To Save A Dying Planet: A Case for Restraint

Annan Nippita ’29

I like to think I use energy responsibly. I turn off lights when I leave a room, keep my showers short, and try not to leave things plugged if they don’t need immediate power. A model citizen. Practically saintly.

Then I looked up how our electricity grid actually works, and I felt a lot less proud of myself.

Here's what nobody explains well: by the time electricity gets from a power plant to your outlet, a massive share of it is already gone. Fossil fuel plants — which still generate about 60% of U.S. electricity — operate at roughly 33% efficiency. That means for every three units of energy burned, only one actually becomes electricity. The other two are released as waste heat into the air. 

Right there, before a single light is switched on, two-thirds of the fuel's energy has vanished. This is offset by other, less commonly used power sources such as wind, solar and hydro power, but add in transmission losses — electricity traveling hundreds of miles through power lines — and the U.S. Energy Information Administration estimates we lose close to 30% of generated electricity before it ever reaches a building. Altogether, the U.S. power sector emits around 1.5 billion metric tons of CO₂ every year, and hundreds of millions of those tons are for electricity that ends up heating the atmosphere rather than powering anything.

To put that another way: if all that wasted electricity were redirected, it could power roughly 40 million homes a year. Or, if you want something more local, it could power every apartment in the five boroughs for roughly 4 decades, without burning a single additional gram of coal.

So we're losing an enormous amount of energy before we even get to how we use it. Fine. That's a grid engineering problem. That’ll be fixable, eventually. But here's where it gets more uncomfortable.

We've been getting more efficient for decades, and we've been using more energy every decade. That's not a coincidence. It's a pattern so well-documented that economists have a name for it: the Jevons paradox. It was first identified in 1865 when a British economist noticed that more efficient steam engines didn't reduce coal consumption, but instead made our coal-powered industries so cheap that demand exploded. The same thing has happened with nearly every energy efficiency improvement since. More efficient cars meant more people drove longer distances. More efficient appliances meant people bought more of them and ran them longer. More efficient data centers meant companies built exponentially more of them for the same cost.

Between 1990 and 2022, U.S. electricity consumption per household actually declined by about 8%. That’s kind of impressive. It meant us switching from incandescent lightbulbs to LEDs, better insulating our homes, and more efficient transportation. But over that same period, total U.S. electricity consumption rose by about 30%, because the number of electricity-using devices, buildings, and industrial processes grew faster than efficiency improved. This is partly, too, due to the fact that the number of households in the U.S. grew 40%. We got better at using energy per unit, but the same time, we used many more units.

This is the demand problem, and it's the one that a conversation about "clean energy" almost never addresses directly.

Right now, the dominant assumption in policy, in the media, and in casual conversation is that the solution to our energy problem is to replace dirty supply with clean supply. Solar panels instead of gas plants. Wind instead of coal. Electrify everything. And to be clear: yes, obviously, we should do all of that. Renewable energy now costs less per kilowatt-hour than new fossil fuel plants in most of the world, and especially powerhouse countries (no pun intended) like the U.S. and China. The economics have flipped. But here's the catch: renewables don't make demand disappear. If anything, cheaper electricity makes demand grow faster.

The U.S. currently uses about 4,000 terawatt-hours of electricity per year, or enough electricity to power an LED for a bit more than three times the age of the universe. Of that, about 24% was generated from renewable sources like wind and solar power. To fully decarbonize our grid with wind and solar energy would require roughly tripling or quadrupling that generation capacity, not to mention having to account for storage, variability, and the electrification of cars, heating, and industry. The U.S. Energy Information Administration projects that electricity demand will grow by 50% by 2050 even under ambitious clean-energy scenarios. We're trying to fill a bucket while the bucket is getting bigger.

And then there's AI.

Artificial intelligence has become the defining example of demand without restraint. Training a single large language model — the kind of model behind AI tools like ChatGPT, Claude, Gemini, and Copilot — consumes roughly 1,300 megawatt-hours of electricity. That's enough to power about 120 New York City apartments for a year, burned through in a matter of weeks. But training is actually the smaller problem. Running these models at scale — billions of queries per day — currently accounts for around 5% of total US consumption. The Electric Power Research Institute projects that AI data centers alone could represent 9-10% of total U.S. electricity demand by 2030.

To meet that demand, tech companies aren't waiting for wind farms. They're recommissioning gas turbines and coal plants that hadn't been operational in years. Mothballed infrastructure is being brought back online because AI needs power now, at scale, and with the kind of reliability that intermittent renewables currently can't guarantee without massive storage investments. We are, in a very real sense, burning coal so that language models can write people's cover letters.

None of this is hidden, exactly: it's in earnings reports, in grid filings, and in trade publications. But it doesn't look like anything from the outside. A data center is just a building. A query is just a text box. The energy behind it is invisible by nature, and that invisibility is doing a lot of work.

This is where that advice of "just unplug your charger" goes terribly off track. Not because individual habits aren’t relevant — they really are — but because framing energy as primarily a personal behavior problem directs attention away from the places where the numbers actually live. The average American household uses about 10,500 kilowatt-hours per year. A single large AI training run uses 120 times that in a few weeks for one model. Individual behavior and corporate demand are not in the same category of the energy consumption crisis. Treating them as if they are is a form of misdirection — sometimes unintentional, sometimes not.

In New York, ConEdison serves about 3.4 million customers across the city. Total city electricity consumption runs around 50 billion kilowatt-hours annually. Every time a new data center cluster comes online, a technology which hundreds of billions of dollars have already been invested in by companies like Amazon, Microsoft, Google and others, it adds load equivalent to the energy use of a small city. And the grid doesn't know whether that electricity is going to charge a student's phone in Seattle or power a server farm in Virginia. It just has to generate it.

So what does any of this mean for how we think about energy?

It means that supply-side solutions — renewables, nuclear, efficiency upgrades — are necessary but not sufficient. If demand keeps doubling, we're running in place. The Jevons paradox suggests that more efficient energy systems historically produce more energy consumption, not less, unless demand is also shaped. The history of the 20th century is largely a history of demand growing to meet whatever supply was available. There's no physical law that says this has to continue, but there's no economic incentive currently in place to stop it either.

The hard version of the energy conversation isn't about which fuel source we use. It's about what we're using energy for, at what scale, and whether that use is actually producing value proportional to its cost. An AI data center burning gigawatts to generate content for TikTok is a different kind of problem than a hospital burning gigawatts to run diagnostic equipment. Both use electricity, but they're not the same thing. Right now, our systems don't distinguish between them. They just generate, transmit, and bill.

I still turn off the lights when I leave a room. I still keep my showers short. These aren't bad habits. But I've stopped thinking of them as the main event. The grid is a system, and systems respond to incentives, regulations, and infrastructure — not to individual virtue. The reason America wastes hundreds of millions of tons of CO₂ equivalent in transmission losses isn't because people forgot to unplug their chargers. It's because we built an infrastructure that treats energy as infinitely abundant and demand as always justified.

Changing that requires asking a different question — not just "where does energy come from?" but "what is it actually for, and how much of it do we actually need?"

Those are harder questions. They don't fit on a recycling poster. But they're the ones that actually matter. And they are the ones that will define how we see the world in 20 years.