The Data Center Frontier Executive Roundtable features insights from industry executives with lengthy experience in the data center industry.

Here’s a look at the Q3 2026 insights from Pierre-Adrien Bel, Product Manager, EMEA, Rehlko.

Pierre-Adrien Bel is currently a Product Manager at Rehlko, responsible for generator sets above 700 kW. A mechanical engineering graduate, he has more than 15 years of experience in the power generation industry. He began his career in the oil and gas sector and in shipbuilding before joining Rehlko's engineering department, where he worked on generator set installation projects. He later advanced to Project Manager for large-scale projects and subsequently served as a Business Engineer focused on advanced applications, including data centers. For the past three years, he has been a member of Rehlko's product management team. With a strong commitment to sustainable development and greenhouse gas emissions reduction, Pierre-Adrien combines his work on the development of high-performance generator sets with the advancement of sustainable energy solutions that support the transition to a lower-carbon future.

Data Center Frontier: What is AI infrastructure exposing that the data center industry could get away with overlooking at lower densities — and which traditional design or redundancy assumptions now need to be reconsidered?

Pierre-Adrien Bel, Rehlko: AI is exposing that traditional redundancy models can look sufficient on paper while still leaving gaps in dynamic system performance. Conventional generator systems were designed around relatively stable loads and isolated transient events, with time to stabilize and recover between disturbances. Large AI training clusters change that operating profile because thousands of GPUs may ramp power demand in synchronized, repetitive cycles, significantly reducing or eliminating those recovery windows.

That means installed megawatts and redundancy alone do not fully define resilience. Operators also need to understand voltage stability, frequency control, dynamic response time and available operating margin under sustained stress.

It also challenges the assumption that oversizing or adding more mitigation layers will solve the problem. UPS systems and battery storage can help smooth fast transients, but they do not eliminate the underlying load behavior. Some of that variability still reaches generators, alternators and controls, which makes validated system-level performance increasingly important.

Data Center Frontier: As data centers become larger, denser and more power-constrained, where is the definition of “critical infrastructure” expanding — and are the interfaces between power, storage, cooling, controls and other systems becoming the real reliability risk?

Pierre-Adrien Bel, Rehlko: The definition of critical infrastructure is expanding to include the interfaces and controls that coordinate individual assets. A generator, UPS or battery system can each perform within specification, but the overall power system can still become unstable if those assets are not responding together as intended.

That becomes especially important in hybrid and islanded architectures. Battery systems can respond quickly to fast transients and support voltage and frequency stability, while generators provide sustained power over longer periods. The power management system has to coordinate real power, reactive power, voltage and operating setpoints across those different technologies.

Those interfaces can absolutely become a reliability risk. If one asset responds too slowly, reaches a protection limit or behaves differently than modeled, the fleet-level response can become unstable even when each individual asset appears acceptable in isolation.

That is why controls, communications and system coordination need to be treated as part of the critical power architecture, not as secondary layers around the equipment.

Data Center Frontier: With time-to-power compressing development schedules, where can operators safely accelerate deployment — and where do commissioning, testing and operational visibility need to become more rigorous rather than less?

Pierre-Adrien Bel, Rehlko: Operators can safely accelerate deployment by moving more modeling and validation work earlier in the project. Validated generator models, digital twins and hardware-in-the-loop testing can help teams evaluate equipment interactions, control logic and system behavior before physical commissioning begins. At Rehlko, this has driven increased investment in generator characterization and model validation, using measured performance data to develop system-level simulations that more accurately reflect real-world operating conditions.

Where operators should not reduce rigor is commissioning and system-level validation. A conventional resistive load-bank test can confirm that equipment accepts load, but it does not fully represent AI-driven conditions such as rapid load steps, load rejection, reactive power requirements or sustained variability.

For prime-power and islanded applications, testing should progress from individual assets to subsystems and then to fleet-level validation under the power management system. That final stage needs to confirm coordinated response across generators, BESS, inverters and controls while maintaining stable voltage and frequency.

The goal is to shift more validation upstream so commissioning confirms a well-understood system rather than becoming the first time teams see how the full architecture behaves under AI-style load conditions.

Keep pace with the fast-moving world of data centers and cloud computing by connecting with Data Center Frontier on LinkedIn, following us on X/Twitter and Facebook, as well as on BlueSky, and signing up for our weekly newsletters using the form below.