Comment: AI ought to pay a price for its environmental damage

By Mark Gongloff / Bloomberg Opinion

Humanity has collectively decided to keep pampering the fossil-fuel industry despite knowing for decades that its products are not only harmful to long-term well-being but also imminently replaceable. It shouldn’t make the same expensive mistake with artificial intelligence.

Last week, Laurence Tubiana, chief executive officer of the European Climate Foundation, a nonprofit research and advocacy group, suggested taxing AI to raise money for climate adaptation. Tubiana helped craft the Paris climate accord and is part of the Global Solidarity Levies Task Force, a group rummaging through the world’s couch cushions for spare change to help it adjust to an environment growing more chaotic and destructive as the planet gets hotter. The group has identified some obvious targets, such as taxing “premium flyers,” cryptocurrencies, fossil-fuel profits and shipping emissions.

Until Tubiana’s comments, the group hasn’t said much about AI. Sutton’s Law — based on the apocryphal claim by Willie Sutton that he robbed banks because that’s where the money was — suggests it might want to give AI a look. The United Nations Trade and Development Agency has estimated the value of this market will explode from $189 billion to $4.8 trillion by 2033, starting to threaten the size of the fossil-fuel market.

All that growth will require dizzying amounts of energy to run data centers around the clock and a lot of water to cool acres of servers. These power-hungry facilities, which are popping up around the world like acne on a teenager, could consume as much as 12 percent of total U.S. electricity by 2028, according to a report last year by the Lawrence Berkeley National Laboratory, up from 4.4 percent in 2023. By 2050, data centers could use as much as 8.7 percent of the entire world’s energy, BloombergNEF estimates.

Much of that power will be generated by fossil fuels, the primary source of the greenhouse gases cooking the atmosphere. Data-center operations could increase global emissions of these gases by 3.5 billion tons in the next decade, BNEF reckons, which is about 10 percent of total global emissions today. The microchips, steel and cement used to build data centers carry their own heavy climate cost. Meanwhile, U.S. data centers could guzzle 74 billion gallons of water a year by 2028, up from less than 6 billion in 2014, according to Lawrence Berkeley.

In light of all this, taxing AI to defray its harm to the climate sounds like a great idea. We’ve already got the perfect example of what happens when you fail to hold an industry accountable for its externalities: fossil fuels.

By one estimate, we give oil, gas and coal producers about $6 trillion every year in implied subsidies by failing to price their products high enough to account for their environmental damage. That’s a lot of money that could go toward steeling ourselves against rising seas and weather disasters.

The rationale of these subsidies is that the societal benefits of fossil fuels — namely, abundant energy to fuel economic growth — outweigh the costs. But this gets more untrue every year as an increasingly chaotic climate wreaks more havoc. Each degree Celsius the planet warms over preindustrial averages cuts global economic growth by 12 percent, according to a National Bureau of Economic Research study last year.

Of course, this wasn’t so apparent in the 19th century, when today’s fossil fuel industry was just a gleam in John D. Rockefeller’s eye. Magical stuff came out of the ground that kept the lights on, and it took us a little while to see the downside. In contrast, the societal benefits of AI are fuzzy at best while the many downsides are already plain. It makes no sense to give this industry its own free pass on environmental destruction.

In theory, a tax on AI, or at least its carbon emissions, would push tech companies to seek efficiency and cleaner energy sources. In practice, making such a tax work would be challenging, Robert Bikel, director of the Socially‚ Environmentally‚ and Ethically Responsible Program at the Pepperdine Graziadio Business School, told me.

If you make the tax too low, then it becomes just another cost of business and loses effectiveness. If you make it too high, then tech companies will just build their data centers in a jurisdiction that’s not so picky about the environment. The net result might be more emissions than without the tax.

Making such a levy universal would help solve that problem. But good luck getting two of the world’s biggest AI enthusiasts — President Donald Trump’s U.S. and China — to play along. For that matter, a global tax on carbon would be the most efficient device of all, taking care of those fossil-fuel subsidies and keeping the AI industry in line in one fell swoop. But that’s even deeper in political fantasyland.

Still, there are other tools. NIMBYism is unhelpful in many ways, but it has put some brakes on runaway data-center growth. In the past two years, projects worth $64 billion have been canceled or delayed by public opposition, according to the research group Data Center Watch. The blowback is bipartisan; 55 percent of the public officials opposing new data centers were Republicans.

Pressure from locals and investors could be enough to make AI purveyors embrace tech that uses less energy and water. They could use their excess heat to keep local homes warm, cutting energy bills and fossil fuel demand. They could invest in renewables, grid upgrades and carbon removal and capture.

Most important, though, they need to stop and think about exactly how much of this rush to make AI as big and resource-consuming as physically possible is truly necessary.

“There’s a lot of techno-optimism that more tech and more growth is inherently good,” Bikel said. “I like to flip that around and say what is the economy serving? What’s AI serving? How is it contributing to human flourishing or even just existential stability?”

Answering those questions would go a long way to making AI’s growth more sustainable and avoiding many more expensive mistakes.