Microsoft vowed to fight climate change — then genAI came along

For years, Microsoft has prided itself on the time, effort, and money it puts into fighting climate change. In early 2020, Microsoft President and Vice Chair Brad Smith promised in a blog post the company would invest $1 billion to make itself carbon-negative by 2030. 

Smith vowed Microsoft wouldn’t use deceptive tricks to get there, for example by pouring carbon in the atmosphere, then buying so-called “offsets” that claim to reduce carbon elsewhere, such as planting trees in a rain forest.

Investigations have found that offsets can be a kind of “greenwashing” that allows large companies to surround themselves with the golden halo of fighting climate change, while doing little or nothing to achieve necessary goals. An investigation by The Guardian and others concluded, for example, that “more than 90% of rainforest carbon offsets by [the] biggest certifier are worthless.”

Kudos to Microsoft for recognizing that. And kudos to the company for beginning to take real action to get to its goal rather than using sleight-of-hand tactics.

But then electricity-hungry generative AI (genAI) came along. Suddenly fighting climate change and bolstering sustainability didn’t seem quite so important to the company.

That’s the message delivered by Microsoft’s spike in water use to cool AI data centers, and the company’s recently proposed deal to reopen Three Mile Island, the site of the worst nuclear power disaster in US history. 

All of the reactors at Three Mile Island were shuttered in 2019 because of financial problems. This new deal would restart the complex, with Microsoft paying plant owner Constellation Energy to reopen a nuclear reactor next to the one that melted down and buying all of its electric output for 20 years. That would help deliver a portion of the vast amounts of electricity required to run Microsoft’s AI data centers. Financial terms weren’t released, although Constellation says the plan would cost it $1.6 billion and will require federal tax breaks. 

(There’s nothing new with that last bit; nuclear power has been propped up by billions in tax breaks and other government aid since its inception.)

Microsoft will no doubt tout nuclear power as providing clean, carbon-free energy. But that ignores myriad other problems. Accidents at Chernobyl, Fukushima, and Three Mile Island remind us the plants remain vulnerable to disasters caused by both humans and nature. Beyond that, there is still no safe way to permanently store nuclear waste, according to former chairman of the US Nuclear Regulatory Commission Allison Macfarlane, currently director of the School of Public Policy and Global Affairs at the University of British Columbia.

Then there’s what might be nuclear power’s biggest problem. The same technology used to make fuel for nuclear plants can be used to make nuclear weapons — and nuclear plants are notably insecure. The Union of Concerned Scientists warns: “The NRC [Nuclear Regulatory Commission] has regularly downplayed the threat of nuclear terrorism, relaxing its requirements for security exercises in response to industry pressure to lower costs.”

GenAI and environmental problems

Use of nuclear power isn’t the only environmental issue posed by the rise of genAI tools and platforms like Microsoft’s Copilot. That’s because genAI requires tremendous amounts of computing power compared to traditional technologies. GenAI first needs to be trained. Once it’s trained, it uses complex calculations to handle each incoming request. That, in turn, requires building massive data centers, which use tremendous amounts of electricity. 

Scientific American reports that researchers say OpenAI’s GPT-3, on which Copilot is based, “has 175 billion parameters, consumed 1,287 megawatt hours of electricity and generated 552 tons of carbon dioxide equivalent, the equivalent of 123 gasoline-powered passenger vehicles driven for one year. And that’s just for getting the model ready to launch, before any consumers start using it.”

Keep in mind that GPT-3 has been superseded by GPT-4 and GPT-4o, which are more powerful and require even more computing power — and therefore more electricity — than GPT-3.

Providing all that electricity is just the start of genAI’s — and Copilot’s — environmental impact. Data centers need to be cooled down because of the heat produced by tightly packed chips. Thanks largely to AI, Microsoft’s water use spiked 34% between 2021 and 2022 — and that was before genAI took off.

All this doesn’t take into account the manufacturing and transportation of the high-end chips that do all the data crunching. And it doesn’t account for how much extra e-waste will be created — older generations of chips and hardware will be continually discarded when they’re replaced by increasingly powerful new generations.

What should Microsoft do?

In the past, Microsoft has shown it’s serious about fighting climate change. AI, especially genAI, might change that. But Microsoft can’t solve the issue by itself. If the company reduces its electricity use by cutting back on its genAI plans, rival companies such as Google, Amazon, OpenAI, Meta and others will fill the breach. The same amount of carbon will still be released into the atmosphere. As evidence, Google already has fallen short of its climate change goals: in 2023 its electric use jumped 13% thanks to its use of AI, rather than fall as Google had projected.

The problem of AI accelerating climate change can’t be solved by Big Tech itself. Governments need to step in and establish realistic regulations about technology’s effects on the climate. Doing that isn’t a pipe dream — outside the US, it’s already happening. 

In mid-September, Google announced it was halting plans to build a $200 million data center in Chile after a Chilean court ruled the project violated the country’s environmental regulations. The company vowed it would redesign the project from the ground up to meet those requirements.

Many people have warned that AI could present an existential threat to mankind if intelligent AI systems run amok. It would be ironic if the existential threat turns out not to be AI itself, but instead climate destruction caused by AI’s ravenous electricity demands.