New generation and grid connections will take years to deliver. Reducing waste from buildings and data centers offers a faster way to ease the pressure.
The energy challenge posed by AI is usually framed as a race to expand generation. Larger models and denser computing systems require vast amounts of electricity, while the facilities that house them need equally formidable cooling infrastructure. As AI spreads through industries, the availability of power appears to dictate how quickly innovation can proceed.
But supply is just one part of the equation. Around 30% of the energy consumed in the built environment is wasted after purchase, exposing a large reserve of capacity hidden inside inefficient buildings and equipment. Unlocking that capacity depends on having a clear picture of where energy is going—and why.
By combining AI with data from connected devices, software and automated controls, organizations can coordinate energy use across buildings, industrial sites, data centers, and transport systems. AI therefore adds to the pressure on power networks while also providing some of the tools needed to relieve it.
“When AI is integrated with deep technical expertise, it can reduce energy use by up to 25% and carbon emissions by as much as 40%,” says David Brennan, vice president of innovation at Trane Technologies, a global climate innovator. “That creates an opportunity to turn energy inefficiency into savings while supporting more sustainable growth.”
Making existing capacity work harder
Demand for the computing power behind AI is climbing at remarkable speed. Eighty-eight percent of surveyed organizations used AI in at least one business function in 2025, up from 78% 12 months earlier, according to Stanford University’s 2026 AI Index. Global corporate investment in the technology reached $581.7 billion over the same period, an increase of 130% in a single year.
Each deployment ultimately relies on physical infrastructure capable of powering advanced processors and carrying away the heat they produce. The International Energy Agency expects data center electricity consumption to nearly double from 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030. More generation and grid connections are key to meeting this surge—but they take time to plan and build, leaving operators under pressure to make better use of what’s already available.
Brennan says that requires a “systems approach,” with computing loads, energy, and cooling managed as parts of the same environment. Improving one component in isolation can merely transfer the strain. Installing more efficient servers, for instance, achieves little if higher heat density requires more cooling, which absorbs the overall savings.
Brennan says that requires a “systems approach,” with computing loads, energy, and cooling managed as parts of the same environment.
That logic underpins what Brennan describes as “connected intelligence,” which brings controls, software, services, and live operational data into the same decision-making process. In the context of a data center, this could mean a connected cooling solution predicting the heat generated by an approaching computing load and adjusting in advance, versus a conventional system that responds only once temperatures start to rise.
Catching waste before it compounds
The same integrated approach that allows a building to anticipate demand can also improve how it is maintained. Instead of waiting for equipment to fail or performance to deteriorate enough to trigger an alarm, building teams can track energy use and operating conditions continuously. AI can identify unusual behavior, adjust settings automatically, and flag problems before they lead to wasted energy or disrupted service.
That predictive capability also applies to day-to-day building performance. By continuously analyzing inputs such as occupancy, weather forecasts, and energy use, AI-enabled controls can predict how a building’s thermal needs are likely to shift over the course of the day. If outdoor temperatures are expected to climb, for example, the system can make gradual adjustments in advance rather than reacting only after indoor conditions begin to change. The result is continued comfort, smoother performance, lower waste, and more efficient operation overall.
In a pilot project at three Amazon Grocery fulfillment centers, software developed by BrainBox AI—acquired by Trane Technologies in 2025—used real-time data to adjust the facilities’ climate systems automatically as conditions changed. The project reduced energy consumption by nearly 15%, more than twice its original target, while maintaining sites’ performance. Amazon now plans to extend the technology to more than 30 U.S. locations.
“At that scale, even a relatively small efficiency gain translates into substantial energy savings,” explains Brennan.
The human expertise behind the system
Technology can identify issues early, recommend responses, and help teams resolve problems remotely. When physical inspection or repair is needed, engineers, and technicians use those AI-driven insights to diagnose faults faster and restore equipment more efficiently.
Trane Technologies is applying AI to help its own teams do that more quickly. ARIA, a digital assistant for field technicians, provides technical guidance while they inspect and service equipment, helping them narrow down the cause of a fault and determine the next step. AI shortens the route to the relevant information, leaving the human to decide what action is ultimately needed.
Researchers at the BrainBox AI Trane Technologies AI Lab in Montreal are also exploring physics-informed neural networks, a form of AI that combines patterns learned from data with established scientific rules. These models incorporate physical principles—such as those governing heat transfer and changes in indoor temperature—rather than relying on historical data alone.
Making room for growth
The next phase of AI infrastructure will require more power, cooling, and physical space. Yet, new generation and grid capacity will not arrive quickly enough to carry the full burden. Some of the room for growth will have to come from using existing assets more intelligently.
That, Brennan says, places a different sort of demand on the buildings, industrial sites, and data centers that both use AI and provide the infrastructure behind it. “Energy demand management, advanced thermal management solutions, and AI-driven optimization all have a role to play, but the real gains come from coordinating and optimizing them across the system.”