August 12, 2026, (Inside AI) — The race to build ever-larger AI models is pushing data centers into space and under the sea, but two UC Berkeley scientists argue this brute-force approach is a business choice, not a technical necessity. In a commentary published Monday in Nature, they contend that smaller, open-source models are rapidly closing the performance gap while slashing energy and water demands.
Professors Carl Boettiger and Fernando Pérez, leaders at the Eric and Wendy Schmidt Center for Data Science & Environment, along with Cassie Buhler of CU Boulder, challenge the narrative that great intelligence requires great power. Their call to action arrives as data center construction faces mounting community opposition and grid strain.
“Data centers don’t comprise a huge amount of the energy footprint of the planet compared to things like air conditioning or other big electrical uses, but they are the most concentrated energy demands we’ve ever created,” Boettiger said. “And because the energy grid is very localized, they spike the power rates in the communities where they are built. In addition, they are too often located in communities that can least afford it, so these impacts are multiplied.”
The commentary hits a nerve as the industry’s appetite for compute collides with climate goals. A single hyperscale facility can consume as much electricity as a small city, and cooling systems guzzle millions of gallons of water. Yet Boettiger and Pérez see a path forward that doesn’t require abandoning AI altogether.
Open models erase the efficiency gap
Open-source large language models now trail proprietary giants by roughly six months on key benchmarks, a lag that is shrinking as optimization techniques improve. The release of DeepSeek in January 2025 sent a shock through markets when it demonstrated competitive performance after training on a fraction of the energy budget of frontier models. That moment, Boettiger said, exposed the economic incentives driving centralized, power-hungry architectures.
“Computing has always followed a pattern where new algorithms are relatively inefficient, and as they get better and better they run on lower and lower power,” Boettiger said. “Right now, AI development has become an economic race. If you’re a company trying to build market share, you don’t want to be selling a model that’s a year behind. That’s immensely far in AI terms.”
This dynamic rewards speed over efficiency. Companies optimize for low latency to keep users from switching providers, even if a slower, more deliberate model could deliver equal or better results with far less energy. Pérez added that centralization also lets companies hoover up user data to further train their models, locking in competitive advantages.
From server racks to laptops
The trajectory of open models mirrors the miniaturization that turned room-sized mainframes into smartphones. In 2023, Meta’s Llama model leaked and within two weeks was running on a laptop, a feat that surprised even its creators. Since then, the resources needed to run capable open models have plummeted. Boettiger predicts that within six months, models matching today’s frontier will run on common hardware, a shift already signaled by laptop makers shipping NVIDIA RTX Spark chips.
“We’re basically at the transition point where the technologies that are six months behind the frontier can do useful work,” Boettiger said. “And this is coming just in time, too, as data centers are going to get harder and harder to construct—the political will is starting to change and there’s so much opposition.”
Pérez likened the situation to transportation: most people don’t need a Formula One car when a Toyota Corolla suffices. For scientific research, open models offer an additional benefit: reproducibility and control. “I believe it is important that scientists actually have control of their scientific instruments,” Pérez said. “Scientists need tools that they can take apart, reassemble, reinvent and reimagine to suit their expertise and their needs.”
The irony, Pérez noted, is that the AI revolution itself was built on decades of openly available tools, data, and software. “The current AI revolution would not have happened if everybody was clicking around in Microsoft Excel using Windows 95,” he said.
The Berkeley team’s commentary, published in Nature, frames the choice not as AI versus the environment but as a matter of picking the right tool for the job. Boettiger drew an analogy: refusing to use AI because of data centers is like refusing all transportation because of airplanes. The lower-footprint options exist, and they are improving fast. The barrier, he said, is business models, not physics.
