New evidence on data center employment effects

This piece has been updated to reflect new results as the research continues with an expanded sample.

随着研究继续扩大样本范围，这篇文章已经更新，以反映新的结果。

Previous findings related to data center effects on employment, wages, and home prices have been revised.

之前有关数据中心对就业、工资和房价影响的研究结果已被修改。

The brief summarizes findings from “Data Centers and Local Labor Markets” by Dany Bahar and Greg Wright.

该简报总结了 Dany Bahar 和 Greg Wright 撰写的《数据中心和当地劳动力市场》的调查结果。

The full paper is available from the authors.

论文全文可从作者处获取。

On March 25, 2026, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced the AI Data Center Moratorium Act, a bill that would pause all new large-scale AI data center construction until Congress passes legislation addressing AI safety, worker protections, and environmental standards.

2026 年 3 月 25 日，参议员伯尼·桑德斯和众议员亚历山大·奥卡西奥-科尔特斯提出了《人工智能数据中心暂停法案》，该法案将暂停所有新的大规模人工智能数据中心建设，直到国会通过有关人工智能安全、工人保护和环境标准的立法。

The bill reflects a sharp turn in the national conversation.

该法案反映了全国对话的急剧转变。

Five years ago, data centers were ribbon-cutting events.

五年前，数据中心还只是剪彩活动。

Today, more than 100 local communities have enacted moratoriums , more than 300 state data-center bills were filed in the first six weeks of 2026, and several states that once competed to offer the largest tax incentives—Virginia, Georgia, and Oklahoma—are now reconsidering those programs entirely.

如今，100 多个当地社区已颁布暂停令，2026 年前六周内提交了 300 多项州数据中心法案，曾经竞相提供最大税收优惠的几个州——弗吉尼亚州、佐治亚州和俄克拉荷马州——现在正在完全重新考虑这些计划。

The backlash is not hard to understand.

这种强烈反对并不难理解。

Data centers consume enormous amounts of electricity and water.

数据中心消耗大量的电力和水。

In the PJM grid region, which serves 65 million people across 13 states, power supply costs jumped from $2.2 billion to $14.7 billion in a single year, with data centers accounting for nearly two-thirds of the increase.

在为 13 个州的 6500 万人提供服务的 PJM 电网地区，供电成本在一年内从 22 亿美元跃升至 147 亿美元，其中数据中心占增量的近三分之二。

Residential electricity rates nationally rose about 32% between July 2020 and July 2025.

2020 年 7 月至 2025 年 7 月期间，全国住宅电价上涨约 32%。

Communities near proposed facilities face noise, strained infrastructure, and the loss of farmland, costs that are immediate and visible.

拟建设施附近的社区面临着噪音、基础设施紧张和农田流失等直接且可见的成本。

Proponents counter that data centers bring high-paying jobs, construction activity, and tax revenue.

支持者反驳说，数据中心带来了高薪工作、建筑活动和税收。

A single hyperscale campus, such as those built by Amazon, Google, and Microsoft, can become one of a county’s largest taxpayers .

单个超大规模园区，例如亚马逊、谷歌和微软建造的园区，可以成为一个县最大的纳税人之一。

Supporters also argue that the United States cannot cede AI infrastructure to geopolitical competitors, and that pausing construction amounts to surrendering technological leadership.

支持者还认为，美国不能将人工智能基础设施让给地缘政治竞争对手，暂停建设就等于放弃技术领先地位。

Critics respond that these benefits are overstated, in part because data centers are among the least labor-intensive structures in the economy.

批评者回应称，这些好处被夸大了，部分原因是数据中心是经济中劳动密集度最低的结构之一。

For instance, large data center projects often promise only dozens to a few hundred permanent workers while the associated construction jobs are temporary.

例如，大型数据中心项目通常只承诺几十到几百名长期工人，而相关的建筑工作都是临时的。

And the tax incentives are costly: In Virginia alone, the data-center sales-tax exemption cost an estimated $1.6 billion in fiscal year 2025.

而且税收优惠的成本高昂：仅在弗吉尼亚州，2025 财年数据中心销售税豁免的成本就估计达到 16 亿美元。

These incentives may simply be subsidizing investments that would have happened anyway.

这些激励措施可能只是补贴无论如何都会发生的投资。

What has been largely missing from this debate is rigorous evidence on the economic effects that data centers actually produce.

这场辩论很大程度上缺少关于数据中心实际产生的经济影响的严格证据。

To this end, we assembled a dataset of approximately 1,500 U.S. data center facilities along with 52 announced-but-canceled projects and we linked these to county-level employment and wage data from the Bureau of Labor Statistics (2003-2024).

为此，我们收集了大约 1,500 个美国数据中心设施以及 52 个已宣布但已取消的项目的数据集，并将这些数据与美国劳工统计局 (2003-2024) 的县级就业和工资数据联系起来。

1

1

We estimate the impact of the data centers on jobs by comparing labor markets with built facilities to labor markets with announced-but-canceled facilities, which controls for the fact that data center sites are chosen with specific characteristics in mind (availability of cheap land, power, fiber, and recent economic growth).

我们通过将已建设施的劳动力市场与已宣布但取消的设施的劳动力市场进行比较来估计数据中心对就业的影响，这控制了数据中心选址时考虑到特定特征（廉价土地、电力、光纤的可用性和近期经济增长）这一事实。

This approach is important because before the facilities arrived, data center counties were growing faster than other counties.

这种方法很重要，因为在设施到达之前，数据中心县的增长速度比其他县更快。

As a result, a naive comparison of treated and control counties, as some industry-sponsored reports have done, would overstate the impact on jobs.

因此，像一些行业赞助的报告所做的那样，对治疗县和对照县进行天真的比较会夸大对就业的影响。

We find that data centers do create local jobs, with caveats.

我们发现数据中心确实创造了当地就业机会，但也有一些警告。

Labor markets that receive their first large data center see employment rise in the data processing sector by 56% over the first decade of operations.

接收第一个大型数据中心的劳动力市场发现，在运营的第一个十年中，数据处理行业的就业人数增加了 56%。

Telecommunications, a key sector in data center operations, jumps 43%.

电信作为数据中心运营的关键行业，增长了 43%。

Wages were unchanged, and we find a modest increase in home prices of 2%-5%.

工资没有变化，我们发现房价小幅上涨了 2%-5%。

At a typical treated county, these estimates imply roughly 100-200 jobs, depending on facility type.

在一个典型的接受治疗的县，这些估计意味着大约 100-200 个工作岗位，具体取决于设施类型。

(Figure 1 shows the yearly estimates, which are noisier than the average for the whole period.)

（图 1 显示了年度估计值，该估计值比整个时期的平均值更加嘈杂。）

Note: Each panel plots the average gap between treated markets and their control markets, with confidence intervals.

注意：每个面板都绘制了处理市场与其对照市场之间的平均差距，并带有置信区间。

Event time 0 corresponds to the year of first large data center opening.

事件时间 0 对应于第一个大型数据中心开放的年份。

The three sector panels share a common y-axis.

三个扇区面板共享一个共同的 y 轴。

One policy-relevant finding is that the employment effects depend on facility type.

一项与政策相关的发现是，就业影响取决于设施类型。

We classify facilities into two types: hyperscale (built by cloud and AI companies—Amazon, Google, Microsoft, Meta—to run their own workloads) and colocation (built by data center landlords—Equinix, Digital Realty, CyrusOne—who lease space to remote tenants).

我们将设施分为两种类型：超大规模（由云和人工智能公司（亚马逊、谷歌、微软、Meta）建造，用于运行自己的工作负载）和托管（由数据中心业主（Equinix、Digital Realty、CyrusOne）建造，他们向远程租户租赁空间）。

This distinction, absent from prior research and current policy, turns out to be critical.

先前的研究和当前政策中没有这种区别，但事实证明这一区别至关重要。

We see in Figure 1 that both types of data centers see growth in the data processing sector.

我们从图 1 中看到，两种类型的数据中心都在数据处理领域实现了增长。

However, hyperscale counties also see telecommunications sector gains while colocation counties do not.

然而，超大规模县也看到了电信行业的增长，而托管县却没有。

This is because a hyperscale campus creates demand for high-capacity fiber connections and network operations centers whereas a colocation facility leases space to remote tenants who may have no local operational presence.

这是因为超大规模园区产生了对大容量光纤连接和网络运营中心的需求，而托管设施则向可能没有本地运营业务的远程租户租赁空间。

Communities have legitimate concerns about energy costs, water use, and environmental impact, and those concerns deserve evidence-based answers of their own.

社区对能源成本、用水和环境影响有合理的担忧，这些担忧应该得到基于证据的答案。

But the labor market dimension of this debate has been conducted largely without data, and our findings suggest several things policymakers should know:

但这场辩论的劳动力市场层面基本上是在没有数据的情况下进行的，我们的研究结果表明政策制定者应该了解以下几点：

The data center debate is moving fast, but the evidence base has not kept pace.

关于数据中心的争论正在迅速发展，但证据基础却没有跟上。

Whether the question is a federal moratorium, a state incentive program, or a local zoning decision, policymakers will require rigorous evidence in order to make these types of decisions.

无论问题是联邦暂停、州激励计划还是地方分区决定，政策制定者都需要严格的证据才能做出这些类型的决定。