芝加哥聯邦儲備銀行行長 Austan Goolsbee 稱 AI 熱潮正擴大通脹壓力

The News

Inflation tied to the AI boom is “jumping out of its lane” and starting to push up prices in unrelated corners of the economy, Chicago Fed President Austan Goolsbee said.

“Prices are going up in spaces that are not just components in data centers and the directly adjacent industries,” he said on Semafor’s . “I’m getting increasingly nervous that that’s just straight, old-fashioned overheating demand.”

Prices for land, electricity, water, construction, and electrical components used in data centers have surged. Those spikes don’t appear to have yet hit consumer pain points — for example, the cost of groceries or a haircut — but economists are closely watching how AI’s wealth, from paper millionaires in Silicon Valley to higher-paid construction workers, is being spent.

Fed Governor Lisa Cook warned last month about “some economy-wide pressure from AI-fueled demand” but said that so far the boom looks like “a shift in demand toward AI-related sectors rather than an increase in economy-wide demand.” Fighting that kind of sector-specific inflation with rate hikes “could be a mistake,” she said.

Goolsbee isn’t a member of the Fed committee that sets interest rates, but will join it in 2027. He and other economists have been trying to parse how much of today’s stubborn inflation — 3.4% in August, with September’s numbers due out this week — is from higher oil prices triggered by the war in the Middle East, which the Fed can’t do much about, and how much is coming from hotter demand across the economy for goods and services.

“If inflation remains persistent, then I think it’s clear what the Fed has to do,” he said, when asked whether the central bank should raise interest rates again when it meets later this month.

Liz’s view

The Fed’s problem is that its best tool for fighting inflation doesn’t seem to be denting the demand for AI inputs. Hyperscalers aren’t slaphappy consumers who might be deterred by higher interest rates. They are locked in a must-win battle and don’t seem to mind higher debt costs. Companies like Alphabet and Meta have cash-cow advertising businesses that can support a few extra basis points of interest.

It’s a broader problem for the Fed: Big parts of the US economy no longer respond to rising interest rates. There may be a pain point for the hyperscalers, but it’s likely to be intolerable for the rest of us.

Room for Disagreement

The issuance of new AI-linked debt fell in August, according to Morgan Stanley, which could show investors tapping on the brakes, even if corporate executives don’t want to.

Transcript

Liz Hoffman: Austan Goolsbee, welcome to the show.

Austan Goolsbee: Thank you for having me.

Liz: You are a great communicator in a world of wonks, literally a champion talker. Kevin Warsh wants the Fed to talk less and to force investors to look at the numbers and make their own decisions. What do you think?

I’m an old friend and fan of Chairman Warsh. I worked with him back in the financial crisis. Before he ever got there, I’m uneasy with people expressing what we’re now calling forward guidance, that is saying, “Here’s what I think will happen at the next meeting. Here’s what I think four months from now the interest rate should change.” Most of the discussion about communication seems to have been about that, and I kind of agree with we should phase out that kind of communication. I’m engaged in a demystification project about the Federal Reserve. Here’s what we do. We’re not the bad guys. We’re the Guardians of the Galaxy. I’m going to stay on that for sure.

Liz: Getting to those numbers that you would all like the market to put more stock in and make their own decisions about, even if they were prepared to be weaned off the commentary, does the data that you’re seeing right now lay out a clear picture?

I like how you phrased that. I would say no, not a super clear picture because inflation’s still way too high. I have been worrying publicly about the potential for supply shocks like tariffs, like oil prices to shove us into a stagflationary kind of a trade-off. There is one sense in which it hasn’t been ambiguous. It hasn’t been stagflationary. The job market still looks very stable to me. Growth has been solid, consumer spending solid. We’re falling down on the inflation side of the mandate. But the ambiguity comes, is the inflation getting better on its own as it might if a supply shock kind of worked through or if you had a tariff that was just a one-time increase in the level of prices that should fade away and it hasn’t really been fading away. And you’ve seen inflation in services kicking up and that’s not really from tariffs, that’s not really mostly from oil. So that’s where you start to get the ambiguity in it and it’s not totally clear.

Rohan Goswami: How much of that is AI, which obviously it’s driving a huge investment boom that’s sending up the price of inputs, power, land up. But at the same time, there is this deflationary promise that it’ll make everything cheaper. How do you think about those two things?

This dynamic where there is great promise to come that is factored into massive stock market valuations and massive incentives for companies to engage in business investment. All of those things, the bigger the hype, the greater the chance that you overheat the economy in the short run, that for the day job of the Fed, the more promise there is, the more temptation, the more tendency to spend today based on the promise of what is to come. And that you got to be careful about because that’s just old-fashioned overheating right now.

Rohan: What would a financial unwind of that trade, of the AI trade actually look like from where you sit?

If it’s a discovery that the fundamentals are not there and it’s a bubble, that we have had bubbles in US history in the past, including in the recent past, and their unwinds can get ugly depending how much leverage is associated with those bubbles. It could be more like the popping of the internet. Now, Rohan, you’re not old enough. You don’t remember it, but if you back up, if you’re older than dirt like me, you’ll remember we had an experience with the internet that they were building out the infrastructure premised on growth rates of 20% per year. And when the growth rate came in, well, it’s only going to be 7% per year, that’s still a pretty strong growth rate, but they were massively overbuilt. And so we kind of had this overhang. That’s slightly different than... Finding out the fundamentals are different is slightly different than just a old-fashioned bubble in my view.

Liz: But Alan Greenspan, I think, said basically that the Fed cannot deflate a bubble in real time without making life miserable for everybody else. Do you think that’s still true?

There is a very significant debate among economists about the question, should the Fed try to prevent bubbles, pop bubbles, just clean up after the bubble pops? All of those things. Anybody citing the 1990s and Alan Greenspan, it’s worth at least just going back to give yourself a refresher on the history because it’s not near as straightforward as how a lot of people want to remember it. It is true that in the mid ’90s, Alan Greenspan identified that he thought productivity growth must be accelerating, but people didn’t know that. It wasn’t showing up in the data yet when he said that. He concluded because he said, “Look, growth is giant, profit margins are really high, but inflation is not going up. It’s very tame and is going down, so there must be productivity growth.” And he concluded we don’t have to raise rates if we’re getting this unexpected productivity growth.

But by the end of the ’90s, you’ll remember it then started showing up the productivity growth and the stock market valuations began soaring and actually the FOMC began raising rates quite significantly in a short period of time. You might not attribute that to trying to pop a bubble, but there were certainly people at the time who said that’s what popped the bubble was that they increased the rates. But the Fed’s expression, if you go back and read the transcripts, they’re just nervous that the economy’s overheating, that inflation is starting to kick up, the unemployment rate is so low and the business investment is so high that they didn’t really feel like they could do anything other than raise the rates.

Liz: As you said, the Fed has two jobs, keeping inflation low and steady and maximizing employment. What does maximum employment even mean if we have a hundred million robots like Elon Musk says we’ll have doing all the jobs?

Oh, geez. Look, these are some deep questions. I will say when I see the technology people, look, they’re really good at their job and what they want is to become economists and they’ve been-

Liz: Well, you make it look so fun, Austan.

They’ve been a lot. Look, it is fun. I see why they want it. I see why they want it. It strikes me that they don’t realize that people have been saying this about technologies for hundreds of years. So many people have said it. We even have a name for it. It’s called the lump of labor fallacy, that all labor is just a lump that cannot move. And as soon as the machines are better than this lump, then everyone will be out of a job. If you asked me to take the over under on will the unemployment rate be 40% in 10 years, I’m taking the under. In a way, I feel like the technology people make more Draconian forecasts almost as a way of bragging about how good their technology is. So it’s like they want to use as a proof point. “My AI algorithm is so good, it’s going to replace everyone’s job.”

It’s going to change a bunch of jobs, but it’s also worth remembering this happens all the time. If you go to the Bureau of Labor Statistics, the average person by age 50 has had 13 different jobs.

Liz: How confident are you in the future of being an economist?

Not 100%.

Liz: That’s about where I am a journalist.

Rohan: That is the sort of high end of the market, the academic side of the market. But there was a McKinsey report a few weeks ago that predicted, and it’s a consultancy, but still, that 16 million US workers, around 10% of the workforce, and that’s primarily low income workers, should be prepared at some point in the next few years to switch their jobs because of AI. And I’m curious what you think the right policy response is if that does come to happen.

I thought you were making some joke. Is the right policy response to a consultant report saying to hire a consulting firm to come look at it? No, probably not.

Rohan: No.

That’s probably not the outcome. Look, the speed at which new technologies displace skill sets of specific workers, as I say, matters a lot to how difficult it is to adjust to the new technologies. Now that said, I liked how you phrased this McKinsey finding that they’ll have to find new skills, tasks, or something in their jobs. I would be puzzled. I’m going to go look at it. A lot of the AI use cases that are trumpeted are knowledge work kind of cases. They’re not low income work.

They’re not in the healthcare space, for example, going in and taking care of people. They are high level knowledge work, reading radiology scans and things like that. So I could be wrong. And I want to just re-highlight, by age 50, right now, these are in the data up to 50 year olds now. So before there ever was AI, 13 jobs people have had, they’re constantly thinking, where is their demand in the marketplace? Where can I build a good career? And they’re moving from thing to thing. So they’re probably right that AI, like computers before it, like a bunch of other technologies before it, and going back farther, the introducing radio, introducing telephones, introducing electricity, introducing a whole bunch of big general purpose technologies did change both the economy and the labor force. But in no way did it lead the unemployment rate to head to 100% because it happened at a speed at which the new people coming out of school, they’re going to try to figure out.

They have every incentive to try to figure out what are the technologies not replacing. And that’s the stuff that we want to focus on, the messier jobs.

Liz: But we want people to have jobs because otherwise they couldn’t buy things and the demand side of the economy would collapse. What about a world where, and there’s some emotional reasons we like people to have jobs because of sense of purpose. It’s what we do all day.

Wow, that was really clinical right there. We want you to have a job, son.

Liz: But I guess I’m asking what about a world where the economic benefits of labor get divorced from actually having a job where we tax the AI, we send everyone a check, the demand side of the economy holds up just fine. Should the Fed care about that?

Nobel Laureate Robert Fogel or John Maynard Keynes each had a theory that as time goes by and we get more and more productive, we’re going to work less and less. So if we came back in 30 years and people were working 10 hours a week at a massively high productivity because they had all these robots supporting them, so their hourly per hour of work income was high enough to satisfy consumption, should the Fed care about that? Only to the extent that they would, if you’re going to maximize employment and what everybody interprets as employment is working 10 hours a week, then fine, that’s the world we would be in. We’re not in that world right now. And I think with the unemployment rate barely above 4%, and that’s for where the majority of people are working full time, that is my conception of what the full employment benchmark should be.

And all of the rest is just speculation. So I’m not going to speculate about what the job market is going to be in 10 years, except to say for the people who say it’s going to cease to exist, I kind of think you got 500 years of history that say that’s probably not true. No, maybe this is different from all the others, but I kind of also want the folks saying that whatever laws of economics from the last 200 years of industrialized society do not apply. I kind of need them to acknowledge that in the last 10 to 15 years, this is like the fifth time that they’re saying that. We had the blockchain was going to revolutionize and change everything. NFTs were going to change it.

Liz: You mean you’re not living in the metaverse?

No, I’m not on the metaverse. I’m not riding a Segway everywhere. Remember the Segway was going to change cities for 15 years. We’ve been hearing that within three years, self-driving cars would eliminate every single professional driver in the economy. And it may still happen, but it didn’t happen in three years as they were predicting at that time. And I would just, again, point folks to the history of general purpose technologies. It takes a long time for those to work their way through the economy. And if they don’t happen in a two-year period, we adjust. And I’m not making light of sectors who are directly impacted. If coders coming straight out of college are replaced or have to change what they’re doing, it can be for sure traumatic and adjustment. And if you look at old people like me, if you’re in your 50s, 60s or beyond, can you keep up with the new technologies?

That’s again, a dynamic that we have seen before in the US and around the world. It comes with transition costs, but it doesn’t blow up the economy and it doesn’t blow up the labor market. So if I had to bet, again, like I say, I would take the under on what the unemployment rate is going to be in 10 years.

Rohan: Let’s forget about 10 years. Let’s talk about today, which is I know where you want to go. We’ve been in a sort of low hire, low fire world for some time. Is that fine?

No, it’s not fine. It’s not fine, but there’s nothing the Fed can do about it or there’s little the Fed can do about it. It’s weird, and I like that you raised it. It’s worth just taking a beat and emphasizing how odd such an environment is. If you have low hiring, you usually have high firing. And the reverse, that’s the business cycle. My explanation has been that’s a thing that characterizes uncertainty, that in uncertainty a business would say, “We’re definitely not going to hire anybody new, but we’re not going to get rid of these people either because what if we got rid of them and then we resolve the uncertainty and now we’re going to have to go out and try to hire them.” But I’m going to have to revise my view. I still think that’s true. That does characterize uncertainty, but it can’t be uncertain forever.

And that’s why in your question, you said it’s been going for a long time. If that goes for five years, then I kind of think I’m going to need a rethink of was it uncertainty.

Liz: So that’s one half of the mandate, employment. You talked a little bit about inflation at the top. How much of what we’re seeing do you think is sort of pure AI? How much is the oil shock from the Middle East, and how much is inflation from the post-pandemic era that never went away?

I don’t know the percentages, but in my heart, I’ve wanted it to be from the supply shocks because they’re supposed to go away. The fact that the services inflation is inflated and kind of going the wrong way makes me think there might be more to it than just the supply shocks. And the part that’s coming from AI, I’m getting increasingly nervous that that’s just straight old-fashioned overheating demand. I don’t think that it has overheated demand in the aggregate sense, but it’s starting to. So I was just in downstate Illinois, met with the Illinois Manufacturers Association, went by University of Illinois, talked to a bunch of business people. Everybody’s got a concern about cost, about prices. There’s nothing about ‘demand is not there.’ So it sounded very much like overheating and/but a lot of it is competition for factors, resources, et cetera, competing against the data centers.

So they’re like, the cost of construction is crazy because we can’t get any electricians. We got a one-year waiting list for all these different kind of electrical component parts, and that’s because of the data centers. To the extent that data centers are up and everything else is down, that’s not aggregate overheating. That’s a change in the relative composition. And you can see why it’s going to agitate people, but that’s slightly different than overheating.

All of that said, this is starting to feel like aggregate overheating, that the AI data center cost increases are jumping out of its lane, that prices are going up in spaces that are not just components in data centers and the directly adjacent industries to data centers, that you’re seeing, if you start to see more of this spreading into general services, that investment goods are just getting GDP growth up faster than potential and generating this kind of spillover inflation, that’s problematic because unlike supply shocks where we tend to assume they’re going to go away on their own in a relatively short period of time, nobody thinks that of overheating demand. It takes a long time for overheated demand to get going, and once it’s going, it’s hard to get rid of that. So that’s why the conventional wisdom among economists has been if you see overheating demand, you act decisively and you act quickly to prevent that from rolling into inflation expectations, rolling into people taking for granted that it’s never going to go away.

Liz: Okay. Well then the big question is, should the Fed keep raising rates?

So back to some forward guidance. Look, as I always say, I don’t like tying our hands. When the dot plot came out, I know what my own dot is. I’m one of the most optimistic people in the dots that if we can just show that we’re on path back to 2%, we don’t have to actually have gotten to 2% inflation, but that was a realistic path back. I think the ultimate settling place for rates is someplace a fair bit below where we are right now.

Now that said, I’m not a voter. Chicago’s not a voter in 2026. I was a voter in 2025. I will be again in 2027. At the end of ’25, I dissented against the rate cut because if you go back and look at what I said, I believed that we could get back to path to 2%, but I was not comfortable assuming that this was going to be temporary, transitory, and just cutting preemptively the rates, assuming that it was going to go away.

I wanted to see some evidence that it was actually going away. And that was the meeting, you’ll recall, when the government was shut down, so we didn’t even have any data. I don’t regret dissenting at that meeting because we have had persistent inflation ever since then. So if inflation remains persistent, then I think it’s clear what the Fed has to do. If we can get some evidence that this tariff inflation is going away, that the oil shock inflation is going away, it’s not too much progress that we need to show, but we need to start showing some progress if we’re going to get back on the kind of rate cutting path that we were on.

Liz: But is the Fed’s primary tool, interest rates, still as powerful as it used to be? A lot of froth, you’re talking about this overheating, a lot of it’s coming from big tech companies-

That’s a good question.

Liz: ... which seem totally immune to the cost of borrowing for the moment.

Yeah, look, it’s an important question. I don’t totally know that we have an answer to that. I’ll give you a couple of places where it’s pretty clearly the transmission and short-run effectiveness have gone down, and that is if you back up not that many years, a large fraction of mortgages were adjustable rate mortgages. So when the Fed changed rates, if it translated into mortgage rates, it did so kind of quickly and that became a channel of transmission. Now the overwhelming majority of mortgages are 30-year fixed mortgages. So when the Fed changes the overnight interest rate, the impact through the mortgage channel is reduced.

That said, if the aggregate economy is overheating, the Fed has to use the tools that it has, and the tool it has is the interest rate, and normally it works because the most cyclical industries are also the most interest-rate-sensitive, and those are housing construction, consumer durables, business investment. One part of the AI landscape is about massive business investment. So far, it hasn’t been very interest-rate-sensitive, but it does seem like they’re getting more debt-financed in these investments. So if it were really overheating and the Fed started driving up rates, I think the cost of investment would get more expensive. There’s always collateral damage.

Liz: But boy, is it going to be wildly unpopular if in the hopes of finding the pain point for these big tech companies, the interest rate at which they will actually change their behavior just becomes intolerable for the rest of us, that if AI is so existential for these companies that they are just going to keep borrowing and investing until mortgages cost 15%, that doesn’t seem great.

Yeah. Look, I agree with that. I agree with that. But in a different way, if you were a person who was really into your own popularity, why would you choose to be a central banker? That’s the dumbest job you could ever have. The biggest majority of Americans say they don’t know what the Fed does but they think it’s doing a bad job at it. And so the law tells us what the job is to maximize employment and stabilize prices. And I like to emphasize, notice there’s nothing in there that says make sure the stock market is happy and there’s nothing in there that says make sure the bond market isn’t surprised or make sure the president or the Congress or whatever.

Our job is to look at the state of the economy, and you can see in the minutes and you will see in the transcript when it comes out word for word, that is what happens at the FOMC meeting. People take the job extremely seriously and it’s all about the economy.

Now that said, as I indicated about the AI heating up, I don’t actually think that they’ve driven the output of the United States much higher than potential so far at this point. And the unemployment rate is 4.2%. We just went through a period where it got down into the lower threes. So I don’t know that this is exactly what overheating looks like. And I am hopeful that the supply side can heal itself and we can kind of go back to what the economists call the divine coincidence, is that normally trying to minimize inflation, not minimize, trying to hit your inflation target while maximizing employment, they’re not in contradiction. You have one tool, but they work together. And so I’m hopeful we’re going to get back in that kind of environment.

Liz: Milton Friedman, I think, had this great analogy for central banks about the fool in the shower who’s constantly scalding and freezing himself by turning the knob faster than the water can adjust. How do you think about the line between being too twitchy and being too slow?

One of the harder degrees of difficulty is to get the timing right when there are moments of transition. It’s not clear we’re in a moment of transition, but that’s what we’re trying to figure out. I lean toward the steady hand and away from the twitchy. I think it’s clearer that way. As you know, I was a protege and friend of Paul Volcker and worked through the financial crisis with him. And his line, which I try to follow, is he was speaking about the market and he said, “Let’s remember, our job is to act and their job is to react and let’s not get the order mixed up.” And I kind of think if you look at the actual interest rate, it smoothly goes like this. And if you look at the market projection of “what do they think the interest rate is going to be,” it kind of goes like this. And it’s better that that is not the Fed, because I kind of fear if the Fed was doing this, then the market would be doing this, rather than turning itself into the smooth part.

Liz: But you’re getting at, I think, another metaphor I’ve always liked about central banks, which I think is a Mervyn King thing, a Bank of England, the Maradona theory of interest rates, which goes back to this goal that Diego Maradona had scored against England in the World Cup, and that he kind of just went straight down the field and the defenders were all expecting him to zig and zag. And so they were trying to zig and zag, and he actually just took it all the way down and scored. And I think that’s kind of the argument you’re making here.

And he just went straight down the mill.

Liz: Straight down the field.

Yeah. Look, I think that’s argument. Mervyn King is an interesting guy. When you can pull that off, that’s the best of all because it’s in a way predictable. The difference between soccer and central banking is it’s not a zero-sum game. We can do this thing together. Let’s all hold hands, get back on the golden path, get inflation headed back to 2% with full employment. Rates, I think, can come down if we could get some evidence that we’re on that. But the more you see services inflation going up and the more you see AI demand jumping out of its lane and overheating the rest of the economy, then it’s going to be a little dicier.

Rohan: You’ve made this point before that there’s nothing in your statutory mandate about weighing the bond market or the stock market in your decision making, but at the same time, how do you weigh the job that you have versus the risk of sparking a market crash or correction?

As long as stable employment, maximum employment, and 2% inflation, the market does not think that that warrants a crash, then good for the market. But the stock market is not the economy. And I always cite the example because it happened on my mom’s birthday in 2020. So I’ll always remember, it was on May the 8th, 2020. That was the day they announced the jobs report for April, that the economy had lost 20 million jobs in a single month and the stock market went up that day. So is that because the stock market thought it was going to be even worse? Is that because... Who knows? We can go way deep into the hall of mirrors to try to figure out the psychology. Our job is to think about those two things. It’s not to think about whether the stock market is going to be upset or what they felt about the dot plot or any of those things.

Liz: So you’re not afraid of the bond vigilantes is what you’re saying? They don’t keep you up at night.

Look, we monitor financial stability. In periods where interest rates rise rapidly, you often see things break in the financial system. So we always got to think about those, but I don’t believe in fiscal dominance over the monetary authority and I don’t believe in financial dominance over the monetary authority. We have our job given to us by the law. We have to do our job. And if knowing that we’re going to do our job changes your behavior, then that’s what it is. If we start basing what we’re going to do on what they think we’re going to do, you can get yourself tied up in a pretzel.

Rohan: Switching gears a little bit. Do you think that the hyperscalers with increasingly longer dated debt are starting to compete with the US Treasury for investors’ money?

It’s an important question. When I’ve talked to people in financial markets about that, there are widely differing views. Each very strongly held. As you know, the history of the question, does the amount of issuance of treasuries in a short period affect the interest rate? And if so, by how much? That is already controversial. So if you’re of the view that the last however many years in which the US debt has gone up a lot and the interest rate hasn’t gone up that much, then it’s a puzzle why. In a way the closest substitute to treasury debt is more treasury debt. Why having extra treasury debt did not increase the interest rate, but having hyperscaler debt of a similar duration did increase it? I don’t think we know the answer to how much it is. There’s a plausible case to be made that it would have some effect because the flow does matter.

But are people really attributing that AI data centers are the same or similar credit to US treasuries? Those don’t seem like perfect substitutes.

Liz: I don’t know, but there were times in the last year that Microsoft bonds traded tighter than treasuries, and certainly-

You have a point.

Liz: ... some group of investors will look at Satya Nadella and think that is a more stable and predictable management team than the one in the US government.

You’re getting me way out of my lane. I have no opinion. I have no opinion. The fiscal policy of the country, they can do whatever they want. I always say our Midwest thing is there’s no bad weather, there’s only bad clothing, and those are just weather conditions. You tell us what the conditions are and we’ll pick the right jacket.

Liz: Okay. It feels like a great place to leave it. Listen, thank you so much, Austan. This was really fun.

Rohan: Thanks, Austan.

Hey, it was great to see you and thanks for having me.

Liz: We’ll see you soon.