AI data centres are increasingly part of the power system reliability conversation.
Across Europe and North America, system operators and industry bodies are moving beyond demand growth alone to examine how emerging large loads behave, how they interact with the grid, and what this means for connection requirements, operations and planning.
The capacity challenge remains real. Data centres need large amounts of power, often in locations where grid capacity is already constrained. But capacity is only one part of the problem. Alongside capacity, grid operators also need to understand, assess and manage the dynamic behaviour of these new loads.
From demand growth to system behaviour
ENTSO-E’s latest report on data centres and the power system describes data centres as “systemically relevant electricity loads” whose behaviour increasingly affects secure grid operation. The report estimates that European data centre electricity demand will grow by more than 50% between 2025 and 2030, driven mainly by colocation and hyperscale expansion.
A 100 MW data centre does not behave like a 100 MW steel mill or chemical plant. Data centres are shaped by software-defined workloads, power electronic interfaces, UPS systems, cooling infrastructure and highly concentrated connection points. AI training, inference, SaaS, batch processing and real-time workloads all have different load shapes, flexibility levels and infrastructure needs.
Visibility is becoming a reliability issue
ENTSO-E highlights that data centres can introduce stability risks “at speeds and frequencies that conventional SCADA systems cannot detect”, requiring high-resolution phasor measurement and operational modelling to identify and analyse dynamic behaviour.
Planning studies and connection assessments remain essential, but they need to be strengthened by real-world measurement. For large, software-driven loads, grid operators need to understand not only what models predict, but how facilities behave during normal operation, disturbances and recovery.
At Reactive Technologies, we see this as a measurement-based assurance challenge. As connection requirements evolve, high-quality measurement will play an important role in validating models, detecting fast-changing behaviours, and building confidence that new large-load connections can operate safely under real-world conditions.
AI workloads can create new dynamic risks
During normal operation, fast, software-driven load swings can create voltage and frequency issues, especially where multiple large data centres are concentrated in the same area.
ENTSO-E also highlights the risk of load-induced forced oscillations. Persistent, periodic load patterns from AI training can excite electromechanical modes of the power system, creating a risk of resonance amplification across wider areas.
UPS-based architectures can amplify grid disturbances
Data centre power architectures are designed to maintain uninterrupted IT operation. However, ENTSO-E notes that UPS-based architectures can instantaneously disconnect hundreds of megawatts during minor grid disturbances, potentially amplifying problems for safe grid operation and power quality.
This makes disturbance response an important part of future connection assessment. Grid operators will need to understand whether large data centre loads can ride through faults where required, avoid sudden large-load disconnection, and recover in a controlled way after an event.
Grid-safe before grid-supporting
ENTSO-E identifies several technical areas under discussion for updated connection requirements, including fault ride-through, RoCoF withstand capability, ramp-rate limits, voltage control, reactive power, oscillation damping and post-fault active power recovery.
This points to an important direction: data centres need to become grid-safe before they can become grid-supporting.
There is real potential for data centres to become active grid resources. UPS batteries, cooling systems, thermal storage, controllable workloads and on-site generation could support flexibility, resilience and market participation. But this future depends on a more basic requirement first: data centres must be able to demonstrate safe and predictable behaviour under real operating conditions.
The discussion around AI data centres is becoming more specific. As large, software-driven loads connect to the grid, visibility into real-world behaviour will become an important part of connection confidence.
For data centres, the next stage of connection readiness will be about proving that their behaviour can be observed, measured and trusted.
Reference:
– NERC, Characteristics and Risks of Emerging Large Loads, July 2025.
– ESIG, Large Load Interconnection Performance Requirements, February 2026.
