comprehensive examination of the metric,” White Paper 49 (2012).
A. Shehabi, S. J. Smith, D. A. Sartor et al., United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, LBNL-1005775 (2016).
Schubart Master Formula: derivation, notation and open questions, pt-schubart.com.

This page is published by the Neutrino Engineering District. Back to Fundamentals