My Attorney Is A Robot

As the adoption of artificial intelligence systems rapidly increases, the demand for resources essential to support and operate these advanced technologies also escalates. This is vividly illustrated by the significant volumes of water required to cool the growing number of servers in newly constructed data centers across the United States. Unlike traditional servers, AI servers incorporate more central processing units (CPUs) and graphics processing units (GPUs), leading to a dramatic rise in power consumption that exceeds the 300-watt capacity of standard air cooling technologies. Remarkably, it is estimated that ChatGPT uses about 500 milliliters of water—comparable to a 16-ounce bottle—each time it processes a set of 5 to 50 prompts or questions. Given that elevated temperatures can impair chip performance, companies such as Microsoft and OpenAI are compelled to adopt liquid cooling solutions. However, the intensive water use required for cooling has ignited debates on the sustainability of AI infrastructure, with critics pointing to data centers in Goodyear, Arizona, and West Des Moines, Iowa, as instances where excessive water consumption exacerbates resource depletion and environmental strain in already water-stressed regions.

Liquid Cooling

Over the last decade, major advancements have been made in how central processing units are built. In turn, this has led to CPUs having more cores and operating at higher speeds. However, this progress has caused the amount of heat these CPUs generate—measured as Thermal Design Power (TDP)—to almost double across several new versions. Furthermore, it’s expected that this trend will continue, making CPUs generate even more heat.

To address this problem, many data centers employ liquid cooling. This is a technique used to pull heat away from server components using a liquid coolant. More specifically, a system pumps the liquid through a loop system inside the server, absorbing heat from the components. After the liquid collects the heat, it moves to a heat exchanger. Here, water that’s been chilled by the facility’s cooling system carries away the heat, effectively removing it from the data center.

Climate Variability in Microsoft Data Centers

An example of such a facility is Microsoft’s data center in Goodyear, Arizona. Spanning 279 acres, the Goodyear campus includes two major buildings, PHX-10 and PHX-11, with plans to expand further. These buildings serve as hubs for cloud computing and AI advancements and represent a critical nexus of technological progress and environmental challenges. At peak operation, the data center’s water usage is projected to reach approximately 56 million gallons annually, equating to the consumption of 670 average local households.

Although 56 million gallons might seem like an immense amount of water, the city of Goodyear has reassured its citizens that Microsoft’s consumption will not affect them. This assurance is based on the fact that Microsoft’s water usage is governed by industrial water rights, ensuring it “does not impact the city’s water portfolio.” Moreover, city officials note that Microsoft employs air cooling for its data center when temperatures are below 85 degrees. However, given that the region frequently experiences temperatures above this threshold, the reliance on air cooling might be relatively limited.

In contrast, in West Des Moines, Iowa, where the climate is less harsh, Microsoft has the advantage of more frequently utilizing ambient air for cooling its data centers. This location also houses the Azure supercomputer, which Microsoft developed for OpenAI and utilizes to support the training of advanced AI models. These models are increasingly vital for a variety of applications, such as analyzing documents, aiding in software development, and even assisting in vacation planning. Hosted on Azure, these systems have played a crucial role in the evolution of OpenAI’s GPT-4 model, which is integrated into several Microsoft technologies, including the new Bing and various AI-powered copilot applications, enhancing productivity and cybersecurity efforts.

However, during periods of peak temperature, the data centers’ dependency on water resources becomes more pronounced, with significant withdrawals from local rivers. For instance, in July 2022, Microsoft’s data center activities constituted about 6% of the total water usage in the local utility district, surging to 11.5 million gallons. This increased usage occurred during a spell of unusually dry weather in West Des Moines, raising pressing questions about the balance of water consumption between industrial facilities and community needs, particularly during times of environmental stress.

Corporate Accountability

To address worries about its impact on local water supplies, Microsoft has pledged to shift from evaporative cooling—a method that conserves energy but consumes a lot of water—to allowing higher temperatures in its data centers. This transition may reduce or eliminate the necessity for water-based cooling in various climates, potentially cutting water use by up to 60%. This move aligns with a broader industry trend towards operating data centers at higher ambient temperatures, bolstered by studies showing that modern servers can operate reliably under such conditions.

Nevertheless, Microsoft’s latest environmental disclosures reveal a 34% surge in global water consumption from 2021 to 2022, escalating to nearly 1.7 billion gallons. This uptick, particularly sharp in the context of the company’s AI initiatives, has ignited scrutiny from external analysts. The spike contrasts with the broader efforts to mitigate resource use, emphasizing the tension between technological innovation and environmental stewardship.

Global Implications and Corporate Commitments

While localized snapshots illuminate the resource challenges communities across the U.S. may face as AI usage increases, the overall impact will be global. It is estimated that AI-driven data centers are poised to consume 1.7 trillion gallons of water worldwide by 2027. As this demand escalates, it is likely that the tech industry’s leaders will face increased pressure to reconcile their rapid growth with environmental sustainability.

Therefore, in response to these challenges, corporations like Microsoft and OpenAI have begun to affirm their commitment to becoming carbon-negative, water positive, and zero-waste by 2030. Ironically, it may be AI’s ability to harness big data more accurately and efficiently that helps these corporations achieve these ambitious environmental milestones.

For example, AI can analyze data to identify patterns and inefficiencies in water use, leading to more sustainable practices. For example, in the building management sector, companies like Schneider Electric use AI to optimize the use of resources, including water, in their facilities. By integrating AI with building management systems, companies can dynamically adjust water usage based on real-time data, significantly reducing waste and improving efficiency.

Moreover, AI can assist in predicting maintenance needs and detecting leaks or inefficiencies in industrial processes, preventing excessive water use, and ensuring systems operate within optimal parameters. Beyond specific industrial applications, AI’s data analysis capabilities can support broader environmental conservation efforts, helping industries comply with regulations and embrace more sustainable practices.

Conclusion

As AI technology rapidly advances, its environmental toll, particularly in water consumption for data center cooling, becomes increasingly conspicuous. The tech industry’s escalation in water use, highlighted by the operational demands of AI-centric data centers, raises critical concerns about sustainable resource management. However, industry leaders like Microsoft and OpenAI are pledging to adopt more sustainable practices, targeting ambitious environmental goals by 2030. This commitment reflects a broader industry trend toward leveraging AI’s analytical prowess not just for business innovation but also for enhancing environmental sustainability.

In essence, while AI’s resource intensity presents a significant environmental challenge, it also offers a pathway to solution, exemplifying a unique convergence of technological advancement and ecological responsibility. The tech sector’s endeavors to harness AI for optimizing resource efficiency and reducing environmental footprints exemplify a crucial balancing act between embracing the benefits of AI and upholding commitments to planetary stewardship. As these companies progress toward their sustainability targets, AI stands as a critical ally, promising a future where technological growth and environmental sustainability coexist harmoniously.