By: Sam Maleki, Principal Advisor PPC & Data Center Controller, EdgeTunePower Inc
Dr. Sam Maleki is an industry leader specializing in data center integration and large inverterbased resource (IBR) interconnections within the North American power grid. At RMS Energy, he focuses on advancing grid reliability, renewable integration, and secure power solutions for emerging technologies. As an Adjunct Professor at McMaster University, Dr. Maleki contributes to education and research in modern power systems. He is also the co-founder of ElectroMentors, a global training platform that empowers electrical and computer engineers with practical skills to tackle today’s grid challenges. His work bridges industry, academia, and innovation to shape the future of reliable and sustainable energy systems.
Power systems were built on simple fundamentals: loads must be served, and generating units must maintain network stability and reliability. For decades, this meant massive synchronous generators with plenty of inertia and spinning reserve. Loads were predictable, ramping up in the morning, peaking in the afternoon, and gradually declining overnight. That predictability allowed us to plan effectively.
But the grid has changed. Large-scale inverter-based resources (IBRs), solar, wind, and energy storage, now dominate in many regions, sometimes providing nearly all the system’s power. Unlike synchronous generators, these units are harder to plan around. Wind and solar output vary constantly, and their controls sometimes produce unexpected behaviors: oscillations and instabilities such as SubSynchronous Control Interaction (SSCI), SubSynchronous Frequency Response (SSFR), and low-frequency swings driven by solar radiation changes.
Commissioning these plants has also revealed challenges. We’ve seen delays because simulation models don’t always match hardware. For example, Power Plant Controllers often behave very differently than modeled, introducing communication delays of several seconds, where simulations assumed just 200 ms. Transient responses, too, often diverged from expectations.
What went wrong?
We trusted simulation files as though they perfectly represented hardware. They didn’t. Models were often inaccurate, sometimes not even close. This led to the rise of hardwarein-the-loop (HIL) testing and unit-level model validation. Vendors were asked to prove that lab-tested equipment matched simulations. But unit-level validation alone isn’t enough. Entire-plant model validation is critical. We need to understand not just individual devices, but how communications, controllers, and protections interact together. Hybrid plants, battery energy storage systems (BESS), photovoltaic (PV), and wind, can create complex dynamics we won’t see in isolated unit tests.
Some argue simulations are sufficient. But in reality, they’ve failed us many times. Others say we can fix issues on-site. True, but at what cost? I’ve seen commissioning delays stretch over a year, and plants curtailed or shut down for months. The lost revenue often reaches millions.
Now, we face a bigger challenge: hyperscaler data centers. And if we repeat the same mistakes, trusting incomplete models, we risk even more. I call these hyperscalers dragons. They are enormous: single campuses demanding up to 1 GW at a single point of interconnection. Data centers in Ireland already consume over 20% of the country’s electricity, forcing the grid operator to pause new connections in some regions. For perspective, the entire city of Toronto consumes around 5 GW. Imagine 20% of that at one site. These dragons are racing to connect, and the winners will be those who move fastest. But getting permission to connect is not the same as actually connecting. Utilities have learned from IBR projects that models and reality often don’t align. Do we really believe they’ll allow loads five to ten times larger than anything before to connect without far more rigorous assurances?
Utilities will demand proof of stable, predictable operation: voltage ride-through, frequency ride-through, and controlled load ramps. The grid cannot tolerate sudden swings like a data center load jumping from 300 MW to 800 MW in seconds, or disappearing entirely. Losing 1 GW instantly could crash system frequency (think about Rate of Change of Frequency [RoCoF]).
So, how do we train these dragons?
- Accurate Models Now We can no longer rely solely on software models. HIL testing must be extensive, ensuring hardware matches models exactly.
- Digital Twins Entire facilities should be replicated in realtime simulation environments, using physical controllers and protection units in the lab. Full-scale UPSs and transformers don’t need to be replicated physically; aggregated models suffice. With this setup, we can test countless scenarios and walk into commissioning with confidence.
- Commissioning & Troubleshooting Digital twins also provide a powerful tool for troubleshooting. With IBRs, disconnections already cost millions. For hyperscaler data centers, with their “six nines” reliability (99.9999%), outages could cost billions.
The question is simple: Are we willing to risk billions of dollars by letting untrained dragons loose on our grid? Or will we take the time to train them, today, so they can operate in harmony with the power system?
