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

Al Drago/Bloomberg/Getty Images

The Federal Reserve has spent years holding off on higher interest rates, wary of doing unnecessary damage to the US economy. Now, as it prepares to raise rates this week, officials are facing a new question: Will they need multiple rate hikes to bring inflation down?

The monthslong conflict in the Middle East has pushed inflation higher this year, and the risk of those price pressures spreading across the economy is one big reason why the Federal Reserve is expected to raise interest rates this week for the first time since July 2023.

But Fed officials are increasingly worried about another inflation threat, one that could prove much harder to tame: the massive build-out of AI infrastructure.

According to the official minutes from the Fed’s policy meeting in July, “several” members of the rate-setting committee said the huge investments in data centers could have “broader effects on prices by pushing up aggregate demand.”

That’s because data center construction is fueling huge spending on everything from semiconductor chips and power to skilled workers. The multibillion-dollar companies driving the boom are aggressively competing for limited resources. And if that demand keeps outstripping supply, it could set the stage for higher inflation for longer.

Trump Goes All in on Data Centers: “Let Data Reign” | The Source with Kaitlan Collins | CNN Audio

New York Fed President John Williams, the vice chair of the central bank’s rate-setting committee and a permanent voting member, identified that as his primary inflation concern.

All of this creates a particular problem for the Fed. AI-driven demand seems to be so strong that even the three quarter-point hikes that Wall Street is expecting in the next few months may not be enough to meaningfully slow the spending boom.

“If you’re a company that’s a technological service provider or a chip maker, your goal is to capture as much market share as you possibly can at the very early stages of this new technological development,” said Jim Caron, chief investment officer of portfolio solutions at Morgan Stanley Investment Management. “The longer you wait, the harder it’ll be to get more ingrained in the field.”

Brandon Bell/Getty Images

The scale of the spending is enormous.

According to a report from PricewaterhouseCoopers released earlier this month, spending on data centers is running at an estimated $800 billion and is expected to swell to $1.1 trillion by 2030. Another report from global research firm Gartner estimated early this year that spending on data centers will reach $1.37 trillion in 2026, on top of billions more spent on software, services, cybersecurity and model development, equaling $2.52 trillion this year alone.

“The demand is so insatiable that these companies, these hyperscalers, will pay almost any price for those inputs, and they need things built yesterday,” Cleveland Fed President Beth Hammack told Yahoo Finance in a July interview. Minneapolis Fed President Neel Kashkari said in a July statement that the “massive investment in data centers has also added a new demand element to the high inflation Americans are experiencing.”

Hammack, Kashkari and Dallas Fed President Lorie Logan dissented from the central bank’s July decision to hold rates steady, instead backing a rate hike. They’re expected to vote for a rate hike this week.

The Fed’s main tool — its key interest rate — functions on the demand side, either by cooling or stimulating the US economy, depending on whether officials want to address high inflation or high unemployment.

But that mechanism is less powerful when facing an AI spending spree. In theory, higher interest rates should cool the economy by reducing spending and bringing down prices. But that’s harder to pull off when companies have so much money to spend and are racing to build out infrastructure.

“Hyperscalers have balance sheets that are in a very strong position,” said Ian Kresnak, senior investment strategist at Vanguard, referring to tech companies Amazon, Microsoft, Google, Meta and Oracle.

“They see this as all about building the future for this technology, so will a marginal increase in interest rates really change that dynamic?” he added.

Further complicating the picture, the hyperscalers are increasingly turning to the bond market to help finance their enormous costs.

US 10-year yield touches highest level since 2023

“Everybody is monitoring these big hyperscalers more closely because they went from companies that had tons of cash to now taking up so much money on the credit market,” said Bjoern Griesbach, head of macroeconomics and capital markets research at Allianz Trade.

If investors start to question whether all that spending will generate enough profit, and if they become less willing to finance it, they could demand higher returns on their loans, adding even more to the costs.

“The million-dollar question is whether all these investments in data centers, chips and software will pay off,” Griesbach said.

That leaves the Fed with a difficult task: While it can raise rates to cool the economy, that may not be enough to slow the AI spending boom, and could even weigh on other parts of the economy, such as the labor market.

Al Drago/Bloomberg/Getty Images

美联储多年来一直避免提高利率，担心会对美国经济造成不必要的损害。如今，随着本周即将加息，官员们面临着一个新的问题：他们是否需要多次加息才能降低通胀？

中东持续数月的冲突推高了今年的通货膨胀，而这些价格压力蔓延至整个经济的风险，正是美联储预计将在本周自 2023 年 7 月以来首次加息的一个重要原因。

但美联储官员越来越担心另一种通胀威胁，这种威胁可能更难控制：人工智能基础设施的大规模建设。

根据美联储 7 月份政策会议的官方记录，“几位”利率制定委员会成员表示，对数据中心的巨额投资可能会“通过推高总需求，对价格产生更广泛的影响”。

这是因为数据中心建设带动了从半导体芯片、电力到技术工人等各个领域的巨额支出。推动这波繁荣的数十亿美元公司正在激烈争夺有限的资源。如果这种需求持续超过供应，可能会导致高通胀持续更长时间。

特朗普全力投入数据中心：“让数据主宰一切” | Kaitlan Collins 的《The Source》节目 | CNN 音频

纽约联邦储备银行行长约翰·威廉姆斯是该央行利率制定委员会的副主席和常任投票成员，他将此视为自己最主要的通胀担忧。

所有这些都给美联储带来了一个特殊的问题。人工智能驱动的需求似乎如此强劲，以至于即使华尔街预期美联储在未来几个月内加息三次（每次加息0.25个百分点），也可能不足以有效减缓消费热潮。

摩根士丹利投资管理公司投资组合解决方案首席投资官吉姆·卡伦表示：“如果你是一家技术服务提供商或芯片制造商，你的目标是在这项新技术发展的早期阶段尽可能多地抢占市场份额。等待的时间越长，就越难在这个领域站稳脚跟。”

Brandon Bell/Getty Images

支出规模极其庞大。

根据普华永道本月早些时候发布的一份报告，数据中心支出估计为8000亿美元，预计到2030年将飙升至1.1万亿美元。全球研究公司Gartner今年年初发布的另一份报告估计，到2026年，数据中心支出将达到1.37万亿美元，此外，软件、服务、网络安全和模型开发方面的支出也高达数十亿美元，仅今年一年就达到了2.52万亿美元。

克利夫兰联储主席贝丝·哈马克在7月接受雅虎财经采访时表示：“需求如此旺盛，这些公司，这些超大规模数据中心运营商，几乎愿意为这些资源支付任何价格，而且他们需要的是昨天就建成的东西。”明尼阿波利斯联储主席尼尔·卡什卡利在7月的一份声明中表示，“对数据中心的大规模投资也为美国人正在经历的高通胀增添了新的需求因素。”

哈马克、卡什卡利和达拉斯联储主席洛里·洛根反对美联储7月份维持利率不变的决定，而是支持加息。预计他们本周将投票支持加息。

美联储的主要工具——关键利率——作用于需求侧，通过冷却或刺激美国经济来发挥作用，这取决于官员们是想解决高通胀还是高失业率问题。

但面对人工智能领域的消费狂潮，这种机制的作用就显得不那么显著了。理论上，更高的利率应该能通过减少支出和降低物价来抑制经济过热。但当企业拥有大量资金并竞相建设基础设施时，这种抑制作用就难以实现。

“超大规模数据中心运营商的资产负债表状况非常稳健，”先锋集团高级投资策略师伊恩·克雷斯纳克 (Ian Kresnak) 表示，他指的是亚马逊、微软、谷歌、Meta 和甲骨文等科技公司。

“他们认为这一切都是为了这项技术的未来发展，那么利率的小幅上升真的会改变这种局面吗？”他补充道。

更复杂的是，超大规模数据中心运营商越来越多地转向债券市场来帮助其筹集巨额资金。

美国10年期国债收益率触及2023年以来最高水平

安联交易宏观经济和资本市场研究主管比约恩·格里斯巴赫表示：“大家都在更加密切地关注这些大型超大规模企业，因为它们从拥有大量现金的公司变成了现在在信贷市场上占用大量资金的公司。”

如果投资者开始质疑所有这些支出是否会产生足够的利润，如果他们不愿意再为之融资，他们可能会要求更高的贷款回报，从而进一步增加成本。

格里斯巴赫说：“最关键的问题是，所有这些对数据中心、芯片和软件的投资是否都会获得回报。”

这使得美联储面临一项艰巨的任务：虽然它可以提高利率来给经济降温，但这可能不足以减缓人工智能支出热潮，甚至可能对经济的其他部分（如劳动力市场）造成压力。