数据中心反对派:接下来该怎么办?

收藏精彩文章, 图片, 视频, 连接等

版主: jack

回复
头像
jack
帖子: 160
注册时间: 周二 12月 12, 2006 10:17 am
联系:

数据中心反对派:接下来该怎么办?

帖子 由 jack » 周六 11月 21, 2026 2:39 pm

考虑一个事实:如今,美国约一半的经济增长都来自庞大的数据中心建设。我们说的不是股票价格,而是实际的消费支出。反过来说,如果停止所有数据中心的扩张,美国的经济增长就会减半。

再考虑另一个事实。2008 年之后的六年间,美国近一半的 GDP 增长都来自页岩能源技术的蓬勃发展,也就是所谓的“水力压裂革命”。如果当时反对水力压裂的人成功颁布禁令,那么美国很可能在那几年里从持续的大衰退(抵押贷款危机和银行业危机)滑向经济萧条。

当时,尽管页岩气禁令运动及其政治盟友不懈游说,但民主党控制的国会和奥巴马总统均未遵守禁令。颇具讽刺意味的是,在2012年与米特·罗姆尼的总统辩论中,奥巴马不仅将页岩气革命的功劳揽于己身(可以说他确实当之无愧,因为他没有扼杀这场革命),而且在其第二个任期内还签署了石油出口法案,进一步推动了页岩气的持续扩张。后者已成为美国乃至全球经济和地缘政治安全的一笔持久财富。

现在,看看我们当前大选后的各种新闻标题,比如《新共和》 杂志的一篇标题是“2026年大选日揭示了美国民众对硅谷数据中心项目有多么厌恶”。或者另一篇标题断言:“选民在全国范围内压倒性地反对数据中心。”[重点为笔者所加] 真的吗?

很少有哪个问题能像这个问题一样,如此直接地揭示竞选和执政之间的区别。

如果说出口民调传达了什么明确的信息,那就是选民最关心的问题是经济。事实上,在选民提及最多的六个问题中,“经济”和“外交政策”是民主党和共和党选民立场几乎完全一致的唯二问题。

盖洛普长期进行的月度追踪民调——该民调收集了“您认为美国面临的最重要问题是什么?”这一问题的答案——显示,对人工智能和科技的担忧并未进入前15名。盖洛普在10月份进行的另一项民调则要求受访者根据重要性对11个问题进行排序,以便决定他们是否会投票给国会。数据中心和气候变化虽然榜上有名,但排名垫底。

没错,11月3日,41项反对或暂停数据中心建设的投票提案中有40项获得通过,这一消息广为人知。但几乎所有这些提案都集中在三个州——其中超过一半在俄亥俄州,另外四分之一在伊利诺伊州和缅因州。这很难说是全国性的回应。(而且这三个州仅占所有规划数据中心总数的10%。)公平地说,在选举前,已有近600个社区对数据中心实施了“积极限制”。但是,背景很重要。美国有3000多个县和近2万个建制城镇和村庄。各地意见不一也是情理之中的事。

与此同时,在州一级,12位当选州长在竞选期间承诺暂停数据中心建设:阿拉斯加州、佐治亚州、堪萨斯州、缅因州、密歇根州、新罕布什尔州、新墨西哥州、纽约州、俄亥俄州、俄勒冈州、德克萨斯州和威斯康星州。(俄勒冈州、纽约州和德克萨斯州此前已禁止建设数据中心。)虽然这12个州并不代表全国趋势,但它们加起来占目前所有已规划(即尚未开工建设)数据中心总数的40%。

在政治上,用“暂停”而非“禁止”或许更为权宜。这在法律上也可能符合州长的职权范围。但是,考虑到数据中心行业建设的规模和速度,暂停与禁止在实际操作上并无本质区别。

一旦大型项目启动,延误的代价将非常高昂。面对不确定的停工期(许多州已明确表示停工将持续到明年),一种理性的应对方式是将资金转移到其他地方。我们可以预见,里斯顿定律将再次发挥作用:资本“会流向需要它的地方,并留在受到优待的地方”。

其他几十个州对开发商更为友好,包括俄亥俄州的邻州宾夕法尼亚州。宾夕法尼亚州议会并未通过一项拟议的多年禁令。相反,州长乔什·夏皮罗在选举前签署的行政命令规定了一些要求,这些要求可以说是任何大型建设项目都应遵循的常规。鉴于这些项目规模庞大,涉及数千亿美元,政治分析人士或许应该开始思考即将到来的资本格局重塑所带来的经济和政治影响。

当然,即使是友好州也会对选址审批附加条件。但城市研究所制定的这份清单列出了25个问题,供社区在数据中心破土动工前向开发商询问,这些问题本身并无不妥。总的来说,我们看到各方在三个共同原则上达成了共识,而大多数人工智能开发商也并不反对这三个原则:取消对数据中心的税收优惠、限制(或报告)用水量以及承担所有相关的电力成本。

尽管如此,所有这些妥协都无法解决推动职业反对运动的两大国民焦虑。对一些人来说,数据中心已成为人工智能对就业和文化影响以及大型科技公司在政治中扮演的角色等担忧的化身。此外,还有人声称数据中心正在对土地、水和能源造成前所未有的物理负担。

我们很难就技术对社会的影响达成共识,更遑论人工智能的利弊,以及大企业干预政治等问题。但那些试图取缔数据中心或阻碍数据中心建设的人,最终将不得不面对一个似乎被所有人忽视的现实:互联网的根基在于数据中心。

如今,大多数人——实际上,97%的美国成年人——每天的许多活动都依赖于互联网,而数据中心使这一切成为可能:电子商务、远程医疗、导航、银行服务、家庭作业和流媒体娱乐等等。即使在经历了数年的人工智能建设热潮之后,现有数据中心超过四分之三的运行容量仍然是传统计算,而非人工智能。

政策制定者面临的难题在于:民众已经用钱包和手机表达了他们的选择。他们自愿使用互联网提供的更多服务。这一趋势与对更多数字硬件的需求直接相关:仅在过去六年里,上​​游数字流量就翻了一番。再次强调,这主要来自传统的数字服务,而非人工智能(至少目前还不是)。

禁止或放缓数据中心扩张将会影响消费者的一切期望,包括他们对互联网服务不断改进和易用的预期。而人工智能能够且将会在带来那些令人兴奋或恐惧的全新事物之前,就带来这种改进。

至于批评人士不断提及建设和运营数据中心对能源、水和土地资源的影响,自从我今年早些时候撰写这篇总结以来,情况没有任何改变:

就能源而言,我们可以比较一下传输信息所需的电力与运输实物货物所需的电力。到2035年,如果美国最终建成所有目前预测的人工智能数据中心,那么它们的总耗电量将不到全国所有8级货运卡车总耗电量的十分之一。数据中心的电力需求似乎并不算高。

至于水和土地,数据中心消耗的水和土地远少于高尔夫球场,更不用说农业了。当初国会强制规定在汽油中添加玉米乙醇并提供补贴时,抗议者都去哪儿了?生产乙醇每加仑所需的水量几乎是汽油的200倍。

最后,关于土地利用:目前美国仓库的总面积几乎是数据中心的100倍。那么,当政客和开发商开始补贴太阳能发电设施,用黑色的、玻璃覆盖的硅片覆盖农田和脆弱的沙漠时,示威者又在哪里呢?

未来的法医心理学家们将会忙着研究为什么反对声浪如此迅速地蔓延开来。但开发商们真正担忧的是眼前的未来。目前,规划中的新产能大约是正在建设产能的两倍,这意味着数千亿美元的资金正处于风口浪尖。可以肯定的是,其中很多项目最终都会转移到对政府友好的州。

但笼罩整个行业的阴影是国会可能采取的行动。正如一则选后新闻标题所言:“民主党人利用人们对人工智能的焦虑赢得了胜利。现在,数据中心限制、传票和可能的前沿模型立法即将到来。”尽管反对和阻止新技术发展的运动鲜有成功,但此前的两个例子——一个成功,一个失败——却颇具启发意义。

如前所述,反对水力压裂法的运动失败了。这并非因为该运动资金不足,也不是因为公众反对的声音消失了。事实上,皮尤研究中心的数据显示,大多数人仍然反对进一步的水力压裂,就像奥巴马总统签署公开支持水力压裂法的法案时一样。页岩油禁令运动失败的最简单原因在于,目前尚无大规模替代石油的能源。石油需求一直在增长,而且仍在增长。(自2008年以来,美国石油产量的增长相当于此后全球石油需求增长总量的一半。)

反例可想而知,20世纪70年代那场政治色彩浓厚、夸张言辞充斥、虚伪做作的反核运动。这场运动在很大程度上终结了美国商业核能的扩张。由于各州和国会采取的行动(以及不作为),美国在1979年后的短短几年内,商业核电站的在建数量从全球近一半骤降至零。尽管人们的热情再次高涨,但美国至今仍未建成任何核电站。国会和整个国家都庆幸,这场事实上的核能禁令并未导致电力短缺——因为他们还有其他选择。美国并没有依赖核能,而是在1980年后的二十年里,将煤炭发电量翻了一番。

然而,对于数据中心而言,却没有其他选择。如果数据中心建设放缓是因为热情“泡沫”破裂,即开发商建设速度超过需求,那还情有可原。但如果是因为政客们强行放缓,即便需求不断增长,情况就完全不同了。

面对各国阻碍人工智能发展的政策,人工智能开发者唯一的选择似乎只有海外建设。值得注意的是,大选次日,谷歌宣布将原定于2024年启动的20亿美元计划扩大至150亿美元,用于在马来西亚建设数据中心。马来西亚拥有丰富的低成本电力资源,其电网约80%的电力来自煤炭和天然气。其他能源丰富的国家也可能向谷歌发出邀请——例如墨西哥,甚至加拿大。

新一届国会将于1月3日召开。届时,国会将不再忙于竞选,而是真正开始执政。目前尚不清楚多数派是否有意愿蓄意扼杀一个贡献了GDP一半增长的行业。一线希望是:据《国会山报》报道,环保人士和大型科技公司在大选前携手合作,致力于制定“负责任的数据中心发展国家共同框架”。正如人们常说的,细节决定成败。但这无疑是一个好兆头。


By Mark P. Mills

Consider a singular fact. About one-half of the nation’s economic growth today is coming from the great data-center buildout. We’re not talking stock prices but real-world spending. Put inversely, stop all data-center expansion and the nation’s economic growth cuts in half.

Consider another fact. For a half-dozen years after 2008, a similar one-half of the nation’s GDP growth came from the great shale energy-tech expansion, a.k.a. the “fracking revolution.” If the fracking opponents of that day had prevailed with a ban, the country would have tilted in several of those years from an ongoing Great Recession (the mortgage collapse and banking crisis) into depression.

Back then, despite the frack-ban movement’s relentless lobbying and its political allies, neither the Democrat-controlled Congress nor President Barack Obama complied. In no small irony, during the 2012 presidential debate with Mitt Romney, Obama not only took credit for the shale revolution (arguably deservedly, for not squelching it), but also in his second term signed the petroleum export legislation that helped fuel continued shale expansion. The latter has been an enduring gift to the U.S. and global economies and geopolitical security.

Now, consider myriad headlines in our current postelection season, such as one from The New Republic crowing that “Election Day 2026 revealed just how much of the country loathes Silicon Valley’s data center projects.” Or another asserting: “Voters overwhelmingly say no to data centers nationally.” [emphasis added] Really?

Few issues will be as immediately consequential as this one in illuminating the difference between campaigning and governing.

If the exit polls made any message clear, it was that the top issue for voters was the economy. In fact, of the top six issues cited, “the economy” and “foreign policy” were the only two with near-identical alignment between Democrats and Republicans.

In Gallup’s long-running monthly tracking poll—cataloging answers to “what do you think is the most important problem facing the country?”—worries about AI and tech don’t show up in the top 15 concerns. A separate Gallup poll in October prompted respondents to rank 11 problems by importance for their congressional vote. Data centers and climate change made the list but were ranked as the bottom two items.

True, November 3 saw the widely reported passage of 40 out of 41 ballot initiatives that oppose or would pause data centers. But nearly all those measures were in just three states—with over half in Ohio alone and another quarter in Illinois and Maine. Hardly a national response. (And those three states are home to just 10 percent of all planned data centers.) To be fair, preelection, nearly 600 communities had already imposed “active restrictions” on data centers. But, again, context matters. The nation has over 3,000 counties and nearly 20,000 incorporated towns and villages. It’s reasonable to expect that opinions may vary.

Meanwhile, at the state level, 12 elected governors made campaign promises involving a data-center pause: Alaska, Georgia, Kansas, Maine, Michigan, New Hampshire, New Mexico, New York, Ohio, Oregon, Texas, Wisconsin. (Oregon, New York, and Texas already had bans.) While those dozen states don’t constitute a national trend, they do collectively account for 40 percent of all data centers now planned, i.e., not under construction.

It may be politically expedient to say “pause” rather than “ban.” It’s also likely legally consistent with a governor’s purview. But, given the financial stakes and the velocity of the data-center industry buildout, a pause is a distinction with no practical difference from a ban.

Once a giga-scale project is under development, delays can be very costly. Facing an uncertain length of pause, which many states have made clear will stretch well into next year, one rational response will be to go elsewhere. We can expect Wriston’s Law to reveal itself again: Capital “will go where it is wanted, stay where it is well-treated.”

Several dozen other states are friendlier to developers, including Ohio’s neighbor Pennsylvania, where the legislature did not enact a proposed multiyear moratorium. Instead, Governor Josh Shapiro’s preelection executive order imposed requirements that arguably fall in the category of normal for large-scale construction of any kind. Given just how big these projects are, entailing hundreds of billions of dollars, political analysts might start thinking about the economic and political fallout from the coming tectonic geographic realignment of capital.

Of course, even friendly states have conditions for site approvals. But it would be hard to take issue with any of the 25 questions the Urban Institute created as a checklist for communities to ask developers about “before a data center breaks ground.” And, in general, we’ve seen a convergence around three common principles, none of which most AI developers oppose: ending tax breaks for data centers, limiting (or reporting) water use, and paying for all associated power costs.

That said, all such accommodations won’t resolve the two big categories of national anxieties fueling the professional opposition movement. For some, data centers have become avatars for worries about AI’s impact on jobs and the culture, as well as the role of Big Tech in politics. And then there are the claims that data centers are imposing unprecedented physical burdens on land, water, and energy.

We will not easily settle debates about social impacts of technology in general, never mind about what AI promises or threatens, nor the issues of big business in politics. But those who would ban data centers or make it harder to build more of them will eventually have to deal with a reality that seems to have escaped everyone’s attention: the internet exists in data centers.

So much of what most people—indeed, 97 percent of U.S. adults—already do every day depends on using the internet, and data centers make it possible: not least e-commerce, virtual doctor visits, navigation, banking, homework, and streaming entertainment. Over three-fourths of the operating capacity in existing data centers is still conventional compute, not AI, even now, after several years of furious AI construction.

Here’s the rub for policymakers. Citizens have voted with their pocketbooks and pocket phones. They are voluntarily using far more of what the internet already offers. That trend is visible in terms with direct relevance for needing more digital hardware: upstream digital traffic has doubled in the past half-dozen years alone. Again, that’s coming mainly from conventional digital services, not because of AI (not yet).

Banning or slowing data-center expansion will hit everything that consumers have come to expect, including the expectation that internet-centric services will continually get better and easier to use. That improvement is precisely what AI can and will deliver long before it can deliver on the entirely new things that excite or frighten people.

As for critics’ constant referral to resource impacts on energy, water, and land, from building and operating data centers, nothing has changed since I wrote this summary earlier in the year:

Regarding energy, we can compare the power needed to move information versus the power required to transport physical goods. By 2035, if the U.S. ends up with all the AI data centers now forecast, the total power required by all of them will be less than one-tenth the total collective power under the hoods of all Class 8 trucks moving goods on the nation’s roads. The data-center demand hardly seems a heavy lift.

As for water and land, data centers use far less of both than golf courses, never mind agriculture. Where were the protesters when Congress enacted mandates and subsidies for corn-ethanol in gasoline? Nearly 200 times more water is used per gallon to produce ethanol than gasoline.

Finally, on land use: the total square footage of warehouses in America is, at present, almost 100-fold greater than for data centers. Again, where were the demonstrators when politicians and promoters started subsidizing solar power installations that blanket farmland and fragile deserts by the square mile with black, glass-covered silicon?

Forensic psychologists of the future will have a field day trying to figure out why opposition became so pervasive so fast. But it’s the near future that developers worry about. Currently, there’s roughly twice as much new capacity planned compared with what’s under construction, meaning hundreds of billions of dollars are in play. You can take it to the bank that a lot of those plans will move to friendly states.

But what looms over the industry is what Congress might yet try to do. As one postelection headline put it: “Democrats rode AI anxiety to victory. Now come data center limits, subpoenas and possible frontier model legislation.” While movements to oppose and halt new technologies have rarely succeeded, two earlier examples—one that succeeded and one that failed—are instructive.

Opposition to fracking, as noted, failed. Not because the movement wasn’t well-funded, or public opposition disappeared. Indeed, as Pew Research shows, a majority still oppose more fracking, as they did when President Obama signed into law overtly pro-fracking legislation. The simplest reason for the failure of the shale ban movement was that there remains no alternative to oil at scale. Demand was, and is, rising. (Increased U.S. oil production since 2008 equals half of all the world’s rise in demand since then.)

For the counterexample, consider the politically animated, hyperbole-infused, and disingenuous anti-nuclear movement of the 1970s. Opposition was responsible, in large measure, for ending U.S. commercial nuclear energy expansion. Through a combination of state and congressional actions, and inactions, the U.S. went, in a handful of years after 1979, from having nearly one-half of the world’s commercial nuclear power plants under construction to none. It’s still none, despite rejuvenated enthusiasm. Congress and the nation were lucky that this de facto nuclear ban didn’t cause an electricity shortage—that’s because there were options. Rather than rely on nukes, the U.S. doubled its coal consumption for electricity in the two decades following 1980.

With data centers, however, alternatives don’t exist. It’s one thing if the data-center buildout slows because a “bubble” of enthusiasm bursts, i.e., developers build faster than demand. But if a slowdown happens because politicians force one, despite rising demand, that’s a horse of a different color.

In the face of national sand-in-the-gears policies, the only option open to AI developers is to build overseas. Notably, the day after the election, Google announced a $15 billion expansion to a $2 billion plan launched in 2024 to build data centers in Malaysia, a country with lots of low-cost electricity, with some 80 percent of its grid fired with coal and gas. Other energy-rich countries might yet beckon—Mexico, even Canada.

The new Congress convenes on January 3. Then it will be time not to campaign but to govern. The jury is out on whether a majority has the appetite to try and deliberately squelch a sector driving one-half of GDP growth. One ray of hope: according to The Hill, environmental activists and Big Tech joined together before the election to work on a “common national framework for responsible data center development.” The devil is, as they say, in the details. But this is a good sign.
宁天下人负我,莫让我负天下人!
http://www.life2fun.com/

回复

在线用户

正浏览此版面之用户: 没有注册用户 和 0 访客