Katie McGinty: The energy economy's biggest waste problem is already inside the system | Fortune

The Hidden Power Plant Nobody's Building

A striking argument is making the rounds in energy circles: the cheapest, fastest new source of electricity capacity isn't a new power plant at all — it's the energy already being wasted inside buildings, grids, and industrial systems. That's the thesis Katie McGinty, an executive at Johnson Controls, lays out in a recent Fortune commentary, timed to projections that data center electricity demand could climb toward 945 terawatt-hours by 2030, largely driven by AI.

The core claim is simple but consequential: roughly a third of electricity generated in the U.S. is lost to inefficiency — through outdated HVAC systems, leaky buildings, aging industrial equipment, and grid transmission losses. If even a fraction of that waste were recovered through efficiency upgrades, it could offset a meaningful share of the new demand AI infrastructure is expected to create, without waiting years for new power plants, transmission lines, or nuclear projects to come online.

Why This Matters Beyond the Power Grid

While this finding is framed around energy policy, its implications ripple directly into the AI industries now driving that demand curve — including video generation, image generation, voice synthesis, and multimodal models. These technologies are computationally expensive by design: training and running large multimodal systems requires sustained, high-density power draw at data centers, and that demand is compounding as generative tools scale toward mainstream consumer and enterprise use.

The efficiency argument matters for this sector in a very practical sense. If AI companies and data center operators face power constraints — whether from grid bottlenecks, permitting delays, or rising electricity costs — the pace of scaling new models could be throttled by infrastructure limits rather than algorithmic ones. That's a real risk already being discussed among hyperscalers, several of which have turned to on-site power generation, nuclear partnerships, and long-term power purchase agreements specifically because grid capacity hasn't kept up with AI's growth curve.

Efficiency as a Quiet Enabler of AI Growth

McGinty's framing suggests a less glamorous but potentially faster lever: efficiency retrofits in the buildings and industrial systems that make up the broader grid, freeing up capacity that could indirectly support data center expansion. This isn't a technology fix for AI itself, but an infrastructure argument that the industry's growth trajectory depends on more than chip supply and model architecture — it depends on the physical grid having room to absorb new load.

The Takeaway

As generative AI tools for video, image, and voice continue multiplying, the conversation around their sustainability is shifting from carbon footprints to raw electron availability. Whether efficiency gains can genuinely offset AI's growing appetite remains to be tested, but the argument underscores a broader point: the next constraint on AI's growth may not be compute, but power.