How is AI influencing interest rates? Investment, productivity, prices, and more

Like many people, monetary policymakers are pondering what artificial intelligence means for them. The minutes of the June meeting of the Fed’s rate-setting body mention AI more than 20 times in the context of stock values, business investment, bank lending, trade flows, and more. References at the June meeting last year? One.

When it comes to the Fed’s dual mandate of inflation and employment, one certain thing about AI is that it makes things uncertain. “The staff continued to view the uncertainty around their forecast as elevated,” the minutes state, “importantly because of uncertainty about … the potential economic effects of AI investment and adoption.”

The aggregate impact of generative AI—AI that uses its training to generate new content—could be the macroeconomic question of our era. A few economists are dabbling in the fantastical future, adapting economic models to explore annual GDP growth topping 70 percent or a bizarre, post-employment economy. Most prefer to focus on the here-and-now: Is AI-driven productivity growth slowing the rate of inflation? How does this stack up against the multiple AI-driven factors pushing interest rates higher? Are households adapting and planning?

Here’s a breakdown of the economic forces to watch, alongside the insights of two Federal Reserve Bank of Minneapolis economists who are tracking this crucial policy issue. The bottom line: AI is moderately heating up today’s economy while we wait for the likely bumpy rollout of productivity gains.

Investment: The dominant—but not yet dominating—effect

From $200 billion in 2024, capital spending by the five largest investors in AI data centers—Alphabet, Amazon, Meta, Microsoft, and Oracle—is projected to approach $1 trillion by 2027. “For reference, total private investment in the economy is about $5.5 trillion dollars,” said Minneapolis Fed Monetary Advisor Alisdair McKay. “We’re talking about 20 percent of investment coming from this one category.”

All else equal, this surge in data center spending and demand for investment funding would constitute strong macroeconomic forces pushing real interest rates higher.¹ But for all the lofty projections, AI-related investment is not moving the needle much at an economy-wide level.

Despite an unmistakable leap in an AI-relevant category like information processing equipment (Figure 1, right axis), the growth path of U.S. aggregate private investment looks similar to the trend since 2010 (Figure 1, left axis).

As a proportion of U.S. GDP, McKay notes that private investment remains roughly flat since 2018. So far, the AI boom does not resemble prior periods of investment growth in the 1990s and 2010s. Breaking investment down into its four primary components shows part of the reason: The shares of housing investment—and, to a lesser extent, investment in nonresidential construction—are falling (Figure 2).

Current nominal interest rates are elevated from the ongoing battle to vanquish inflation, which depresses or postpones investment in those rate-sensitive construction categories. Intense investment demand for AI data centers also drives up costs for construction inputs and could attract funds that might otherwise go into housing. The net result is something of a wash from a macroeconomic perspective. “You would think that if you have this great opportunity” to achieve future growth by investing in AI, McKay said, “you would increase the amount you invest. But we have not done that in the aggregate.”

Minneapolis Fed Monetary Advisor and Assistant Director of Policy Cristina Arellano notes that while the technology sector represents about 15 percent of U.S. output, it comprises only 7 percent of U.S. consumption. As many economists understand it, this investment-driven economic growth puts less pressure on underlying interest rates than if spending were driven by a more consumption-heavy category.

“The natural rate [of interest] is more linked to the consumption growth rate, in terms of the frameworks we use to think about this,” Arellano said. Growth focused in the tech sector “may have a smaller effect on the natural rate because it’s not affecting consumption so much.”

AI “definitely is, in the short and medium run, a force that increases both natural rates and potentially price pressures,” Arellano said. But other shifting pieces of the U.S. economy appear to be significantly offsetting the effect of AI investment, for now. If accelerating AI investment were to outpace the residential slowdown—or if rates were to fall and residential investment rebound—spiking aggregate investment would mean strong demand and even more upward pressure on rates.

Household consumption: Optimism, anxiety, and stock wealth

In many economic models, household and investor expectations of the future can make a big difference today. If people expect to be richer down the road, they spend more today and even borrow against that future income. This would tend to increase real interest rates as the supply of savings shrinks, especially in the context of high investment demand.

If, on the other hand, people worry about their jobs or the prospects for their children, they might tend to save more as a precautionary measure, having the opposite economic effect. Pessimism about the future tends to keep rates in check today.

This classic “consumption smoothing” dynamic comes up often in speeches and papers about the macroeconomic impact of AI. Researchers analyzing significant movements of bond yields around major AI announcements interpret them under this theory.

For all the utopian-to-existential talk about AI around American dinner tables, Arellano and McKay are skeptical that households are behaving like the economic models. “I don’t think there are that many people who connect that future—where all of us, where the economy is richer—with, ‘I’m going to be richer,’” said McKay.

As for the scenario of fear-based savings pushing rates down, Americans are showing no evidence of precautionary saving. The U.S. personal saving rate has been generally falling since AI hit the public consciousness and sits now near historically low levels.

However, beliefs about the future of AI are likely influencing current consumption and interest rates through another channel: Soaring stock wealth. Since ChatGPT debuted to the general public in November 2022, the S&P 500 stock index has risen 80 percent (as of late July 2026), driven by shares of tech companies associated with AI. “We think that the marginal propensity to consume out of stock wealth is about 3 cents on the dollar,” said McKay. “So that would mean, ballpark, one-half to 1 percent of GDP in consumption each year from this extra wealth. That’s pretty big.”

The wealthiest 10 percent of U.S. households own almost 90 percent of American stock and mutual fund holdings; the richest households also account for a disproportionate amount of spending. Their consumption, supported by these equity gains, has helped sustain demand despite low sentiment among consumers overall.

On the heels of attending a conference in Europe, Arellano contrasts European pessimism about the economy with greater optimism here, which could motivate relatively stronger U.S. consumption from AI-related wealth. This demand keeps the economy and inflation running hotter, an argument for higher policy rates.

Aggregate consumer demand is held somewhat in check, however, by the concentrated nature of AI-based wealth and by caution among consumers with lower wealth and income. AI-inspired spending is “not for every segment,” Arellano said, “especially for young people graduating from college.”

Prices: Percolating cost pressures and “algorithmic pricing”

Some products are affected directly by the voracious demand of AI data centers for the same inputs. In June, Apple announced price increases across its computers and tablets, including a 25 percent hike on the iPad Air. Nintendo is raising the price of its popular Switch 2 gaming console, citing, like Apple, the rising cost of memory and storage. Steel makers blame AI data center operations for driving up their cost of electricity, even as data center construction pushes up the price of steel from the demand side.

AI-relevant categories are running ahead of their broader consumer and producer inflation indexes (Figure 3).

Relative price changes happen all the time; they do not necessarily portend general inflation. But with core consumer inflation persistently above the Fed’s 2 percent annual target, policymakers might take note of categories where price increases are not only above historic averages but accelerating—as is the case for computer-related equipment. Rising prices for metals, power, and information technology could pass more widely into the business costs for firms.

Policymakers generally “look through” supply shocks expected to temporarily affect relative prices, such as a one-time increase in tariffs or the war in Iran. “The AI impact seems like it could be more persistent,” said McKay, with data center investment possibly reaching into trillions of dollars and stretching years into the future. If so, this might incline policymakers toward higher interest rates to contain wider price increases and keep inflation expectations anchored.

Another effect to ponder: AI tools could bring a leap forward in helping companies adjust prices more frequently and precisely. AI could turbocharge what economists call “price discrimination”—think of it as personalized pricing—“by facilitating the real-time analysis of consumer demand and price elasticities,” wrote European Central Bank official Piero Cippolone, who says the change would not tend to benefit consumers. “Algorithms consistently learn to charge collusive prices that are higher than competitive ones.”

The advent of such “surveillance pricing” would primarily affect the price level, not necessarily the ongoing rate of change (that is, inflation). But a world of instantaneous price adjustments and pass-through of costs could amplify inflationary events. It could also make central banks’ jobs more difficult. “Those frictions shape the transmission of monetary policy,” said Arellano. What economists call “nominal rigidities” of prices (and wages) are understood to play a crucial role in translating the Fed’s policy moves into reactions across the economy. While AI-driven, algorithmic pricing might not be—or remain—strictly legal, a recent report from the Federal Reserve Bank of Kansas City finds evidence such pricing practices are already spreading across many sectors of the economy.

Against all this, there is a prominent counterargument that AI will restrain price increases or even drive many prices down. Recent findings by European researchers found that a higher share of AI adoption by firms corresponded with lower inflation in those sectors, with the productivity gains from AI a possible “structural force dampening inflation.” The notion surfaces at each Fed policy meeting in 2026, including, per the June minutes, “Some participants remarked that productivity gains associated with AI adoption would eventually reduce production costs and increase aggregate supply, which should put downward pressure on inflation, though they noted this effect would likely take time to materialize.”

Indeed, these effects are not yet meaningfully apparent at a macroeconomic level, where headline and core price indexes remain elevated. Nor are they evident for the task where AI has been most immediately and heavily put into action: computer coding. The consumer and producer price indexes for software, historically deflationary, instead show flat-to-rising prices since generative AI came on the scene (Figure 4). Importantly, these measures also reflect AI-driven hardware price pressures.² Nonetheless, they display no ground-level signal of productivity leading to disinflation—quite the opposite.

But AI use is spreading rapidly across companies, with nearly 90 percent telling McKinsey they were already using it in 2025. “Let’s say in the next five years we start to have these productivity gains in different sectors,” said Arellano. “That could have effects that are disinflationary in the medium term.”

Everything depends on the pace, scale, and nature of productivity gains from AI.

Productivity: Uncertain timing and direction of overall effect

There are, broadly speaking, two stories about the relationship between AI-driven productivity growth and interest rates.

In the story that leads to lower rates, AI constitutes a huge positive supply shock, allowing companies to do much more with the same resources. Businesses could raise output and profit margins without stoking inflation or courting labor shortages. In some variations, AI primarily replaces workers, leading to a “jobless boom” or growing long-term unemployment that could depress consumption and put further pressure on the employment side of the Fed’s mandate. In any case, the productivity effect of AI is strong enough that wage gains do not pass through to wider price pressures.

A more conventional economic story sees these supply-side, disinflationary effects of productivity outweighed by the accompanying demand-side pressures of economic growth, pushing in the opposite direction. Ongoing investment and rising consumption would mean higher real interest rates. If there is a “goldilocks” outcome of falling prices and rising living standards, this view sees it likely far in the future, beyond a rocky period of adjustment where most forces point to higher rates.

We might not know for decades which story prevails in the long run, where additional factors include the effects of AI on our lifespans and the burden of the national debt. For policymakers today, the debate is largely academic until productivity gains start to appear in the macroeconomic data.

We might expect to see these productivity gains first in the businesses where AI is most heavily and immediately implemented. For the information services sector and the subsector that includes software, rough calculations of productivity through 2025 (real output-per-worker) are consistent with levels over the past 20 years (Figure 5). A large jump in 2023 coincides with the introduction of ChatGPT. But that is followed by four quarters of productivity declines. Rising productivity through 2025 is tantalizing, but within normal, historic fluctuations.

Even rocketing productivity growth in the information services sector—about 6 percent of GDP—might have little effect on productivity across the economy. Three alternate measures of overall productivity through early 2026 do not show a budding productivity boom (Figure 6). Recent readings are lower than prior periods of sustained high growth, like the mid-2000s. Arguably the most refined of these measures, utilization-adjusted productivity from the Federal Reserve Bank of San Francisco, is trending down since 2024.

“If you look back at other types of technological adoption—electricity, computers—they took decades to manifest in terms of changing production processes and getting the productivity benefits,” said McKay. “From a macro perspective, the main thing we are doing right now is ‘building the machine.’”

McKay points out a disconnect between the tasks most likely to benefit initially from AI and the physical investment that puts demand-side pressure on the economy. “I don’t see that the productivity benefits are going to show up in a way that makes it easier to build a data center,” he said.

Researchers have documented a J-shaped pattern to the adoption of general-purpose technologies. Measured productivity actually decreases at first, as companies implement investments in retraining, reorganization, and updating equipment. Although AI awareness and experimentation are arguably spreading faster across the economy than prior technologies, similar frictions are already appearing. These could be compounded by regulatory, cultural, and natural resource constraints that crop up on the path to transformative AI.

Other factors: Many manual jobs might see little ultimate impact from AI. And productivity changes from new technology can cut both ways: Economist Daron Acemoglu suggests that “bad new tasks” supercharged by AI—deepfakes, fraud, cyberattacks—could do “sizable” societal harm alongside productivity gains.

Focus on the present, focus on the aggregate

Price stability is in the foreground for many of the above factors. But it is also essential to note that there are no clear or dramatic signals from the employment side of the Fed’s mandate, AI or otherwise. The unemployment rate has held steady amid a persistent “low-hire, low-fire” labor market. AI has been cited for some notable layoffs in the tech sector, even as a broader study of 21,000 U.S. firms found AI adoption is associated with additional hiring. Big developments in the labor market would naturally shift the balance of factors for policy.

Minneapolis Fed economists McKay and Arellano stress that monetary policy has little influence over structural changes wrought by technology. Nor do central banks tend to respond to scenarios in the uncertain future. Fed policymakers are focused on today’s data, where productivity gains are a work in progress and jobs are holding steady. Data center investment, wealth-driven consumption, and possibly pricing forces are creating heat—but this appears moderate in the aggregate, for now.

“In the near term, whatever increase in the productive capacity of the economy AI has brought, AI has brought a larger increase in demand,” McKay said. Big changes could still be on the way for the economy, maybe sooner than later given the speed of AI investment, awareness, and diffusion. But we are still near the start of the journey.

“It’s hard to just implement things really quickly, then adapt and change,” said Arellano. “These processes are sort of slow. But I do think there will be a lot of gains in the medium term.”

Endnotes

1 This article primarily describes economic pressures—affecting real interest rates and inflation—with which Fed policymakers contend when deciding how to pursue the Fed’s dual mandate of full employment and low-and-stable inflation. The nominal interest rates that prevail in the economy for savings and lending transactions are a result of Fed policy actions interacting with these pressures.

2 As a general matter, software producers must spend on hardware and on “tokens” to the extent they employ AI tools. Further, it is essential to note that the consumer “software” categories of the CPI and PCEPI (which currently uses CPI data) include portable blank storage media (such as flash drives). These have jumped in price because of input competition from the AI buildout. Fed economists estimated that storage media may have accounted for about a quarter of the rise in the PCEPI software category from November 2025 to March 2026. For a clearer view of software prices, the BEA will eliminate storage media from its methodology starting with August 2026 data, retroactive five years.