For climate In this regard, the rise in the number of Internet queries based on artificial intelligence is a cause for concern. Many people turned to ChatGPT and other services for simple questions. And even a basic Google search includes AI-powered results.
Depending on how you crunch the numbers, you can get a wide range of energy consumption and associated climate costs per AI query. ChatGPT provides evaluate up to 0.34 watt-hours per tip, which is equivalent to using a household light bulb for 20 seconds, while one group of researchers came to the conclusion that some models can use 100 times more for longer tips.
On Thursday, Google released its own data: The average search using Gemini, the company's ubiquitous artificial intelligence tool, uses 0.24 watt-hours. This is equivalent to watching TV in about nine seconds. This search produces 0.03 grams of carbon dioxide equivalent. Perhaps more interesting is that Google says Gemini's text queries have gotten cleaner over time. Over the past year, energy consumption per request has dropped by approximately 97% and carbon emissions have dropped by 98% per request, according to the company. Separate report A report from Google published earlier in the summer showed a breakdown of data center energy consumption and associated emissions. (Of course, it's worth noting that simple text queries are less intensive than other functions like generating images, audio, or visuals, and these numbers don't include model training—numbers that aren't included in Google's report due to problems with accurately calculating them.)
Read more: Some AI tips can cause 50 times more CO2 emissions than others
Whether this downward trajectory can continue is a critical question for anyone watching the energy and climate future in the United States. This will have implications not only for the future of US emissions, but also for hundreds of billions of dollars of investment in the energy sector. Across a variety of related industries, leaders will have to try to get the ball rolling: meeting growing demand for AI while avoiding overbuilding infrastructure as AI models become more efficient.
Google's progress comes down to two levers: cleaner power and more efficient chips, and query processing.
The clean energy strategy is impressive, but quite simple. The company buys a lot of renewable energy to power its operations, signing contracts to purchase 8 GW of clean energy last year alone. This is equivalent to the power of 2,400 utility-scale wind turbines, according to the Department of Energy. The company has further invested in helping to bring into operation other future clean technologies, such as nuclear fusion.
But there are also measures of company efficiency. In energy circles, efficiency usually means simply using less energy and making energy equipment more efficient—think climate control or better insulation. While Google has done some of this, the most impressive efficiency gains have come from the AI ​​ecosystem, not the energy system. The company has created its own chips, which it calls TPUs, as opposed to the commonly used GPUs. These chips have become more efficient over time—about 30 times more efficient since 2018, according to Google's sustainability data. report. The company has also improved the efficiency of its models by using techniques that process queries differently, thereby reducing the processing power required. And a few weeks ago, the company announced a program aimed at shifting demand for data centers to times when the power grid is less stressed.
Read more: AI could change everything we know about climate change
The question—not just for Google, but for any company deeply invested in artificial intelligence—is whether these programs and the associated efficiency gains can continue. Deeper efficiency gains would be a huge win in the fight against climate change—as long as usage growth doesn't outpace efficiency growth.
Improved efficiency will also have significant implications for the entire energy sector. Right now, energy companies are betting big on new sources of power generation, anticipating that artificial intelligence will continue to drive demand growth. But it is very difficult to predict exactly how quickly demand will grow. Promising efficiency gains are a major reason for this, and Google's results should at least make you pause and consider the known unknown potential..
To get this story in your inbox, sign up for the TIME CO2 Leadership Report newsletter. Here.
