NextFin News - Artificial intelligence is moving into the inflation debate in a way central banks cannot ignore: not because it instantly makes prices fall, but because it can keep costs and demand uncomfortably high before productivity gains arrive. That is the question now hanging over the Federal Reserve and the Bank of Korea. AI may lower inflation over time by lifting output per worker, but in the near term it can do the opposite by driving capital spending, wages, and infrastructure costs faster than supply can catch up. The result is a policy problem, not just a technology story.
The Federal Reserve’s latest Beige Book said economic activity increased at a slight to moderate pace in 11 of 12 districts, with several districts noting growth in data center building. That matters because data centers are not an abstract software trend. They require land, power, steel, chips, cooling systems, engineers, and construction labor. The Bank of Korea is seeing a similar macro shape from a different angle. South Korea’s government has lifted its 2026 growth forecast to 3.0%, the strongest pace since 2021, and raised its inflation forecast to 2.6% from 2.1% as the semiconductor and AI investment cycle feeds through the economy. The common thread is that AI is now showing up in real activity, real bottlenecks, and real price pressure.
That combination matters because inflation is not only a demand problem. It is a speed problem. When investment rushes into a narrow set of inputs - chips, electricians, transformers, power capacity, and advanced manufacturing space - prices can rise before the promised productivity gains appear. The market usually treats AI as a future efficiency gain. Central banks have to treat it as a current allocation shock. That difference explains why the same technology can look disinflationary in theory and inflationary in practice.
For the Fed, the relevant question is whether AI is broadening supply enough to offset the extra spending it is creating. For the Bank of Korea, the question is sharper: can the semiconductor-led AI cycle sustain growth without keeping inflation sticky above target? The answer is not the same in both economies. Korea’s industrial structure makes it more exposed to an export-led AI boom, while the United States feels the effect more through investment, labor demand, and the service side of the economy. But the mechanism is similar. AI boosts activity first, and only later, if diffusion is broad enough, does it relieve price pressure.
Why AI Can Push Inflation Up Before It Pushes It Down
The first-order story is simple. AI raises productivity if it allows firms to produce more with less labor and less waste. The second-order story is more important for policymakers. In the early phase of the cycle, firms do not buy AI because they want lower inflation. They buy it because competitors are buying it, because clients expect it, and because they fear falling behind. That triggers a wave of capital expenditure. It also pushes up demand for scarce inputs. If those inputs are constrained, the economy sees more spending before it sees more supply.
That is why the inflation effect can come before the productivity effect. The buildout of chips and data centers consumes steel, cement, power equipment, networking hardware, and specialist labor. In the short run, those inputs are not infinitely elastic. The result is a familiar bottleneck pattern: higher capex raises activity, but it also bids up the cost of the very inputs needed to build the new productive capacity. That is especially visible when the technology requires a physical footprint. AI is digital only at the application layer; underneath, it is heavy industry.
The Fed’s Beige Book gives a concrete sign that this is no longer a theoretical concern. A modestly firmer economy with data center construction expanding is exactly the kind of setting where inflation can stay sticky even if consumer demand is not overheating. The mechanism is indirect. Construction raises employment and wages in local markets. Power demand can lift utility and infrastructure costs. Financing conditions can tighten if corporate spending remains strong. Those are not the classic signs of an old-fashioned demand boom, but they still matter for the inflation path.
This is why the AI story looks cyclical at first and potentially structural later. The cyclical part is the investment wave itself. Every capex cycle eventually runs into capacity constraints, and those constraints usually ease once supply adjusts. That is a mean-reverting process. But if AI permanently changes how much investment the economy needs to generate a given amount of output, then the inflation process also changes. The structural shift would come not from the first wave of spending, but from the new baseline of higher digital and physical capital intensity.
History argues for caution before calling it structural. Electrification, railroads, and the internet all produced inflationary buildouts before they produced large productivity gains. The early phase was often marked by shortages and rising costs. Only later did the economy reap the broader efficiency dividend. That pattern does not prove AI will behave the same way, but it is enough to say the near-term inflation impulse is likely cyclical even if the long-run productivity effect is structural.
That distinction matters because central banks set policy on current inflation, not on hoped-for productivity. If the economy is spending aggressively on AI infrastructure now, the Fed and the Bank of Korea have to lean against the demand side even if they believe the supply side will eventually improve. Policy has to respond to the transition, and transitions are where inflation often lives.
Why Korea Feels The Pressure Faster Than The U.S.
The Bank of Korea is closer to the AI inflation channel because South Korea sits at the center of the semiconductor cycle. When chip demand surges, Korean exports, corporate earnings, income, and domestic investment all feel it more quickly. That can lift growth at the same time it keeps inflation from cooling. The government’s 2026 growth forecast of 3.0% and inflation forecast of 2.6% show that the domestic economy is absorbing the AI boom as both a growth dividend and a price risk.
The United States is less concentrated, but it is not immune. The Fed’s Beige Book points to data center construction as one of the notable sources of activity. That means the inflation transmission is less about a single export industry and more about broad-based capital formation. The U.S. can diffuse the shock across more sectors, which makes the effect slower and more varied. But the same physical constraints apply: land, electricity, construction crews, and equipment are all finite in the short run.
The policy difference between the two central banks is therefore one of speed, not principle. Korea’s narrower industrial base makes the inflation impact more visible faster. The U.S. sees it through a wider lens, which makes it easier to miss until the bottlenecks show up in the data. In both cases, the AI boom can keep nominal growth firmer than conventional models expect.
The Federal Reserve said in its latest Beige Book that economic activity increased at a slight to moderate pace in 11 of 12 districts, with several districts noting growth in data center building.
That line is important because it tells you where the mechanism begins. The inflation question is not whether AI software itself is expensive. It is whether the physical economy needed to build and run the software is expensive enough to matter. If the answer is yes, then AI can keep services and construction inflation from easing as quickly as headline models assume.
The second-order implication is even more important. Once firms and households see a wave of AI spending, they may revise expectations for wages, rent, energy, and financing costs. Expectations can become self-reinforcing. Firms set prices with an eye to future input costs, workers bargain for higher pay if they think demand is running hot, and bond markets demand compensation for sticky inflation risk. The effect is not just on current prices. It is on how quickly prices move in the first place.
The Strongest Case Against The Inflationary Reading
The best argument against this view is that AI should ultimately lower inflation by raising productivity. That argument is not weak. It is the central structural bull case for the technology. If firms can automate routine work, improve forecasting, reduce error rates, and cut waste, then each unit of output should require less labor and less overhead. Over time, that should compress unit costs and ease price pressure. The longer the technology diffuses across the economy, the stronger that effect should become.
The problem is timing. Central banks cannot price the economy that exists in five years. They have to respond to the economy that exists now. If the early stage of AI adoption requires a large burst of investment, the near-term inflation effect can dominate even if the long-term productivity effect is real. That is particularly true when the gains are concentrated in a few large firms or sectors while the costs are spread across the wider economy. The broad price level feels the costs first.
There is also a market test embedded in this debate. If AI were already broadly disinflationary, the most visible evidence would be easing price pressure in the data and a clearer move lower in inflation-sensitive rates. Instead, the policy conversation is still about whether AI is adding to bottlenecks and demand. That does not prove the inflation case, but it does show that the burden of proof has not shifted fully to the disinflation camp.
The strongest falsifying signal for the inflationary thesis is a sustained and measurable easing in core price pressure while productivity gains accelerate. If U.S. core PCE were to slow to a clearly subtrend pace for several consecutive prints at the same time that productivity growth strengthened materially, the case that AI is keeping inflation sticky would weaken. In Korea, a similar falsification would be a return of inflation toward target while AI-linked investment remains strong. That would indicate productivity is arriving quickly enough to offset the demand and bottleneck effects.
Until that happens, the disinflation story remains a future promise, not a present fact. AI may eventually make the economy cheaper to run. Right now, it is making the economy more expensive to build.
What To Watch Next
In the short term, the beneficiaries are the firms supplying the buildout: chipmakers, data-center operators, power equipment producers, grid builders, and construction-related industries. The exposed groups are the rate setters and price takers - central banks, utilities, labor-intensive service sectors, and any economy that is already close to capacity. If AI spending keeps lifting wages and input prices before productivity gains show up, those are the places where inflation pressure will appear first.
In the medium term, the key test is whether productivity begins to outpace the investment burst. If it does, the inflation impulse should fade, and central banks may find they can tolerate stronger growth without tightening as aggressively. If it does not, both the Fed and the Bank of Korea may have to treat AI as a reason inflation stays above target for longer than models suggest.
In the long term, the question is structural. If AI permanently lifts trend productivity, it can lower the economy’s inflation ceiling. But structural change does not arrive cleanly. It comes through a cyclical phase of bottlenecks, wage pressure, and capital scarcity. That is why today’s inflation debate is really a debate about transition speed. The faster AI spreads, the sooner the supply benefit shows up. The slower it spreads, the more the inflation cost dominates.
The base case is that AI remains mildly inflationary in the near term, especially in economies where the buildout is concentrated, and only gradually turns disinflationary as productivity gains diffuse. The upside case for disinflation is that adoption broadens quickly enough to offset higher capex and wage demand. The downside case is that the physical buildout keeps running into scarce labor and power capacity, forcing central banks to keep policy tighter for longer.
For the Fed, the critical signals are the Beige Book, core inflation prints, and productivity data. For the Bank of Korea, the key markers are inflation, exports, and policy guidance tied to the semiconductor cycle. If inflation cools only after AI investment slows, the cyclical view wins. If inflation stays elevated even as productivity rises, the structural case gets stronger.
AI is not a clean disinflation machine. It is a powerful economic force that first makes the supply side harder to read. That is why central banks are treating it less like a theme and more like a risk factor.
The market may still want a simple story. The inflation data is likely to deliver a messier one.
Explore more exclusive insights at nextfin.ai.
