Why the AI boom makes inflation harder to tame

Brendan McDermid/Reuters

The US economy may have a new problem: It’s too strong.

America’s cost-of-living concerns and a bond market meltdown aren’t functions of a sputtering economy that’s running out of gas. Quite the opposite.

Prices and interest rates are rising too quickly for comfort for three primary reasons: the global energy shock from wars in Iran and Ukraine, an escalating trade war and an unprecedented corporate spending spree on artificial intelligence.

Policy decisions (wars and tariffs) are garnering significant attention as gas and diesel prices surge and a US trade war buds with Canada. But economists are increasingly concerned about the mind-boggling amounts of Big Tech money committed to the AI data center buildout.

The many trillions of dollars expected to be spent on AI infrastructure over the next several years would facilitate a fundamental reshaping of the US economy. All that spending risks overheating the economy at a time when it’s already running hot: The unemployment rate is low, and consumer spending is strong.

That’s a recipe for higher inflation. And inflation is already too high .

It’s hard to overstate how extraordinarily massive the AI spending boom has become after it rapidly exploded onto the scene.

AI infrastructure spending – on data centers and the chips and servers that go inside them – is expected to hit around $1 trillion this year, according to JPMorgan. That’s more than the federal government spends annually on the military.

But we’re just in the beginning stages of the AI buildout.

Through 2032, that number will reach $10.3 trillion, claims Columbia University economist Stijn Van Nieuwerburgh in a paper published Wednesday by the Brookings Institution. For perspective: Remember the $1.2 trillion bipartisan infrastructure law signed by then-President Joe Biden in 2021 that became an inflation bugaboo? Projected AI expenditures are equivalent to spending every last dollar allocated from one of those bills each year over the course of a decade.

Benjamin Fanjoy/Getty Images

That spending – just on AI infrastructure, not on the technology itself – is on pace to comprise 1.9% of all US economic activity this year, according to Goldman Sachs. But that’s forecast to effectively double: On average, AI infrastructure spending will top 3.6% of America’s gross domestic product each year through 2032, according to Van Nieuwerburgh.

That projection is gargantuan, equivalent to the output from mammoth American economic engines, such as transportation, restaurants and hotels. But those are all mature industries. Even more shockingly, Van Nieuwerburgh projects the AI buildout will make up a larger share of total US economic output than any of America’s previous investment booms – bigger than when the United States built its canals, railroads, electrical grid, highways and telecommunications networks.

“All of this will result in a structural transformation of the US economy to one organized around artificial intelligence,” said Joe Brusuelas, chief economist at RSM US.

OK, so it’s big. Really big.

There’s nothing inherently wrong with that. If the tech companies are building something customers want, that wouldn’t necessarily create an inflation problem. Supply and demand would be in balance. And if AI yields its promised productivity gains, that should prevent inflation from rising out of control.

But companies are still very much in the “if you build it” phase of this “Field of Dreams” analogy. The promised revolution is still a long way away. The foundation needs to be constructed first.

In the meantime, the massive AI spending supercycle has created a developing economic problem.

“A demand shock is creating inflation,” said Daniel Yue, who teaches about the economic effects of AI at Georgia Tech.

Data center spending is already causing inflation to rise, because of off-the-charts demand for memory and storage chips, building supplies, electricity and laborers – construction workers, plumbers and electricians – to build the data centers. Prices for certain goods and services are going through the roof.

Brandon Bell/Getty Images

Supply constraints, regulations, the immigration crackdown, permitting issues and many other practical realities have prevented all those goods and labor supplies from coming into balance with the extreme demand. And all of that is starting to permeate throughout the economy, showing up particularly in labor shortages and higher costs for building supplies across other industries.

That has raised alarm bells for Austan Goolsbee, president of the Federal Reserve Bank of Chicago.

In a speech in London on September 21, Goolsbee said he is closely watching whether “AI data center construction is spilling out of its own lane” and creating more economic activity than the economy can absorb. If so, it could be a sign that demand is overheating.

“And if demand overheats, there is no ambiguity about how the Fed needs to respond,” Goolsbee said, predicting the central bank would have to further raise its target interest rate to help bring inflation down.

The massive amount of AI spending is coming on top of an already hot economy that never cooled down after its post-pandemic sugar rush.

US manufacturing activity in August was the highest recorded since July 2021, according to a Purchasing Managers’ Index reported by S&P Global last week. Unemployment is at 4.1% , a level economists consider “full employment.” US retail sales surged by 1.2% in August.

Those are not signs of a weak economy.

In large part, the already-hot economy has been fueled by an AI-fueled stock market boom. The eight most valuable stocks in the S&P 500 are all effectively AI companies, and they make up a combined 38.8% of the total value of the stock market.

Portfolio gains for wealthier investors gave permission to the top 40% of earners to keep spending , despite all the cost-of-living concerns Americans in the lower-income thresholds have faced. Those higher earners are responsible for 70% of all consumer spending, according to the New York Fed.

That’s given companies the ability to raise prices, particularly as they contend with record diesel and shipping costs on top of a resurgence in tariffs.

That’s why inflation, which has sat well above the Fed’s 2% target for more than five years, remains a problem that won’t be easily solved, even if the Iran war ended and tariffs went away.

Al Drago/Bloomberg/Getty Images

The Fed is preparing to raise rates. What if it doesn’t work?

“Anything that touches AI is on fire – tech investment, data center construction, even manufacturing of parts to go into these data centers,” said Heather Long, chief economist at Navy Federal Credit Union. “But most of Main Street is just getting the cost increases without much of the financial gain. Rising borrowing costs only compound that feeling of someone is getting rich here, and it’s not me.”

Those rising bond yields and higher target rates from the Fed are increasing the cost of borrowing for businesses, too. That should, in theory, slow down spending and ultimately reduce inflation.

There’s a problem with that premise, though: AI is a runaway freight train.

“To be sure, rising borrowing costs may lead to some second thoughts and cancellations in planned business investment,” said Oren Klachkin, an economist at Nationwide. “But AI investment is rate-insensitive and unlikely to slow down soon.”

布伦丹·麦克德米德/路透社

美国经济可能面临一个新问题：它太强劲了。

美国的生活成本担忧和债券市场崩盘并非经济步履蹒跚、后继乏力的结果，恰恰相反。

物价和利率上涨速度过快令人不安，主要有三个原因：伊朗和乌克兰战争造成的全球能源冲击、不断升级的贸易战以及前所未有的企业在人工智能领域的支出狂潮。

随着汽油和柴油价格飙升，以及美加贸易战愈演愈烈，政策决策（战争和关税）备受关注。但经济学家们越来越担忧大型科技公司投入巨资建设人工智能数据中心的惊人规模。

预计未来几年将在人工智能基础设施上投入数万亿美元，这将从根本上重塑美国经济。然而，所有这些支出都有可能导致经济过热，而目前美国经济已经处于过热状态：失业率低，消费支出强劲。

这只会加剧通货膨胀。而目前的通货膨胀率已经过高了。

人工智能领域的支出热潮在迅速兴起后，其规模之庞大令人难以置信，这一点怎么强调都不为过。

据摩根大通预测，今年人工智能基础设施支出——包括数据中心以及内部的芯片和服务器——预计将达到约1万亿美元。这比美国联邦政府每年的军费开支还要多。

但我们目前还处于人工智能建设的初期阶段。

哥伦比亚大学经济学家斯泰恩·范·尼乌韦尔伯格（Stijn Van Nieuwerburgh）在布鲁金斯学会周三发表的一篇论文中指出，到2032年，这一数字将达到10.3万亿美元。为了便于理解：还记得2021年时任总统乔·拜登签署的1.2万亿美元两党基础设施​​法案吗？该法案后来引发了通货膨胀的担忧。预计人工智能领域的支出相当于将该法案中每年拨出的每一美元都花在十年内。

Benjamin Fanjoy/Getty Images

据高盛预测，仅人工智能基础设施方面的支出（不包括技术本身）今年将占美国经济总量的1.9%。但据范·尼乌沃伯格预测，这一数字预计将翻一番：到2032年，人工智能基础设施支出平均每年将超过美国国内生产总值的3.6%。

这一预测规模庞大，相当于美国经济中一些巨型引擎——例如交通运输、餐饮和酒店——的总产出。但这些都是成熟行业。更令人震惊的是，范·尼乌沃伯格预测，人工智能建设在美国经济总产出中所占的份额将超过美国以往任何一次投资热潮——甚至超过美国当年修建运河、铁路、电网、公路和电信网络的总和。

RSM US首席经济学家乔·布鲁苏拉斯表示：“所有这一切都将导致美国经济的结构性转型，使其围绕人工智能而运转。”

好吧，它很大。真的很大。

这本身并没有什么错。如果科技公司开发的产品正是消费者需要的，那未必会引发通胀问题。供需关系会保持平衡。而且，如果人工智能能够真正实现其承诺的生产力提升，那就应该能够防止通胀失控。

但企业目前仍处于“如果你建造它”的阶段，就像电影《梦幻之地》里的情节一样。人们期待的变革还遥遥无期。首先需要打好基础。

与此同时，人工智能领域的巨额支出超级周期也造成了一个正在发展中的经济问题。

“需求冲击正在造成通货膨胀，”佐治亚理工学院教授人工智能经济影响的丹尼尔·岳说道。

由于对内存和存储芯片、建筑材料、电力以及建造数据中心所需的劳动力（包括建筑工人、水管工和电工）的需求远超预期，数据中心建设支出已导致通货膨胀加剧。某些商品和服务的价格正在飞涨。

Brandon Bell/Getty Images

供应限制、监管、移民管制、许可问题以及许多其他实际因素，导致所有商品和劳动力供应无法与极高的需求达到平衡。所有这些问题正开始渗透到整个经济领域，尤其体现在其他行业的劳动力短缺和建筑材料成本上涨上。

这引起了芝加哥联邦储备银行行长奥斯坦·古尔斯比的警觉。

9月21日，古尔斯比在伦敦发表讲话时表示，他正密切关注“人工智能数据中心建设是否超出其自身范畴”，是否创造了超过经济消化能力的经济活动。如果真是如此，这可能表明需求过热。

“如果需求过热，美联储需要如何应对就毫无疑问了，”古尔斯比说，他预测央行将不得不进一步提高目标利率以帮助降低通胀。

人工智能领域的巨额支出是在一个已经过热的经济环境下进行的，而这个经济在疫情后的繁荣期过后从未降温。

根据标普全球上周发布的采购经理人指数，美国8月份制造业活动创下2021年7月以来的最高水平。失业率为4.1%，经济学家认为这一水平已达到“充分就业”。美国8月份零售销售额增长1.2%。

这些并非经济疲软的迹象。

在很大程度上，本已火热的经济是由人工智能驱动的股市繁荣推动的。标普500指数中市值最高的八只股票实际上都是人工智能公司，它们的总市值占股市总市值的38.8%。

富裕投资者的投资组合收益使得收入最高的40%人群得以继续消费，尽管低收入人群面临着生活成本上涨的种种担忧。据纽约联邦储备银行统计，这些高收入人群的消费支出占总消费支出的70%。

这使得企业有能力提高价格，尤其是在柴油和运输成本创历史新高，以及关税回升的情况下。

这就是为什么通货膨胀（五年多来一直远高于美联储 2% 的目标）仍然是一个难以解决的问题，即使伊朗战争结束，关税取消，这个问题依然存在。

Al Drago/Bloomberg/Getty Images

美联储正准备加息。如果加息失败怎么办？

“凡是跟人工智能沾边的都炙手可热——科技投资、数据中心建设，甚至数据中心零部件的生产，”海军联邦信贷联盟首席经济学家希瑟·朗表示。“但普通民众却只能承受成本上涨，而没有获得多少经济收益。不断上涨的借贷成本只会加剧这种感觉：有人从中发了财，而我却一无所获。”

债券收益率上升和美联储提高目标利率也增加了企业的借贷成本。理论上讲，这应该会减缓支出，并最终降低通货膨胀。

但这个前提存在一个问题：人工智能就像一列失控的货运列车。

“诚然，借贷成本上升可能会导致一些企业重新考虑并取消原定的投资计划，”全国住房金融公司(Nationwide)的经济学家奥伦·克拉奇金(Oren Klachkin)表示，“但人工智能投资对利率并不敏感，而且不太可能很快放缓。”