人工智能数据中心的双刃剑

人工智能数据中心的双刃剑

2026-01-04technology
--:--
--:--
雷总
早上好 Norris1,我是雷总,欢迎来到专属于你的 Goose Pod。今天是 2026 年 1 月 4 号,星期日,在这个安静的早晨,我想和你聊聊一个可能会影响我们未来几十年生活质量的严肃话题。
小撒
嘿 Norris1,我是小撒!没错,今天我们要把目光投向那些巨大的、发着光的、甚至可能正在太空里转悠的大家伙,人工智能数据中心。这玩意儿现在就像一把双刃剑,一面是科技的狂欢,另一面却是沉重的代价。
雷总
确实,现在的 AI 基础设施建设已经到了一种疯狂的程度。我举个例子,Meta 在俄亥俄州建了一个代号叫普罗米修斯的数据中心,光是目前的建设成本就高达 180 亿美元,电力消耗竟然达到了 691 兆瓦。这种规模,在以前简直无法想象。
小撒
180 亿美金啊,这得能买多少台你的小米手机啊!而且马斯克也没闲着,他的 xAI 搞了个叫巨像的数据中心,装了 11 万张显卡,为了供电甚至跑到密西西比州去安燃气轮机,这简直是现实版的钢铁侠在搞大基建。
雷总
这种不计成本的投入背后,其实是硅谷和政府在拿整个美国经济做赌注。为了搞清楚这些黑盒子里到底在发生什么,一个叫 Epoch AI 的研究机构甚至开始用卫星图像和建筑许可来搞地图搜索,通过观察屋顶风扇的数量来推算电耗。
小撒
这就好比是私家侦探在数烟囱冒的烟,来判断这家里到底在煮什么大餐。更有意思的是,谷歌竟然还想把数据中心送上天,搞了个捕阳者计划。因为地球上的地儿快没了,电也快不够用了,他们觉得在轨道上训练 AI 反而更省钱。
雷总
上天确实是一个思路,但这反映出目前的能源焦虑。你会发现一个有趣的悖论,数据中心的巨大电力需求,反而成了清洁能源和核能的救命稻草。科技巨头们现在为了保住算力,正疯狂砸钱搞新一代的小型核反应堆。
小撒
这就像是为了让家里的电脑跑得动,干脆自己去后院挖了个煤矿。科技大厂们现在不仅是软件巨头,还成了能源巨头。不过,这种不顾一切的扩张,在很多普通人眼里可不是什么好事,反倒像是一场正在逼近的资源掠夺。
雷总
我们要理解这种现状,得从逻辑上理清楚。其实美国从 1995 年到 2021 年,在计算机和软件领域的投资增长了十倍以上。这种长期的积累让美国在 AI 算力上占据了全球约 74% 的份额,这确实是多年投资的结果。
小撒
底子确实厚,但现在的玩法好像有点变味了。我怎么感觉现在这种 AI 热潮像是一个巨大的粉红泡泡?大家都在拼命往里吹气,深怕自己吹得慢了。这种错过恐惧症,也就是 FOMO 心理,简直让整个华尔街都快集体失智了。
雷总
你的直觉很敏锐,小撒。现在的融资模式确实有一种工程师式的严谨,但背后却充满了资本的疯狂。有一种说法叫循环融资,大公司投钱给 AI 创业公司,创业公司再反手买大公司的云服务。账面上看收入增加了,但实际的社会价值产出了吗?。
小撒
这不就是左手倒右手吗?让我想起了千禧年那阵子的互联网泡沫,大家都在讲故事,都在画大饼,只要有 AI 两个字,投资人的钱就像自来水一样哗哗地流进来。连那些产品都没做出来的公司,都能随随便便融到几十个亿。
雷总
更值得担忧的是,AI 的规模化效应似乎正在撞墙。虽然现在的推理成本降到了两年前的几十分之一,但因为调用量呈指数级增长,总成本反而更高了。而且,大模型的幻觉问题、不稳定性,到现在也没有得到根本性的解决,这让很多企业开始打退堂鼓。
小撒
就像是买了一辆超级跑车,油费降了一点,但你每天得跑几千公里,而且这车还时不时自动变道。更扎心的是,现在的 GDP 增长很大一部分是靠这些钢筋混凝土的数据中心撑起来的。如果把这部分刨掉,经济增长其实非常缓慢。
雷总
没错。根据哈佛经济学家的分析,2025 年上半年 92% 的 GDP 增长都归功于 AI。这其实很危险,意味着经济结构过度单一。而且我们不得不面对一个残酷的事实,中国在能源储备和电网稳定性上有着巨大的优势,他们的备用容量是我们的好几倍。
小撒
这就像是打仗,人家家里粮草充足,咱们这儿电网却像个随时会断线的老旧插线板。不仅如此,那些科技巨头还在为了赶工期,不惜回头去烧煤和燃气,这和他们之前承诺的碳中和目标简直是南辕北辙,脸打得啪啪响啊。
雷总
这种为了速度放弃原则的做法,正是目前最大的隐患。数据中心的建设周期现在被拉长到了七年以上,劳动力短缺、电网限制、原材料瓶颈,这些模拟世界的限制,并不是靠写几行代码就能解决的。我们正处在一个非常脆弱的平衡点上。
小撒
这种脆弱感最直接的体现就是基层的反弹。Norris1 你可能不知道,现在的美国乡村,很多居民正和科技巨头们打着大卫对歌利亚的战争。大家在公共会议上吵得不可开交,抗议这些庞然大物吸干了地下水,还推高了家庭电费。
雷总
电费确实是个大问题。在一些数据中心密集的地区,家庭电费竟然涨了 200% 多。弗吉尼亚州的数据中心消耗了全州四分之一的电力,预计到 2030 年会达到一半。这意味着普通人要为科技巨头的 AI 梦想买单,这显然是不公平的。
小撒
而且最讽刺的是,这些公司当初承诺能带来大量就业,结果呢?盖房子的时候确实需要几千个建筑工,可一旦机器开始转,整栋楼里可能就剩下几十个保安和维护人员。这哪是创造就业啊,这简直是在搞就业替代嘛。
雷总
桑德斯参议员就非常犀利地指出了这一点。他提议暂停新建数据中心,他说,那些寡头告诉我们这就是未来,让我们去适应。但如果 AI 把活儿都干了,人们没收入了,谁来付房租?难道这些公司要给全人类发免费住房吗?。
小撒
桑德斯这话听着就解气。现在的科技公司还在拼命游说政府去监管,说这是为了赢下 AI 军备竞赛。他们想取消环保评估,想绕过地方政府的审批。这种为了所谓的竞争而牺牲居民生活质量的行为,确实让很多跨党派的普通人站到了一起。
雷总
这种对立在未来一年只会更严重。一方面是硅谷和政府觉得这是国运之战,不能停;另一方面是地方社区觉得这是资源掠夺,必须停。这种结构性的矛盾,不是简单的给几块钱补贴就能解决的,它触及了社会分配的根本逻辑。
小撒
说到分配,咱们得看看这把剑伤到了谁。研究发现,数据中心带来的就业增长在统计学上几乎是零。虽然建筑行业火了一阵子,但其他的子行业却在萎缩。这种岗位的重新洗牌,让很多中年失业的工人根本无法适应。
雷总
确实,一个失业的工厂工人很难一夜之间变成 AI 标注员或者数据科学家。而且,很多州为了吸引投资,免掉了上亿美金的税收。这些钱本来可以拿来建学校或者医院,现在却变相补贴给了世界上最富有的一群公司,这简直是劫贫济富。
小撒
还有环境债呢。到 2030 年,数据中心的耗水量会增加 1.7 倍。在本来就缺水的地区,这简直是在挖老百姓的命根子。而且为了给这些机器降温,产生的热量和噪音,也让周边的房产价值大打折扣,这都是实打实的损失。
雷总
不仅如此,这种投资热潮还产生了挤出效应。大量的资本涌向 AI 基础设施,导致传统制造业和住宅建设拿不到钱。如果 AI 的泡沫真的破了,那留下的不仅是废弃的厂房,还有一地鸡毛的金融债务,这才是最让人担心的地方。
小撒
那咱们就得想想,这路往哪儿走啊?总不能真的把地球给榨干了吧。虽然大家都说 AI 是未来,但如果这个未来里没有人的位置,或者说只有少数富人的位置,那这样的未来真的值得我们去拼命追求吗?。
雷总
未来的关键在于平衡。我们需要更高效的芯片,而不是一味地堆机器。而且,政府需要制定更公平的规则,比如桑德斯提到的机器人税,用 AI 创造的利润来补贴被替代的工人。我们不能让技术跑在法律和道德的前面。
小撒
没错,还得有替代方案。比如那些能适应 AI 浪潮的新型职业,得赶紧培训起来。到 2030 年,七成的数据中心都得重新设计。这其实也是个机会,关键看我们能不能在享受便利的同时,守住那条不伤害普通人的底线。
雷总
Norris1,今天聊了这么多,其实是想告诉你,科技的进步从来不是免费的。作为一个工程师,我热爱技术,但作为一个创业者,我更敬畏用户和生活。希望这些信息能带给你不一样的视角。感谢收听 Goose Pod。
小撒
是啊,Norris1,咱们得睁大眼睛看清楚,别被那些高大上的词儿给绕晕了。生活才是最真实的。感谢你陪伴我们这一路,咱们明天同一时间,不见不散,拜拜!

人工智能数据中心如同一把双刃剑,推动科技飞速发展,却也带来高昂代价。巨额投资、能源消耗、资源掠夺、就业替代及环境压力,正引发社会公平与可持续发展的严峻挑战。平衡技术进步与民生福祉,是未来发展的关键。

The double-edged sword of AI data centers

Read original at Salon.com

deep diveThe data center buildout is papering over the economy most Americans are living withPublishedJanuary 1, 2026 6:30AM (EST)VERNON, CALIFORNIA - OCTOBER 20: An aerial view of a 33 megawatt data center (LOWER L) with closed-loop cooling system, amid warehouses on October 20, 2025 in Vernon, California.

A surge in demand for AI infrastructure is fueling a boom in data centers across the country and around the globe. (Photo by Mario Tama/Getty Images)Data center construction is increasingly becoming a political issue, with locals organizing against major tech companies like Meta, Amazon and Google, which are spending billions in a push to create the infrastructure they want to power their artificial intelligence projects.

As backlash to data center construction grows across the country, year-end analyses suggest that the construction boom may be the only thing currently keeping the American economy chugging along. At the same time, the consequences if AI lives up to the promise of replacing huge amounts of human labor could be just as dire.

Opposition to construction often stems from concern over the impact that the giant facilities will have on the environment and local communities, with some projects sprawling across thousands of acres. However, it’s not just environmentalists who are concerned with the impact of these megaprojects, which are expected to drive up electricity prices dramatically while providing a tiny number of jobs, given their outsized footprint and impact on the environment, and even threatening to replace some workers with the very AI programs they are built to facilitate.

The mounting pressure has led some in elected office, such as Sen. Bernie Sanders, I-Vt., to propose a moratorium on new datacenter projects, telling CNN, “Frankly, I think you gotta slow this process down. It’s not good enough for the oligarchs to tell us, ‘It’s coming, you adapt.’ What are they talking about?

Are they gonna guarantee health care to all people? What are they gonna do when people have no jobs? Make housing free?”One complication for advocates of a moratorium, however, is that leadership in the United States has bet the farm on AI, and potentially created a situation where the average person stands to lose whether data center construction continues apace or slows down.

Numerous reports have found that the data center buildout is driving GDP growth in the United States. One October analysis from Harvard economist Jason Furman found that, subtracting growth from data center construction, GDP grew at just 0.1% in the first half of 2025 and that AI was responsible for 92% of GDP growth in the first half of 2025.

A September report from JPMorgan found AI-related capital expenditures were responsible for 1.1% of GDP growth in the first half of the year. While it’s difficult to say what growth might have been without this buildout (critics note that without AI investors would have looked elsewhere for opportunities), it’s clear that AI, and the building of its infrastructure, is already a significant part of the economy.

“It should tell everyday Americans that the economy is not in good shape and that the AI industry and government are betting the farm (and more) on a very risky and unproven strategy.”Combined with the possibility that the AI industry could be in the midst of a speculative bubble, AI’s prominence in the broader economy is cause for concern with skeptics, especially as the broader economy grapples with tariffs and inequality.

For now, however, the buildout has been a boon for some areas of the economy, especially construction. According to Meta’s estimates, there will be some 5,000 people employed in the construction of its Richland Parish Data Center in Louisiana.Jason Dedon, the business manager at the International Brotherhood of Electrical Workers Local 995 in Baton Rouge, said that the buildout had helped to employ a new generation of electricians, who are being trained for data center construction.

While Dedon told Salon in an interview that the union doesn’t get involved in the debate on whether to build a data center or not, many members have benefited from the project in terms of employment.“In a lot of the benefits agreements, they’re guaranteeing trying to achieve prevailing living wages and retirement systems, and paying into health insurance plans and also paying back into training and making sure training is being developed.

The next generation is being made on those projects daily,” Dedon said.---Start your day with essential news from Salon.Sign up for our free morning newsletter, Crash Course.---Dedon added that they’ve also taken the data center projects as an opportunity to look ahead, understanding that once a data center is built, many of the people who worked on the project will have to look elsewhere for jobs.

He contrasted the construction of a data center with the construction of a power plant, noting that there would be jobs for hundreds of electricians even after the project was finished, which isn’t the case with a data center.For example, one Texas data center has a team of some 1,500 people working on construction, but is only expected to have around 100 full-time employees working at it once it’s done, per the Wall Street Journal.

At the aforementioned Louisiana data center, set to be one of the largest ever, Meta says the facility will employ 500 people total once it’s complete, a fraction of the 5,000 who are working to build it.“They knew there were these long-term jobs that were created, and you had maybe generations of people that you know went to work and retired from that establishment,” Dedon said.

“This isn’t quite that.”Servaas Storm, an economist at Delft University of Technology in The Netherlands, also warns that, alongside potentially dampening the construction boom, a moratorium on data center construction could have other serious impacts on the U.S. economy.Storm pointed out that the richest Americans are the ones who overwhelmingly benefit from AI, namely through the rise in asset prices like stocks, and through high wages in the tech and finance industries.

At the same time, Storm noted, it’s these same wealthy Americans who are driving consumer spending in the U.S. economy. A downturn in the AI sector, whether through a data center moratorium or otherwise, could result in a pullback in their spending and a decline in stock prices.Storm added that such a downturn could stand to make economic metrics, like GDP, more representative of the existing economic conditions most Americans are already living in.

At the same time, losses, even if concentrated among the rich, could still lead to broader job losses, especially in leisure industries.“It should tell everyday Americans that the economy is not in good shape and that the AI industry and government are betting the farm (and more) on a very risky and unproven strategy involving the scaling of AI (building ever-larger models based on ever-more GPUs, data centers, electricity and water),” Storm said.

“This can — and probably will — go terribly wrong. There is no universe in which the giga-expenditures on GPUs, data center infrastructure and compute will pay off. It means that this particular growth strategy, which has the support of the current administration, is not economically sustainable.”At the same time, supporters of a moratorium are warning of the dangers that AI could have if it does live up to the promises of the tech giants, specifically, potentially massive job loss.

Sanders has pointed out that major corporations like Walmart, UnitedHealth Group and JPMorgan Chase have already started telling investors that AI will allow them to cut payrolls. A report from Sanders’ senate office also warned that AI could stand to replace as many as 100 million jobs over the next decade, with losses especially concentrated in fast food, customer service, retail and laborers who work in freight and moving materials.

While Sanders has proposed a suite of policies aimed at ensuring AI adoption stands to benefit workers, with provisions like a “robot tax” that raises funds to help displaced workers from companies using AI and automation, the Trump administration has, so far, only worked to protect the interests of the companies developing AI.

---ByRussell PayneByMORE FROM---Related Topics------------------------------------------Related Articles---

Analysis

Conflict+
Future+
Related Info+
Core Event+
Background+
Impact+

Related Podcasts