Data centers face an unexpected crisis: they have enough servers and graphics processing units (GPUs) to expand, but not enough electricity to power them. As artificial intelligence adoption accelerates, the global demand for data center electricity is projected to roughly double from 485 terawatt-hours (TWh) in 2025 to 950 TWh by 2030, according to the International Energy Agency (IEA). Consumption from AI-focused data centers alone is expected to triple over the same period.

This power crunch is reshaping how the industry thinks about growth. Facility operators can acquire land, install servers, and deploy cutting-edge GPUs, but without sufficient and reliable electricity, they cannot add computing capacity. The constraint has shifted from hardware availability to grid capacity and power supply reliability.

Why Is Power Becoming the Primary Limit for Data Center Growth?

The explosion in AI workloads has created an unprecedented demand surge. Large language models (LLMs), which are AI systems trained on vast amounts of text to generate human-like responses, require enormous computational resources. Training and running these models consumes electricity at scales that traditional data centers were never designed to handle. Even colocation facilities with available physical space and equipment cannot expand without access to more power from the grid or on-site generation.

This bottleneck has caught the attention of technology companies and energy optimization firms. PRF Technologies, a company that develops AI-driven energy management software, announced that its GridFeed platform is exploring the data center market as a strategic growth opportunity. GridFeed was originally built to help renewable energy operators maximize the value of every megawatt they produce, but the company believes the same optimization logic can help data center operators make better use of limited power capacity.

How Can Data Centers Optimize Power Use to Keep Growing?

Intelligent energy management is emerging as a competitive advantage. Rather than simply consuming whatever power is available, data center operators can use software to coordinate when and how they run compute workloads, manage battery storage, and respond to grid conditions. Here are the key strategies being explored:

• Workload Shifting: Moving compute tasks to times when power is cheaper or more abundant, such as when renewable energy generation peaks or electricity prices drop.
• Energy Cost Optimization: Analyzing real-time electricity prices and grid conditions to determine the most cost-effective times to run different workloads.
• Battery Charge and Discharge Strategies: Using on-site battery storage to absorb power when it is cheap or plentiful, then discharging during peak demand or high-price periods.
• Demand Response: Coordinating with utilities to reduce power consumption during grid stress events in exchange for financial incentives.
• Real-Time Power Allocation: Dynamically deciding which workloads get priority access to available power based on business value and operational requirements.

GridFeed's optimization engine follows a "predict, simulate, optimize and recommend" process. For data centers, this means using weather forecasting to anticipate cooling demand, predicting on-site renewable generation, monitoring grid availability, and then recommending the action expected to produce the best overall business outcome while protecting operational reliability.

"The growth of AI is driving a sharp increase in the power that data centers need, and access to electricity and grid capacity is becoming a limiting factor for many operators. We believe that intelligent coordination of compute, storage and energy will become an important competitive advantage for data center operators," said Efraim Cohen-Arazi, Interim Chief Executive Officer of PRF Technologies.

Efraim Cohen-Arazi, Interim Chief Executive Officer, PRF Technologies

What Role Does Cooling Play in the Power Crisis?

Cooling is a hidden driver of data center power consumption. High-density GPU servers generate enormous amounts of heat, and removing that heat efficiently is critical for both performance and reliability. Conventional air-cooling systems are increasingly unable to handle the thermal load, pushing operators toward more advanced cooling technologies.

South Korea's state-run Korea Research Institute of Chemical Technology and SK Innovation have partnered to develop next-generation cooling solutions using two-phase refrigerants. In two-phase cooling, a liquid refrigerant absorbs heat from servers, vaporizes, and then condenses back into liquid, transferring large amounts of heat in the process. This approach is far more efficient than air cooling but requires specialized refrigerants that can absorb heat rapidly while remaining stable over extended use.

The two organizations plan to use machine learning to predict the thermal and physical properties of candidate refrigerant substances, then synthesize and test the most promising ones. The collaboration will span from initial refrigerant discovery through synthesis, performance verification at the server and rack level, and culminate in validating cooling performance in a 500-kilowatt-class AI data center environment. By improving cooling efficiency, data centers can reduce the power needed to remove heat, freeing up more electricity for actual computing work.

What Does This Mean for the Data Center Industry?

The power constraint is forcing a fundamental shift in how data centers operate. Rather than viewing energy as an unlimited commodity, operators are beginning to treat power as a scarce resource that must be managed strategically. This opens opportunities for software companies, materials scientists, and energy providers to collaborate on solutions.

PRF Technologies has announced plans to establish GridFeed as a standalone, wholly owned subsidiary, giving the platform more flexibility to pursue partnerships across the data center ecosystem. The company is actively exploring collaboration opportunities with workload orchestration providers, data center infrastructure management (DCIM) software vendors, energy management system (EMS) platforms, facility management providers, storage and uninterruptible power supply (UPS) vendors, hyperscale operators, colocation facilities, and utilities.

The convergence of AI adoption, power scarcity, and advanced cooling technology suggests that the next wave of competitive advantage in data centers will belong to operators who can intelligently balance compute demand, energy availability, and thermal management. As the IEA projects electricity demand to nearly double by 2030, solving the power puzzle has become as important as acquiring the latest hardware.