Consider the transistor, the basic unit of computer processors. Transistors can be tiny, down to single-digit nanometers in size. Billions can fit on a computer chip.
Though they have no moving parts, they devour electricity as they store and modify bits of information. “Ones and zeros are encoded as these high and low voltages,” said Timothy Sherwood, a computer science professor at the University of California Santa Barbara. “When you do any computation, what’s happening inside the microprocessor is that there’s some one that transitions to a zero, or a zero that transitions to one. Every time that happens, a little bit of energy is used.”
Sign up here to explore the big, complicated problems the world faces and the most efficient ways to solve them. Sent twice a week.
When you add that up — across the billions of transistors on chips and then the billions of these chips in computers and server farms — they form a significant and growing share of humanity’s energy appetite.
According to the International Energy Agency, computing and storing data accounts for somewhere between 1 and 1.5 percent of global electricity demand at the moment.
With the growth of artificial intelligence and cryptocurrencies that rely on industrial-scale data centers, that share is poised to grow. For instance, a typical Google search uses about 0.3 watt-hours while a ChatGPT query consumes 2.9 watt-hours. In 2024, the amount of data center capacity under construction in the US jumped 70 percent compared to 2023. Some of the tech companies leaning into AI have seen their greenhouse gas emissions surge and are finding it harder to meet their own environmental goals.
How much more electricity will this computation need in the years ahead, and will it put our climate change goals out of reach?
AI is injecting chaos into energy demand forecasts
The IEA estimates that data center energy demand will double by 2030. McKinsey estimates somewhere between a tripling and a quintupling. As a result, major tech players are desperately trying to shore up their power supplies. Over the past year, they’ve been some of the largest purchasers of energy sources that produce few greenhouse gas emissions. Amazon is the largest corporate buyer of renewable energy in the world. Companies like Microsoft are even reviving old nuclear plants while also investing in the next generation of nuclear technology.
But some of these companies aren’t picky about where their power is coming from. “What we need from you,” former Google CEO Eric Schmidt told the House Energy and Commerce committee earlier this month, is “energy in all forms, renewable, non-renewable, whatever. It needs to be there, and it needs to be there quickly.”
Already, energy demand from data centers is extending a lifeline to old coal power plants and is creating a market for new natural gas plants. The IEA estimates that over the next five years, renewables will meet half of the additional electricity demand from data centers, followed by natural gas, coal, and nuclear power.
However, a lot of these energy demand forecasts are projections based on current trends, and well, a lot of things are changing very quickly. “The first thing I’ll say is that there’s just a lot of uncertainty about how data center energy demand will grow,” said Jessika Trancik, a professor at the Massachusetts Institute of Technology studying the tech sector and energy.
Here is some context to keep in mind: Remember that data centers are less than 2 percent of overall electricity demand now and even doubling, tripling, or quintupling would still keep their share in the single digits. A larger portion of global electricity demand growth is poised to come from developing countries industrializing and climbing up the income ladder. Energy use is also linked to the economy; in a recession, for example, power demand tends to fall.
Climate change could play a role as well. One of the biggest drivers of electricity demand last year was simply that it was so hot out, leading more people to switch on air conditioners. So while AI is an important, growing energy user, it’s not the only thing altering the future of energy demand.
We’re also in the Cambrian explosion era of crypto and AI companies, meaning there are a lot of different firms trying out a variety of approaches. All of this experimentation is spiking energy use in the near term, but not all of these approaches are going to make it. As these sectors mature and their players consolidate, that could drive down energy demand too.
The good news is that computers are getting more efficient. AI and crypto harness graphical processing units, chips optimized for the kinds of calculations behind these technologies. GPUs have made massive performance leaps, particularly when it comes to the ability of AI to take in new information and generate conclusions.
“In the past 10 years, our platform has become 100,000 times more energy efficient for the exact same inference workload,” said Joshua Parker, who leads corporate sustainability efforts at Nvidia, one of the largest GPU producers in the world. “In the past two years — one generation of our product — we’ve become 25 times more energy efficient.”
Nvidia has now established a commanding lead in the AI race, making it one of the most valuable companies in history.
However, as computer processors get more efficient, they cost less to run, which can lead people to use them more, offsetting some of the energy savings.
“It’s easier to make the business case to deploy AI, which means that the footprint is growing, so it’s a real paradox,” Parker said. “Ultimately, that kind of exponential growth only continues if you actually reach zero incremental costs. There’s still costs to the energy and there’s still cost to the computation. As much as we’re driving towards efficiency, there will be a balance in the end because it’s not free.”
Another factor to consider is that AI tools can have their own environmental benefits. Using AI to perform simulations can avoid some of the need for expensive, slow, energy-intensive real-world testing when designing aircraft, for example. Grid operators are using AI to optimize electricity distribution to integrate renewables, increase reliability, and reduce waste. AI has already helped design better batteries and better solar cells.
Amid all this uncertainty about the future, there are still paths that could keep AI’s expansion aligned with efforts to limit climate change. Tech companies need to continue pulling on the efficiency lever. These sectors also have big opportunities to reduce carbon emissions in the supply chains for these devices, and in the infrastructure for data centers. Deploying vastly more clean energy is essential.
We’ve already seen a number of countries grow their economies while cutting greenhouse gases. While AI is slowing some of that progress right now, it doesn’t have to worsen climate change over the long term, and it could accelerate efforts to keep it in check. But it won’t happen by chance, and will require deliberate action to get on track.
“It’s easy to write the headline that says AI is going to break the grid, it’s going to lead to more emissions,” Parker said. “I’m personally very optimistic — I think this is credible optimism — that AI over time will be the best tool for sustainability the world has ever seen.”
