Something quietly significant is happening to electricity, and it doesn't yet have the world-in-peril focal point that tends to attract attention. There are no rolling blackouts, no emergency declarations, no fuel lines stretching around the block. But for the first time in a generation, the world is genuinely uncertain whether it can produce enough power to support the next wave of economic growth.

AI data centers, electric vehicles, heat pumps, and industrial electrification are driving a rapid increase in electricity demand, in which any one of these trends would be significant on its own. Together, they are pushing against the limits of a grid built for a different century.

The response so far has focused largely on supply, with governments revisiting nuclear power, utilities proposing massive transmission expansions, and investors pouring money into a new generation of energy technologies ranging from geothermal energy to grid-scale storage. Underlying all of these efforts is a straightforward assumption: if demand is rising, the answer is more electricity. Given how tightly economic growth and energy consumption have been linked throughout modern industrial history, it's a conclusion that feels almost self-evident.

It mostly is. But it's also incomplete—and fixating on it risks missing something important.

The conversation dominating boardrooms, legislatures, and engineering departments is almost entirely about supply. How do we produce more? What we're not asking with nearly enough urgency is how intelligently we use the infrastructure that already exists. Because one of the largest untapped energy resources in the world isn't buried underground, waiting to be extracted.

It's the built environment itself, and it's already here.

The United States uses about 4,000 terawatt-hours of electricity every year, and demand projections for the coming decades describe growth that, if they materialize, will look unlike anything the country has experienced since the postwar industrial boom. The infrastructure required to meet that demand will be expensive, slow, and politically complicated to build. None of that is surprising. What is surprising is a figure that tends to get buried in the footnotes of the same reports forecasting all that demand growth: the Department of Energy has estimated that advanced building controls could reduce total U.S. energy consumption by more than 3%.

That number sounds modest enough to ignore, but it shouldn't be. Applied across the American economy, 3% works out to roughly 120 terawatt-hours per year—more than many entire countries consume, enough to power around 10 million homes, and equivalent to approximately 14 gigawatts of continuous generation capacity. To put that figure in perspective, recovering 14 gigawatts from existing buildings would be equivalent to bringing roughly fourteen large nuclear reactors or more than twenty typical utility-scale natural gas plants online. Building that much generation capacity would require years of permitting, financing, construction, and grid interconnection work, all while navigating queues that are already backlogged across much of the country.

The electricity represented by that 3% already exists. It's being generated right now, flowing through the grid right now. It's just being wasted, quietly and continuously, inside buildings that have no particular interest in using it wisely.

The conventional view treats buildings as consumers of electricity, but a building is not a passive load. It is a complex physical system, continuously balancing thermal dynamics, equipment operation, occupancy patterns, weather conditions, and energy flows. Collectively, buildings constitute one of the largest controllable systems in the economy. Yet despite consuming most of the electricity in developed countries, the majority still operate with remarkably little intelligence, reacting to conditions after they occur rather than understanding, predicting, and optimizing their behavior in advance.

Walk into almost any large commercial building and the energy story is some version of the same. Equipment runs on schedules that were programmed years ago and haven't been meaningfully updated since. Heating and cooling systems work against each other in ways that would seem almost comedic if the consequences weren't real—one system spending energy to cool a space while another spends energy to warm it, both running because a calendar says so rather than because anyone needs either one. The building doesn't know what the weather will do this afternoon. It doesn't know whether electricity is cheap or expensive right now, or whether the grid is under stress. It has no model of its own future state, no ability to anticipate what it will need, and no mechanism for coordinating its own systems toward a coherent outcome. It reacts to conditions after they occur rather than preparing for them, and as a by-product wastes a substantial amount of energy.

For most of the past century, this was an understandable arrangement. Electricity was cheap, abundant, and rarely constrained, making inefficiency little more than a line item on an operating budget. Today, electricity is becoming something different: a strategic resource. Data center developers in parts of North America and Europe are sitting on fully financed projects that cannot secure enough grid capacity to move forward. Utilities are fielding demand requests at levels they have not seen in decades, while some projections suggest AI infrastructure alone could consume as much electricity as Japan by the end of this decade. In that environment, the energy wasted inside commercial buildings stops looking like a minor operational inefficiency and starts looking like stranded economic capacity.

What's emerging now is a fundamentally different way of thinking about what a building can be. Advances in modeling, computation, and artificial intelligence make it possible for buildings to understand themselves as dynamic physical systems rather than collections of independent equipment. For the first time, a building can maintain a living model of its own behavior—continuously reasoning about thermal dynamics, occupancy, weather, equipment performance, and energy costs as parts of a single interconnected system. Instead of reacting to conditions after they occur, an intelligent building can continuously model its future, anticipate changing circumstances, and optimize its behavior in real time. The result introduces the ability to convert the latent flexibility of the built environment into infrastructure capacity—shifting loads, coordinating with the grid, and adapting to changing conditions autonomously. In effect, the building becomes a form of physical intelligence embedded directly within the energy system.

A commercial building that reduces its energy consumption by 30%—a target achievable with existing technology—isn't merely saving energy; it's creating capacity. A building consuming 1.5 million kilowatt-hours annually would recover roughly 450,000 kilowatt-hours through that reduction, representing approximately $45,000 to $90,000 in annual energy savings depending on local electricity prices. Across a portfolio of 10,000 buildings, that scales to more than 4.5 terawatt-hours per year and between $450 million and $900 million in avoided energy costs. From the grid's perspective, that recovered electricity is functionally equivalent to new supply, creating roughly 500 megawatts of continuous capacity—enough to support hundreds of thousands of homes or several large AI data centers—without building a power plant, transmission line, or substation.

The energy industry calls resources like this virtual power plants. It's a useful concept, if a slightly bloodless one. But it may actually undersell what intelligent buildings represent. A power plant is a machine for producing electrons. An intelligent building is something closer to a machine for producing economic capacity—freeing existing electricity for uses that generate more value, transforming waste into availability, and doing it continuously, at scale, using infrastructure that's already built and already connected.

For a century, expanding the grid meant adding physical assets: power plants, transmission lines, substations, and generation capacity. Those investments remain essential, and the world will need an enormous amount of new infrastructure over the coming decades. But increasingly, grid expansion will also come from intelligence applied to physical infrastructure. The next generation of capacity will not come exclusively from building more assets; it will come from enabling existing assets to understand themselves, predict their future behavior, and coordinate their operation autonomously. In that sense, intelligence is becoming a new category of infrastructure.

The countries that lead the next era of economic growth won't simply be the ones that generate the most electricity. They'll be the ones that extract the most value from every electron they produce, treating intelligence as a form of infrastructure and the built environment as something more than a load on the grid. Office towers, warehouses, hospitals, and data centers collectively consume most of the world's electricity, yet they remain one of the largest untapped sources of capacity in the energy system.

The dominant narrative of the energy transition is that society needs more power, and it does. But it also needs more intelligence applied to the physical systems that already consume that power. The fastest way to expand the effective capacity of the grid may not be to build more of it, but to make the infrastructure already connected to it intelligent enough to operate at its full potential.

The fastest power plant we build this decade may not be a power plant at all. It may be the intelligence layer that transforms millions of buildings from passive consumers of electricity into active participants in the physical economy.