The AI Revolution Is Changing the Rules of the Data Center
For decades, the mission of a data center was straightforward: provide enough electrical power, sufficient cooling capacity, and highly available IT infrastructure to support business applications.
The assumption was simple. Compute demand dictated infrastructure requirements, and the energy system’s job was to satisfy those demands. Artificial Intelligence has fundamentally changed this equation.
Modern AI clusters consume unprecedented amounts of electricity. A single GPU server can require several times more power than a traditional enterprise server. Entire AI racks may consume over 100 kW, while next-generation AI facilities are being designed with capacities measured in hundreds of megawatts. For the first time in the history of computing, energy is becoming the primary constraint on computational growth.
Around the world, utilities struggle to provide new grid connections. Renewable generation introduces periods of energy abundance followed by periods of limited availability. Electricity prices fluctuate throughout the day, while battery energy storage systems are becoming standard components of modern facilities.
The challenge is no longer simply building larger data centers. The challenge is learning how to operate them intelligently.
A Fundamental Change in Thinking
For nearly forty years, the industry has invested enormous resources in adapting the energy infrastructure to meet increasing compute demand.
We built larger substations. We installed bigger UPS systems. We added diesel generators. We expanded cooling capacity. We upgraded transformers. Whenever computational demand increased, the energy system expanded accordingly.
But perhaps we have been asking the wrong question. Instead of continuously adapting the energy infrastructure to satisfy every computational demand, what if certain computational workloads could intelligently adapt themselves to the available energy?
This represents a profound shift in the way future data centers will operate.
Not Every Workload Is Equally Urgent
One of the most interesting observations emerging from AI infrastructure is that not every workload has identical timing requirements. Some applications are mission critical. Database transactions, industrial control systems, financial applications, healthcare systems, and real-time AI inference often require immediate execution.
Others do not.
These jobs may tolerate delays ranging from minutes to several hours without affecting business operations. That flexibility creates an entirely new opportunity. Instead of treating electricity as an unlimited resource, we can begin matching computational demand to energy availability.
From Energy Management to Compute Management
Traditional Energy Management Systems (EMS) optimize energy generation, storage, and consumption.
Data Center Infrastructure Management (DCIM) platforms monitor power distribution, cooling systems, and physical assets. Hypervisors and Kubernetes orchestrate computational workloads. Each platform performs its own role exceptionally well. Yet none of them was originally designed to coordinate all of these domains simultaneously. As AI facilities continue to grow, a new layer of intelligence becomes necessary. A layer capable of making real-time decisions based on both computational priorities and energy availability. I believe this new category deserves its own name.
Energy Compute Management System (ECMS)
An Energy Compute Management System (ECMS) is a software platform that continuously optimizes computational workloads according to available energy resources while maintaining operational service levels and business priorities. Rather than managing only energy or only compute, an ECMS manages the relationship between them.
How an ECMS Works
An ECMS continuously receives information from multiple systems across the facility.
These may include:
The ECMS analyzes all these inputs and continuously determines the optimal allocation of computational resources. Its objective is simple: Execute the right workload at the right time using the most appropriate energy source.
Practical Examples
The concept becomes easier to understand through practical scenarios.
Running AI Training When Solar Energy Is Available
Instead of starting AI training immediately, the ECMS may postpone execution until midday, when on-site solar production reaches its peak. The same computational result is achieved while consuming significantly more renewable energy.
Battery-Aware Computing
When battery storage is fully charged, the ECMS may accelerate non-critical computational workloads. If battery reserves become limited, lower-priority tasks are postponed until additional renewable energy becomes available.
Demand Response Participation
Utilities increasingly encourage large consumers to reduce demand during periods of grid stress. Rather than shutting down critical services, the ECMS can temporarily pause AI training, backup operations, or analytics workloads while maintaining business continuity. The data center effectively becomes an intelligent participant in grid stability.
Geographic Workload Optimization
Organizations operating multiple data centers may dynamically relocate computational workloads. An AI training task could execute in Israel during periods of abundant solar generation, then move to Northern Europe when wind generation becomes more favorable.
Energy availability becomes another scheduling parameter alongside latency and computational capacity.
Dynamic GPU Power Optimization
Not every AI workload requires maximum GPU performance. The ECMS can temporarily reduce GPU power limits during periods of constrained energy availability and restore full performance once additional energy becomes available.
This creates valuable operational flexibility without interrupting running workloads.
Why This Matters
The AI revolution is placing unprecedented pressure on electrical infrastructure. Many regions already experience long waiting periods for new grid connections. At the same time, renewable generation continues to expand, introducing natural variations in energy availability throughout the day. Battery storage systems are becoming standard components of modern facilities. Electricity markets are becoming increasingly dynamic. Future data centers must therefore optimize not only computational performance but also energy utilization. This transformation requires a new operational philosophy. Energy is no longer simply a utility. It becomes an active scheduling parameter.
The Next Evolution of EMS
As someone who has spent many years working with SCADA and Energy Management Systems across industrial facilities and critical infrastructure, I believe we are witnessing the next major evolution of energy management.
Traditional EMS platforms answered questions such as:
The next generation of systems must answer an entirely new question:
Given the available energy, what should the data center compute right now?
This is a fundamentally different challenge that requires deep system integration capabilities across energy systems, IT infrastructure, and orchestration platforms. It requires integrating operational technology, energy systems, IT infrastructure, AI scheduling, battery storage, renewable generation, and business priorities into a single decision-making platform.
That platform is what I define as an Energy Compute Management System (ECMS).
Looking Ahead
Artificial Intelligence will continue increasing global electricity demand at an extraordinary pace.
At the same time, renewable generation, battery storage, and increasingly constrained electrical grids will make energy availability more dynamic than ever before. Tomorrow’s most efficient data centers will not simply consume electricity. They will continuously adapt their computational behavior to available energy.
I believe the emergence of Energy Compute Management Systems (ECMS) represents the next logical step in the evolution of digital infrastructure.
Just as SCADA transformed industrial automation, EMS transformed energy optimization, and DCIM transformed data center operations, ECMS has the potential to become the intelligent layer that brings these disciplines together.
At Contel, this vision aligns naturally with our long-standing expertise in industrial automation, SCADA, Energy Management Systems, and mission-critical infrastructure. We believe the future of AI-ready data centers lies not only in faster processors or larger electrical connections, but in intelligent coordination between energy and computation. Developing technologies that bridge these two worlds is a strategic direction we are actively pursuing.
The future of computing will not be defined solely by processing power. It will be defined by how intelligently we manage the energy that powers it.
By Shai Gershon
