作为一名环保记者，在报道人工智能对生态环境的影响时，我经常遇到阻力。有些读者认为，将数据中心和人工智能模型描述成“妖魔”是不公平的。然而，有确凿的证据表明，为人工智能提供动力是一项资源消耗巨大的事业，尤其是在用水方面，这些环境担忧绝非危言耸听。人工智能对水的巨大需求是一个亟待解决的关键问题。 最惊人的统计数据之一显示，人工智能一年内可能消耗高达7640亿升水。形象地说，这个水量与全球人口同期饮用瓶装水的总量不相上下。这并非遥远的未来预测，而是对当前人工智能对地球淡水资源巨大消耗的估算。这一惊人的数字凸显了人工智能对全球水资源的巨大影响，其用水模式与通过瓶装水满足的基本人类需求不相上下——而瓶装水本身也存在着众所周知的环境问题。 展望未来，水资源需求将急剧增长。预测显示，到2030年，作为人工智能运营支柱的美国数据中心，在高峰期每天将额外需要14.5亿加仑的水。这一巨大的日用水量相当于纽约这样大都市的每日总供水量，表明人工智能将对现有水利基础设施和供水系统造成前所未有的压力。 这种巨大的水资源需求有双重且相互关联的原因。首先，为人工智能模型供电需要大量的电力。全球很大一部分电力生产仍依赖热电厂（例如燃煤、核能、天然气），这些电厂严重依赖水进行冷却。随着人工智能的能源足迹扩大，其间接发电所需的用水量也随之增加。其次，数据中心本身会产生大量热量，因为其强大的图形处理器（GPU）正在处理复杂的计算。为防止过热并保持最佳性能，这些设施采用复杂的冷却系统，其中许多系统利用大量水进行蒸发或直接冷却。随着人工智能模型变得更加复杂、规模更大、应用更广，能源和冷却需求都会按比例升级，从而加剧水资源的紧张状况。 然而，水资源消耗仅仅是人工智能更广泛环境代价的一个方面。作为人工智能计算基础的专用硬件，特别是图形处理器（GPU），在其整个生命周期中都带有着巨大的生态足迹。这首先源于从全球各地开采各种原材料（包括稀土元素和其他关键矿物）的破坏环境的过程，这些开采往往发生在引发伦理和环境质疑的条件下。随后这些高科技组件的制造过程耗能巨大，并产生大量工业废弃物。此外，在人工智能能力不断提高的推动下，计算硬件的快速淘汰周期转化为一个巨大且日益增长的电子垃圾挑战。这些废弃的GPU和其他电子元件可能含有有害物质，如果处理不当，将对土壤和水造成重大风险。从原材料的初始开采到制造过程，再到最终作为电子垃圾处理，GPU对环境造成了严峻的问题。 作为一名报道这些问题的环保记者，我们必须超越那些将人工智能视为“妖魔”的轻蔑言论，而是要关注其切实的风险以及对可持续解决方案的迫切需求。全面了解这些环境成本——从巨大的水足迹到原材料采购、能源消耗和电子垃圾的复杂挑战——是缓解这些问题的关键第一步。公众和政策制定者必须充分认识到我们对人工智能日益增长的依赖所带来的全部影响。

As an environmental journalist, I often face pushback when reporting on the ecological impact of artificial intelligence, with some readers suggesting that data centers and AI models are unfairly cast as "boogeymen." However, the evidence unequivocally demonstrates that powering AI is an incredibly resource-intensive endeavor, particularly concerning water usage, and these environmental concerns are far from exaggerated. The sheer scale of AI's thirst is a critical issue that demands immediate attention. One of the most striking statistics reveals that AI could consume an astonishing 764 billion liters of water globally in a single year. To put this into perspective, this volume is comparable to the entire amount of water people consume globally in bottled water within the same timeframe. This isn't a distant future projection; it's a current estimate of AI's substantial draw on our planet's freshwater resources. This staggering figure underscores AI's significant impact on global water resources, highlighting a consumption pattern that rivals a fundamental human need currently met by bottled water – a commodity with its own well-documented environmental issues. Looking ahead, the demand for water is set to escalate dramatically. Projections indicate that by 2030, U.S. data centers, which form the backbone of AI operations, will require an additional 1.45 billion gallons of water *per day* during peak usage. This colossal daily requirement is equivalent to the entire daily water supply of a major metropolis like New York City, demonstrating the unprecedented pressure AI will place on existing water infrastructure and supplies. The reasons for this immense water demand are twofold and interconnected. Firstly, powering AI models requires vast amounts of electricity. A significant portion of global electricity generation still relies on thermoelectric power plants (e.g., coal, nuclear, natural gas), which heavily depend on water for cooling purposes. As AI's energy footprint expands, so too does the water needed for its indirect electricity generation. Secondly, data centers themselves generate substantial heat as their powerful Graphics Processing Units (GPUs) process complex computations. To prevent overheating and maintain optimal performance, these facilities employ sophisticated cooling systems, many of which utilize significant quantities of water for evaporation or direct cooling. As AI models become more complex, larger, and more widespread, both energy and cooling demands escalate proportionally, intensifying the strain on water resources. However, water consumption is merely one facet of AI's broader environmental toll. The specialized hardware, particularly GPUs, which are fundamental to AI computation, carry a substantial ecological footprint throughout their entire lifecycle. This begins with the environmentally destructive processes of mining various raw materials, including rare earth elements and other critical minerals, from diverse global locations, often under conditions that raise ethical and environmental questions. The subsequent manufacturing of these high-tech components is energy-intensive and generates substantial industrial waste. Moreover, the rapid obsolescence cycle of computing hardware, driven by ever-improving AI capabilities, translates into a massive and growing e-waste challenge. These discarded GPUs and other electronic components can contain hazardous materials, posing significant risks to soil and water if not properly recycled or managed. From the initial extraction of raw materials to the manufacturing processes and ultimately to their disposal as e-waste, GPUs contribute to a significant environmental "mess." As an environmental journalist reporting on these issues, it is imperative to move beyond dismissive comments about AI being a "boogeyman" and instead focus on the tangible risks and the urgent need for sustainable solutions. Understanding the full scope of these environmental costs – from the massive water footprint to the complex challenges of raw material sourcing, energy consumption, and e-waste – is the essential first step toward mitigating them. It's crucial for the public and policymakers to grasp the full implications of our rapidly expanding reliance on AI.