数据中心无序扩张引发的社区反弹
过去一年,美国各地正在发生一场不同寻常的事件。从佐治亚州到马里兰州再到密歇根州,社区居民们意外发现,建在他们家附近的数据中心正在以惊人的规模消耗水、电和资金,而这些消耗既未经批准,也未受到监控,在某些情况下甚至无人知晓。在短短一年内,超过140个地方团体阻止或推迟了总价值超过600亿美元的数据中心项目。本立法周期内,至少有12个美国州提交了暂停数据中心建设的法案。城镇居民正在罢免市议会成员,并通过《信息自由法案》(FOIA)获取信息,民众的愤怒情绪日益高涨。本文将深入探讨这些现象背后的驱动因素。这并非要评判人工智能的善恶,而是一个更紧迫和具体的问题:数据中心正以史无前例的速度和规模快速建设。无论人们认为人工智能是杰出、可怕,还是介于两者之间,其背后支撑的物理基础设施正在真实社区落地,消耗真实资源,并产生真实后果。本文将通过三个涉及水、噪音和金钱的案例,探讨数据中心带来的影响,并援引一些令人不安的历史相似案例,同时审视其他国家如何应对同样的挑战,因为事实证明,情况并非必须如此。
Original English Source
Over the past year, something extraordinary has been unfolding across the United States. Communities from Georgia to Maryland to Michigan have been discovering, often by accident, that data centers built near their homes are extracting water, electricity, and money at scales nobody approved, nobody metered, and in some cases nobody even knew about. More than 140 local groups have blocked or delayed over $60 billion in data center projects in just one year. At least 12 US states have filed moratorium bills this legislative cycle. Towns are ousting their city councils. Residents are filing FOAS. People are angry and I understand why. I'm Mel. I have a PhD in computer science and I analyze AI developments to understand what is actually happening beneath all of this hype. This is not about whether AI is a force for good or a force for evil; that is a complex conversation for another time. Today is about something more immediate and concrete: data centers are being built right now at unprecedented speed and scale. Whether you think AI is brilliant, terrifying, or somewhere in between, the physical infrastructure supporting it is landing in real communities, consuming real resources, and generating real consequences. In this video, I'm going to walk through three ongoing cases involving water, sound, and money. I'll draw some historical parallels and look at how other countries are handling the same challenge, because it turns out it doesn't have to be this way.
数据中心:从基础设施到“工业巨兽”的演变
首先,我们需要快速了解数据中心的本质。最简单地说,数据中心就是一座装满了计算机的建筑。它们已经存在了几十年。每次你查询银行余额、在Netflix上观看节目、将照片备份到云端或发送电子邮件时,你都在使用数据中心。它们并非新生事物,也并非天生邪恶。它们是基础设施,就像发电站或水处理厂一样。然而,如今改变的是其规模。人工智能工作负载,例如训练大型语言模型和进行大规模推理,是极度计算密集型的。一块现代人工智能 GPU 每年消耗高达3.7兆瓦时的电力。作为对比,普通美国家庭每年大约消耗10.5兆瓦时。这意味着,一块手掌大小的芯片,其耗电量约为整个家庭年耗电量的三分之一。数据中心将数千个这样的芯片堆叠在服务器机架中,产生巨大的热量。这些热量需要散热,而冷却系统消耗的数据中心总电量近40%。其结果是,这些设施以工业规模消耗水、电和物理空间。然而,它们却常常建在为住宅生活设计的区域。问题也由此而生,并非因为建筑物内部运行着什么,而是因为当建筑物如此庞大且扩张如此迅速时,却没有人制定相应的规则。
Original English Source
Let's start with what data centers actually are. A data center is, at its simplest, a building full of computers. They have existed for decades. Every time you check your bank balance, stream a show on Netflix, back up your photos to the cloud, or send an email, you are using a data center. They're nothing new; they're not inherently sinister. They are infrastructure the same way a power station or a water treatment plant is infrastructure. What is new now is the scale. AI workloads, training large language models, and running inference at scale are extraordinarily computationally intensive. A single modern AI GPU consumes up to 3.7 megawatt hours of power annually. To put that into perspective, the average American household uses about 10.5 megawatt hours per year. So, one chip, roughly the size of your hand, consumes about a third of what an entire home uses in a year. Data centers stack thousands of these into server racks that generate enormous amounts of heat. That heat needs cooling, and cooling accounts for nearly 40% of a data center's total power consumption. The result is facilities that consume water, electricity, and physical space at industrial scales. Except they're often being built in places designed for residential life. And that's where the problems start, not because of what's running inside the building, but because nobody built the rules for what happens when the building gets this big this quickly.
案例一:佐治亚州费耶特县的水资源盗用
让我们来谈谈第一个案例:水。在亚特兰大南部的佐治亚州费耶特县,正在发生一件荒谬的事情。一个高档住宅区的居民开始注意到他们的水压异常低。县政府调查后发现了原因:由全球最大私募股权公司之一的黑石集团(Blackstone)旗下子公司品质科技服务公司(Quality Technology Services, QTS)运营的一个占地615英亩的数据中心园区,通过两条县政府甚至不知道存在的连接管道,抽取了大约2900万加仑的水。其中一条连接管道是在未告知公用事业部门的情况下安装的;另一条则未与计费账户关联。一家管理着超过1万亿美元资产的子公司的确在未告知县政府的情况下安装了一条工业用水连接,竟然无人察觉。2900万加仑的水,相当于44个奥林匹克标准游泳池的容量,远远超过了QTS在获得规划批准时承诺的最高用水量。县政府最终向QTS追溯收取了14.7万美元的费用,但没有罚款,没有惩罚,除了补开发票之外没有任何后果。水务系统主管瓦妮莎·泰格特(Vanessa Tigert)对此的解释非常特别,她说:“他们是我们最大的客户,我们必须成为合作伙伴。这叫客户服务。”这番言论是在干旱期间对未经授权的工业用水抽取行为提供“客户服务”,因为没有人希望这件事被发现。在QTS未经计量抽取水资源的同时,县政府却在告诉居民们节约用水,停止浇灌草坪。县政府实际上在要求个体业主减少用水量,而其最大的单一消费者却在“账外”运营。之所以这一切能公之于众,是因为当地居民詹姆斯·克利夫顿(James Clifton),一位正在竞选县委员的律师,通过公共记录申请获得了公用事业部门发给QTS的信函,并将其发布到Facebook上。县政府原本并不打算主动公开这些信息。正是**《信息自由法案》**(FOIA)和社交媒体才揭露了正在发生的事情以及导致这一切的基础设施漏洞。泰格特本人解释说:“费耶特县是一个郊区,主要是住宅区,我们的系统里没有多少商业用水表。”她的部门只有一名员工负责检查和规划审查——一个人要管理一个面积相当于51个沃尔玛超级购物中心大小的设施。佐治亚州目前正经历中度至重度干旱,州长布莱恩·坎普(Brian Kemp)上个月因野火宣布进入紧急状态。该州拥有200多个数据中心设施。费耶特市议会此后已禁止在所有分区内新建数据中心,但QTS的园区(代号“王者之剑计划”,Project Excalibur),却将继续建设3到5年。QTS和黑石集团还因弗吉尼亚州一个数字门户项目被其合同持有者起诉,该项目位于马纳萨斯国家战场公园附近,是为200万人供水的饮用水流域的源头。QTS表示,这2900万加仑的水用于施工、混凝土、抑尘和场地准备,一旦投入运营,其冷却系统将采用闭环,不会抽取市政用水。这或许是真的,但他们确实超出了约定限额,并安装了未经授权的连接。施工用水并不能完全解释为何需要一个县政府毫不知情的连接。
Original English Source
Let's talk about the first case: water. In Fayette County, Georgia, just south of Atlanta, something absurd is happening. Residents of an upscale subdivision started noticing their water pressure was unusually low. The county investigated and found the cause: a 615-acre data center campus operated by Quality Technology Services (QTS), a subsidiary of Blackstone, one of the world's largest private equity firms, had been drawing roughly 29 million gallons of water through two connections the county didn't even know existed. One hookup had been installed without the utility's knowledge; the other wasn't linked to a billing account. Let me say that again: a subsidiary of a company managing over $1 trillion in assets installed an industrial water connection without telling the county, and nobody noticed. 29 million gallons, for context, is equivalent to 44 Olympic-size swimming pools and far exceeded the peak water usage QTS had agreed to when they got planning approval. The county eventually billed QTS $147,000 US in retroactive charges. No fine, no penalty, no consequences beyond a backdated invoice. The water system director, Vanessa Tigert, explained the decision with an extraordinary statement: 'They're our largest customer, and we have to be partners. It's called customer service.' Customer service for unauthorized industrial water extraction during a drought. Nobody was meant to find out. While QTS was draining the supply unmetered, the county was telling residents to stop watering their lawns to conserve water. The county literally leaned on individual homeowners to cut back their consumption while its single largest consumer was operating off the books. The only reason any of this became public is that a local resident, James Clifton, an attorney now running for county commissioner, obtained the utility's letter to QTS through a public records request and posted it on Facebook. The county was not going to volunteer this information. It took FOIA and social media to surface what was happening and the infrastructure gap that allowed it. Tigert herself explained it: 'Fayette County is a suburb. It's mostly residential, and we don't have much commercial meters in our system.' Her department has a single employee handling both inspection and plan reviews—one person for a facility the size of 51 Walmart Super Centers. Georgia is currently experiencing moderate to severe drought. Governor Brian Kemp declared a state of emergency last month over wildfires. The state hosts more than 200 data center facilities. The Fayette City Council has since banned new data centers in every zoning district, but QTS's campus, code-named Project Excalibur, will continue building for another 3 to 5 years. QTS and Blackstone are also being sued by their own contract holders at a digital gateway project in Virginia, which sits near Manassa's National Battlefield Park and the headwaters of a drinking supply watershed serving up to 2 million people. QTS says the 29 million gallons were for construction, concrete, dust suppression, and site preparation, and that once operational, their cooling will be closed-loop and won't draw municipal water. Maybe, but they exceeded their agreed limits and installed a connection nobody authorized. Construction water doesn't exactly explain why you'd need a hookup the county doesn't know about.
案例二:次声波与噪音污染的隐形威胁
我有一个问题:当一种噪音真实存在,但技术上却“不存在”时,会发生什么?数据中心产生的噪音,其特性与现有法规格格不入。《环境与能源研究学会》(Environmental and Energy Study Institute)报告指出,这些设施发出的高低频噪音可达96分贝,大致相当于割草机在你耳边全天候24小时、每周7天不间断地轰鸣。噪音来源多样且持续不断:大规模工业冷却风扇、可达105分贝(比雪地摩托车还响)的备用柴油发电机,以及日益增多的现场天然气涡轮机。这些涡轮机本质上是固定在地板上的喷气发动机,为了给那些认为电网不够可靠的设施提供持续电力。据报道,埃隆·马斯克(Elon Musk)在孟菲斯的“巨型超级计算机”(Colossal Supercomput)运行了30多台移动燃气涡轮机。想象一下一个机场,但航班从不停止,而你却从未选择住在跑道附近。更令人不安的是,数据中心附近的居民越来越多地报告一种被称为次声波(infrasound)的现象——这种频率极低的声音人耳无法听到,但身体却能真切感受到。人们会因此感到头痛、失眠、恶心、焦虑和眩晕。这就是监管难题所在:用于执行噪音条例的标准分贝计无法捕捉次声波。这些测量工具本身并非为这个频率范围设计。因此,当居民投诉,县政府派人检查时,读数却显示在可接受的范围内。声音确实存在,身体能感知到,但仪表却测不到。这种模式值得关注:那些真实存在但结构上对我们测量工具“隐形”的危害。水通过无人计量连接管道流失。噪音身体感知得到,但专门检测噪音的设备却测不到。稍后我们将在电费问题上再次看到这种模式。与备用发电机偶尔在停电时运行不同,许多这些噪音源(冷却系统、现场涡轮机)是持续不间断运行的。没有安静期,没有喘息的机会。一旦设施投入运营,居民们所熟悉的社区便不复存在。
Original English Source
Here's a question for you: what happens when a noise is real but technically doesn't exist? Data centers generate sound in ways that don't play nicely with existing regulations. The Environmental and Energy Study Institute reports that the high and low-frequency sounds emitted by these facilities can reach 96 dB, roughly as loud as a lawnmower at your ear, 24 hours a day, 7 days a week. The sources are varied and relentless: industrial cooling fans deployed at enormous scale, backup diesel generators that can hit 105 dB (louder than a snowmobile), and increasingly, on-site natural gas turbines. These are essentially jet engines bolted to the floor, running continuously to generate power for facilities that have decided the electrical grid isn't reliable enough. Elon Musk's Colossal Supercomput in Memphis reportedly ran over 30 mobile gas turbines. Picture an airport, but the flights never stop, and nobody chose to live near the runway. But it gets more unsettling. Residents near data centers are increasingly reporting a phenomenon called infrasound—frequencies so low they're inaudible to the human ear but can be physically felt. People describe headaches, insomnia, nausea, anxiety, and vertigo. And here's the regulatory problem: standard decibel meters, the kind used to enforce noise ordinances, don't capture infrasound. The measurement tools themselves are not designed for this frequency range. So when residents complain and the county sends somebody to check, the reading comes back within an acceptable limit. The sound is there; bodies register it; the meter doesn't. This is a pattern worth paying attention to: harms that are real but structurally invisible to the tools we use to measure them. Water that flows through connections nobody metered. Sound that registers in your body but not on the equipment designed to detect it. We'll see it again in a moment with electricity costs. Unlike backup generators, which run occasionally during power outages, many of these noise sources (the cooling systems, the on-site turbines) operate continuously all the time. There is no quiet period; there is no respite. The neighborhood, as these people knew it, simply ceases to exist once the facility goes online.
案例三:电费飙升与跨州成本转嫁
这就引出了第三个案例:金钱。如果说佐治亚州的水资源案例让人觉得触手可及,次声波案例则让人觉得几乎能听到却又无法完全捕捉,那么马里兰州的案例则是以静默的方式出现在你的电费账单上,并永远留存。马里兰州人民委员会办公室(Maryland Office of People's Council, 简称OPC),一个代表公用事业消费者的州机构,向联邦能源管理委员会(FERC)投诉了美国最大的电网运营商 PJM互联公司(PJM Interconnection)。PJM斥资220亿美元升级电网以适应数据中心激增的需求,并向马里兰州收取了其中20亿美元的费用。然而,这个惊人的症结在于,驱动这些需求的大多数数据中心甚至不在马里兰州。该州却要为主要服务其他州设施的基础设施买单。根据OPC的备案文件,这将在未来十年内额外花费马里兰州消费者16.6亿美元。大约每户居民客户345美元,每户商业客户673美元,每户工业客户超过1.5万美元。这些并非一次性费用,而是固化在费率结构中,成为每张电费账单的永久性组成部分。为了理解这种飙升的速度,PJM的容量拍卖成本在一年内从大约20亿美元飙升至近150亿美元,几乎完全由现有和预计的数据中心需求驱动。最近的拍卖额更是超过了160亿美元。独立市场监测机构的分析证实,这一涨幅几乎完全源于数据中心的负荷。因此,我们看到这样的景象:与数据中心行业毫无关联、未从中获得任何利益、也从未投票支持的居民,发现自己的电费账单上多了一笔额外费用,用于资助其他州运营并服务企业客户的设施的电网升级。这并非政策失误,而是一种成本结构设计,旨在让消耗资源的人不必支付这些资源所需的基础设施成本。
Original English Source
Which brings us to case number three: money. If the Georgia water case feels like something you can touch, and the infrasound case feels like something you can almost, but not quite, hear, the Maryland case is the version that arrives silently in your electricity bill and stays there forever. The Maryland Office of People's Council, the state agency that represents utility consumers, filed a complaint before the Federal Energy Regulatory Commission (FERC) against PJM Interconnection, the largest grid operator in the United States. PJM spent $22 billion upgrading its grid to accommodate surging demand from data centers and is billing Maryland $2 billion of that total. The catch, and it's a spectacular catch, is that the majority of data centers driving the demand are not even in Maryland. The state is being charged for infrastructure that primarily serves facilities in other states. According to the OPC's filings, this will cost Maryland consumers an extra $1.66 billion over the next decade. Approximately $345 per residential customer, $673 per commercial customer, and over $15,000 per industrial customer. These are not one-off charges; they are baked into the rate structure. They become permanent features of every single electricity bill. To appreciate how quickly this escalated, PJM's capacity auction costs jumped from roughly $2 billion to nearly $15 billion in a single year, driven almost entirely by existing and projected data center demand. The most recent auction hit over $16 billion. An analysis from the independent market monitor confirmed the hike stemmed almost entirely from data center loads. So here's the picture: residents who have no connection to the data center industry, who derived no benefit from it whatsoever, who never voted for it, find an additional charge on their electricity bill, funding grid upgrades for facilities that operate in other states and serve corporate customers. This is not policy failure. That is a cost structure designed so that the people consuming the resources don't have to pay for the infrastructure those resources require.
历史的教训:“隐形危害”模式的重演
如果这一切让你感到熟悉,那是因为它本就该如此。其模式是:工业规模的资源开采,对当时测量工具而言“隐形”的危害,在缺乏有效监管能力的体系内运营的企业实体,成本转嫁给公众而利润归私人所有,以及只有当公民决定调查时才被发现的真相。这并非新鲜事,它有着漫长而详尽的历史。在1990年代,没有受过正规培训的法律文员艾琳·布罗克维奇(Erin Brockovich)揭露了太平洋煤气电力公司(Pacific Gas and Electric, 简称PG&E)在加利福尼亚欣克利(Hinckley)的地下水体中用六价铬污染了超过26年。PG&E知情,但监管机构却不知。居民们在不明白原因的情况下生病。随后,PG&E支付了3.33亿美元的和解金,但这仅仅是因为一个人拒绝接受官方说法。如果把六价铬换成2900万加仑未经计量抽取的水,把欣克利换成费耶特县,其结构是完全相同的。杜邦公司(DuPont)在西弗吉尼亚州的PFAS污染也遵循同样的逻辑。该公司数十年来明知故犯地向水体排放合成化学品。他们自己的内部研究显示了健康风险,但周边社区却一无所知,而监管机构也未被设计来检测这些排放物。这种危害一直“隐形”,直到它被揭露的那一刻。还有阿片类药物危机,普渡制药公司(Purdue Pharma)通过医疗系统分销 OxyContin,其中每份处方看起来都完全合理、合法。没有一个医生,没有一家药房能看到整体模式。系统性危害只有在大规模显现时才被看见。而当有人最终描绘出全貌时,损害已是灾难性的。这与数据中心资源开采的情况惊人地相似。每个独立的设施都获得了许可,每个独立的连接看起来都可控。但直到水压下降或电费账单到来,才没有人追踪整个区域的总需求。这些并非完美的类比,我深知这一点。没有人会因为未经计量抽取的水管而死亡。但其运作机制——通过缺乏发现整体危害能力的系统进行资源开采,并默默地将成本转嫁给公众——却是相同的套路。认识到这种套路至关重要,因为它能指明所需的解决方案。
Original English Source
If any of this feels familiar, it should. The pattern is industrial-scale resource extraction, harms invisible to the measurement tools of the era, corporate entities operating inside regulatory systems that lack the capacity to see what's happening, cost distributed to the public while profits accrue privately, and discovery that only happens when a citizen decides to investigate. This is not a new thing; it has a long and well-documented history. In the 1990s, Erin Brockovich, a legal clerk with no formal training, uncovered that Pacific Gas and Electric (PG&E) had been contaminating the groundwater in Hinckley, California, with hexavalent chromium for over 26 years. PG&E knew; the regulators didn't. Residents were getting sick without understanding why. A $333 million settlement followed, but only because one person refused to accept the official story. Just replace hexavalent chromium with 29 million gallons of unmetered water and Hinckley with Fayette County, and the structure is identical. DuPont's PFAS contamination in West Virginia followed the same logic. The company knowingly discharged synthetic chemicals into water supplies for decades. Their own internal research showed the health risks. The surrounding community had no idea, and the regulatory apparatus was not designed to detect what was being released. The harm was invisible right up until the moment that it wasn't. And then there's the opioid crisis. Purdue Pharma distributed OxyContin through a healthcare system where individual prescriptions looked perfectly reasonable, totally legitimate. No single doctor, no single pharmacy could see the aggregate pattern. The systemic harm only became visible at scale. By the time anyone mapped the full picture, the damage was catastrophic. That is uncomfortably close to what's happening with data centers' resource extraction. Each individual facility clears its permits. Each individual connection looks manageable. And nobody is tracking the total demand across a region until the water pressure drops or the electricity bill arrives. These are not perfect analogies; I'm well aware of that. Nobody's dying from unmetered water hookups. But the mechanism – extraction through systems that lack the capacity to see the aggregate harm, with cost silently transferred to the public – is the same playbook. Recognizing the playbook matters because it tells you what kind of solution is needed.
国际经验借鉴:爱尔兰的监管框架
现在,我想把话题转向解决方案。此时此刻,人们很容易举手投降说:“那干脆别建了。”但数据中心不会消失,它们支撑着每个人都依赖的关键基础设施。问题不在于它们是否应该存在,而在于是否存在一个框架来确保它们不会碾压所落地的社区。事实证明,其他国家已经开始构建这样的框架。爱尔兰就是一个非常有趣的案例。数据中心消耗了爱尔兰全国计量电力的22%,高于2015年的5%,预计到2026年将上升到约三分之一。特别是在都柏林和米斯(Meath)地区,数据中心占电力消耗的约50%。因此,其挑战规模与美国正在发生的任何情况相比,有过之而无不及,甚至可能更糟。都柏林从2021年到2025年末,对新的数据中心连接实际上实行了长达5年的暂停令(moratorium)。但其不同之处在于:当爱尔兰解除暂停令时,它并没有完全放开。公用事业监管委员会(Commission for Regulation of Utilities, 简称CRU)引入了一项具有约束力的义务,即数据中心每年80%的电力必须来自爱尔兰境内新增的可再生能源,并设定了六年的过渡期。新的数据中心必须将电力反馈到国家电网。如果设施不符合性能和可用性标准,系统运营商可以拒绝拟议的选址并降低其进口容量。如果数据中心未能提供所需的能源服务,电网运营商可以降低其进口容量或将其转为非固定状态。这意味着存在实际的后果。在欧盟层面,监管架构甚至走得更远。修订后的能源效率指令(Energy Efficiency Directive)强制要求500千瓦时以上的数据中心进行标准化可持续性报告,并每年向欧洲数据库提交。欧盟分类法规(EU Taxonomy Regulation)根据气候缓解标准对数据中心活动进行分类。《人工智能法案》(AI Act)本身也包含通用人工智能模型的能源消耗透明度要求。欧盟委员会正在提出一项数据中心能源效率一揽子计划,旨在到2030年实现数据中心的碳中和。它完美吗?当然不。爱尔兰的环境组织正在高等法院挑战CRU的决定,他们认为20%的非可再生能源配额加上化石燃料备用发电机仍将使该国排放量增加。爱尔兰自己的环境保护局预测,该国将在2026年至2030年碳预算期内超支7700万至1.14亿吨。这些都是合法的担忧,正在通过民主和法律程序积极辩论。但爱尔兰拥有而美国许多司法管辖区所缺乏的是:一个框架、强制性报告、附带执行机制的可再生能源义务、电网运营商的否决权、法律挑战途径和民主问责制。当出现问题时,爱尔兰有解决问题的结构。
Original English Source
Now, here's where I want to redirect the conversation. It's tempting at this point to throw your hands up in the air and say, 'Well, just stop building them.' But data centers are not going away. They support critical infrastructure that everybody relies on. The question isn't whether they should exist; the question is whether there is a framework to ensure they don't steamroll the communities they land in. It turns out other countries have already started building that framework. Ireland is a very interesting case study. Data centers consume 22% of Ireland's total metered electricity, up from 5% in 2015, with projections that this rises to roughly a third by 2026. In the Dublin and Meath region specifically, data centers account for about 50% of electricity consumption. So the scale of the challenge is comparable to, or arguably worse than, anything happening in the United States. Dublin effectively had a moratorium on new data center connections for 5 years, from 2021 to late 2025. But here is where it diverges. When Ireland lifted the moratorium, it didn't just throw the doors open. The Commission for Regulation of Utilities (CRU) introduced a binding obligation that 80% of data centers' annual electricity must come from additional renewable energy generated in Ireland, with a six-year glide path. New data centers must generate power back into the national grid. The system operators can reject proposed sites and reduce import capacity if facilities don't meet performance and availability standards. If a data center fails to deliver its required energy services, the grid operator can reduce its import capacity or move it to a non-firm status. There are actual consequences. At the EU level, the regulatory architecture goes even further. The revised Energy Efficiency Directive mandates standardized sustainability reporting for data centers above 500 kilowatt-hour with annual submissions to a European database. The EU Taxonomy Regulation classifies data center activities by climate mitigation criteria. The AI Act itself includes energy consumption transparency requirements for general-purpose AI models. And the European Commission is proposing a data center energy efficiency package aimed at making data centers carbon neutral by 2030. Is it perfect? Of course not. Environmental groups in Ireland are challenging the CRU decision in the high court, arguing that the 20% non-renewable allowance plus fossil fuel backup generators still lock the country into emission increases. Ireland's own environmental protection agency projects the state will overshoot its 2026 to 2030 carbon budget by 77 to 114 million tons. These are legitimate concerns that are being actively debated through democratic and legal processes. But here's what Ireland has that the United States in many jurisdictions simply doesn't: a framework, mandatory reporting, renewable energy obligations with enforcement mechanisms, grid operator veto power, legal challenge pathways, democratic accountability. When something goes wrong, there is a structure for addressing it.
美国日益加剧的反弹与前行之路
与此形成对比的是佐治亚州的费耶特县,那里只有一名检查员试图监控黑石集团子公司建设的、规模相当于一个小镇的园区。而当发现干旱期间未经授权的抽水行为时,机构的反应基本上是:“这叫客户服务。”这种反弹并未减弱,反而日益演变为选举问题。在密苏里州,一个小镇在批准了一个60亿美元的数据中心项目后,罢免了一半的市议会成员。在密歇根州,尽管一项价值160亿美元的“星门计划”(Stargate project)被投票否决,但仍被建设,此后城镇居民纷纷行动起来,阻止新的此类法案。在俄亥俄州,组织者正在征集签名,以发起一项全州范围的投票倡议,旨在禁止需要超过25兆瓦电力的数据中心。西雅图刚刚宣布了一项紧急暂停令,此前有五个拟议的数据中心将从城市电网额外抽取369兆瓦电力。丹佛、明尼阿波利斯、达勒姆,这个名单每周都在增长。在拥有全球最大数据中心集群的弗吉尼亚州,公众对新建数据中心的支持率已从2023年的69%降至目前的35%。这并非反技术情绪,而是社区发现工业巨型项目通过为住宅区设计的治理体系在其管辖范围内建设,并决定他们已经受够了。愤怒是真实的,不满是合理的,而且解决方案已经存在,只是它们被部署在其他国家。数据中心将继续建设,但问题在于,它们所落地的社区能否获得透明度、问责制和真正的发言权,还是下一个费耶特县居民必须通过《信息自由法案》才能知道为什么水压下降了。
Original English Source
Compare that with Fayette County, Georgia, where one inspector is trying to monitor a Blackstone subsidiary building a campus the size of a small town. And the institutional response to discovering unauthorized water extraction during a drought is basically, 'It's called customer service.' The backlash isn't slowing down; it is becoming electoral. In Missouri, a small town ousted half its city council after it approved a $6 billion data center. In Michigan, towns rush to block new bills after a $16 billion Stargate project was built despite being voted down. In Ohio, organizers are collecting signatures for a statewide ballot initiative to ban data centers requiring more than 25 megawatts of power. Seattle just announced an emergency moratorium after five proposed data centers would have demanded 369 megawatts from the city grid. Denver, Minneapolis, Durham, the list is growing week by week. In Virginia, the state that hosts the largest data center cluster in the world, voter support for new data centers has collapsed from 69% in 2023 to just 35% now. This is not anti-technology sentiment. It is communities discovering that industrial mega-projects were built inside their jurisdictions through governance systems designed for residential oversight and deciding they have just had enough. The anger is real. The grievances are legitimate. And the solutions already exist; they are just deployed in other countries. Data centers will continue to be built. But the question is whether the communities they land in get transparency, accountability, and a genuine seat at the table, or whether the next Fayette County resident has to file an FOIA request just to find out why the water pressure dropped.