When Chinese entrepreneur Liang Wenfeng unveiled DeepSeek’s latest Artificial Intelligence (AI) app earlier in January, the tech world was quick to marvel at its potential. Here was an AI model that promised to rival the likes of OpenAI’s ChatGPT and Google’s Gemini, but at a fraction of the cost and resource consumption. To a world increasingly concerned about the environmental toll of AI, DeepSeek’s emergence seemed like a beacon of hope. However, is it truly a breakthrough, or just another step in a race that risks outpacing our planet’s capacity to sustain it?
The environmental cost of AI is becoming impossible to ignore. Data centers, the backbone of AI infrastructure, already consume 2% of all global electricity usage, a figure projected to reach 3.4% by 2026. Training a single large language model (LLM) like GPT-3 can consume millions of litres of water and emit as much carbon as five cars over their lifetimes. This voracious appetite for energy and water has led some tech giants, like Google and Meta, to make decisions that starkly contradict their public sustainability commitments. In Omaha, Nebraska, for instance, the Omaha Public Power District had to abandon plans to decommission two coal-burning generators due to rising energy demands from nearby data centers operated by these companies. This decision not only perpetuates greenhouse gas emissions but also endangers local air quality, undermining public health.
Against this backdrop, DeepSeek’s promise of a cleaner, more efficient AI model feels like a breath of fresh air. But is it enough?
David Rolnick, an assistant professor at McGill University’s School of Computer Science, cautions against jumping to conclusions.
“There is almost no information available about either ChatGPT or DeepSeek, so any numbers [about their environmental impact] are speculation,” he notes. While DeepSeek’s efficiency claims are enticing, the real question is how this model will be used.
DeepSeek’s appeal lies in its open-source, “open-weight” model, which allows it to be downloaded and run locally on devices, potentially bypassing the need for massive data centers. This could reduce the energy and water consumption typically associated with AI operations. Moreover, DeepSeek’s development process, which sidestepped the need for Nvidia’s energy-intensive AI chips, could represent a significant step towards reducing AI’s environmental footprint.
Here’s the twist: efficiency gains often lead to increased consumption, a phenomenon known as the Jevons paradox. As AI becomes cheaper and more accessible, its use could skyrocket, offsetting the environmental benefits of improved efficiency. DeepSeek’s innovations might not reduce AI’s overall impact but instead enable its proliferation, leading to even greater energy and resource consumption in the long run. This raises a critical question: are we solving or simply delaying the problem?
Precisely, the environmental impact of AI extends far beyond energy consumption. For instance, the manufacturing of AI chips requires thousands of gallons of water per chip, and data centers rely on vast quantities of water for cooling. A 2023 study from the University of California, Riverside, found that training a large-language model like GPT-3 can consume millions of litres of water. By 2027, AI could be withdrawing 6.6 billion cubic meters of water annually. To compare, this represents six times Denmark’s total annual water usage. In drought-prone regions, this poses a serious threat to already strained water supplies.
While DeepSeek’s efficiency might mitigate some of these impacts, it does not address the root problem: the unchecked growth of AI. As Rolnick points out,
“the biggest differences in terms of the energy consumption aspects of large AI algorithms are how much you use.”
In other words, the environmental benefits of one model over another pale in comparison to the decision of whether to use AI at all.
So, can AI become more sustainable? The answer is yes and no. Technological advancements made by DeepSeek are essential but must be paired with systemic change in how we develop and deploy AI. Data centers must shift away from fossil fuels and embrace renewable energy sources like solar and wind. Pairing data centers with solar parks, as seen in Texas, is a promising start. However, the intermittent nature of renewables necessitates investment in large-scale battery storage, which remains a significant challenge. Additionally, AI companies must prioritize water conservation, particularly in drought-prone regions. This includes adopting closed-loop cooling systems, reusing water, and scheduling energy-intensive tasks, such as data processing and industrial operations during periods of low public water use. Thus, without careful regulation and a commitment to environmental stewardship, AI’s growth could exacerbate the very crises it seeks to solve.
In the end, the question is not whether AI can become more sustainable, but whether we have the collective will to make it so. DeepSeek’s innovations offer a glimpse of what progress is possible, while also reminding us that technology alone cannot save us. It’s up to us to ensure that the future of AI is not just smarter, but also greener. The genie is out of the bottle, now we must decide how to wield its power.
Edited by Alexandra MacNaughton
Victoria Forte is in her third and final year at McGill University, where she is completing her B.A. in Political Science and Gender, Sexuality, Feminist, and Social Justice Studies. As she navigates her final year, Victoria brings a fresh perspective as a new staff writer for Catalyst. Her academic and journalistic pursuits are driven by a deep interest in exploring the intersections of power, identity, and social justice.