The Federal Reserve Bank of San Francisco's own model shows zero growth in underlying US productivity over the past 18 months, even as AI investment surges.
Fed chairman Kevin Warsh argued before taking the job that AI would be inherently deflationary, creating room to cut rates. Matt Klein says the 1990s precedent points the other way: a real productivity boom pushes interest rates up, not down.
"But even if you were to sort of naively say it's the same thing, the implication that interest rates would go down, I think is not correct. I think it's very plausible that if that were to happen, interest rates would go up."
Klein spent more than a decade covering the intersection of economics and markets at the Economist, Bloomberg, Barron's and FT Alphaville before starting The Overshoot, and co-wrote "Trade Wars Are Class Wars" with Michael Pettis.
I listened to the full interview so you can skip it. 62 minutes of audio, 20 minutes of reading.
Here are the 15 takeaways that matter.
👤 Guest: Matt Klein, who writes The Overshoot and spent more than a decade covering economics at the Economist, Bloomberg, Barron's and FT Alphaville, and co-wrote "Trade Wars Are Class Wars" with Michael Pettis
🎙️ Host: Alan Dunne, founder of Archive Capital, who spent a decade as Managing Director at Abbey Capital and sat on its investment committee
📰 Published: 8 September 2026
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 1 hr 2 min | ✅ Time saved: 42 min
Key Takeaways
A real AI productivity boom would push interest rates up, reversing what Kevin Warsh argued before he took the Fed chair
Real rates rose once the 1990s productivity boom actually started, from a higher base than today
Underlying US productivity growth has been at zero for the past 18 months, based on the San Francisco Fed's own decomposition
Its model implies people are working harder, not smarter, so far
AI's productivity gains will take years to spread past software engineers, the way it took decades for businesses to reorganize around electric motors
Most of the incremental AI capex money is flowing straight out of the US as imports, mainly to Taiwan and Korea
That is the reverse of the 1990s, when the investment boom mostly stayed domestic
A 5% return on a 30-year bond is a hard sell against an economy growing near 7% a year in dollar terms
Higher interest rates are not automatically bad news — Japan's 10-year yield just hit a 30-year high because the country may have finally escaped decades of near-zero growth
The scenario actually worth worrying about is Brazil's, where extraordinarily high real rates and a hyperinflation legacy trap a country in permanently high debt service
Consumer spending in China relative to what the country produces is lower than in almost any other economy in history, comparable only to corporate tax havens, oil states and Stalin-era Soviet Russia
Some 50 to 70 million rural migrant workers lost their city jobs in China's initial 2020 lockdowns
China's trade surplus has grown from roughly $100–200 billion a year before the pandemic to more than $1 trillion now
A new Plaza Accord to force the yuan higher is a non-starter, because Beijing believes the 1985 original was a US trick that crippled Japan
Much of the actual US tariff effort has landed on countries not driving global imbalances, while China has "comparatively been let off the hook"
The Treasury's yen purchases were framed as $5 to $10 billion, but Klein's former colleagues at FT Alphaville put the real figure at $500 million
The Fed's balance-sheet task force is the one to watch, precisely because it is the question officials are most divided on internally
1. Warsh's Jackson Hole Debut
Dunne opened by asking for Klein's read on Fed chairman Kevin Warsh's Jackson Hole speech, delivered a week before the recording.
Klein called the speech itself reasonable, but said it had ground to make up. "I thought the speech itself was very reasonable." Before Warsh's nomination, Klein said there were real concerns he would simply cut rates to please the administration, and his Productivity and Jobs Task Force looked like an attempt to engineer a predetermined conclusion — Warsh had said "the advent of AI-related technologies would inherently be deflationary and create the scope for lower interest rates"
Jackson Hole was clearer than his first press conference on two points that matter. Warsh stated plainly that the inflation target is the core PCE 2% figure, and acknowledged there is still a gap between actual and target inflation to close
Klein's own read of the economy lines up with what he heard from Warsh: things have stalled, not improved. The job market is stable, businesses are investing, consumers are spending, and inflation has made little further progress in the past three years, after making real progress off the 2022 peak. Klein said Fed officials had been slow to accept this because they expected the aftereffects of 2021 and 2022 to simply unwind on their own
Markets read the speech as opening the door to a rate rise rather than a cut, repricing toward the Fed raising short-term rates from here
2. The 1990s Productivity Test
Dunne asked how closely the current AI moment tracks the 1990s productivity boom that Klein studied while helping Sebastian Mallaby research his biography of Alan Greenspan.
No productivity acceleration has shown up in the data yet, and history says that can take years even when the technology is obviously real. In the 1990s, computers and the early internet were already visible by 1995, but the productivity data did not start accelerating until 1997. "He didn't start making that argument until the middle of 1996," Klein said of Greenspan — ahead of the official data, but not by much, and far less than today's AI narrative is already running ahead of the numbers
Interest rates were meaningfully higher before the 1990s boom even started, despite lower inflation. Overall nominal growth then was similar to today's because stronger real growth offset the lower inflation
Once the boom began, real interest rates rose — imperfectly measurable given how new inflation-protected Treasuries were at the time, but the direction is clear, and Klein thinks they arguably should have risen more than they did given how much investment demand the new technology created
The 1990s slowdown in inflation had almost nothing to do with the productivity boom itself. "The productivity boom in the 90s mostly manifested as more growth rather than less inflation." What disinflation did occur came from emerging-market crises crushing commodity prices, Japan's post-bubble slump, and a rising dollar — not from productivity
Klein's conclusion, even on a favorable reading of the AI story, runs opposite to the deflation case Warsh made before taking office: a genuine repeat of the 1990s argues for higher rates, not lower ones, and countervailing forces since Covid and the wars in Ukraine and Iran — countries stockpiling goods and building domestic capacity — point the same way
3. Why AI Diffusion Takes Years
Dunne asked whether AI's productivity impact might show up faster than the internet's did, given how quickly sentiment shifted this year.
The gap between inventing a technology and it showing up in productivity statistics is usually long, Klein said, even for a technology that clearly works. Software engineers are the clearest early adopters, because AI is demonstrably good at helping write code
Two open questions determine how far that goes: whether the coding-assistance use case has already hit its ceiling, and how long it takes other industries to find comparable gains
The electricity analogy is his standard reference for how slow that second step can be. "People discovered how to use electricity, electric motors and stuff, but it took decades before people understood all the ways they could update their existing processes to be more efficient and fully take advantage of what the advent of electric power and electric motors will let you do."
Driverless cars show the same lag, playing out in real time. San Francisco allows Waymo vehicles and they are now a common sight, while many other jurisdictions have not authorized them at all — and Klein said most people are not even aware the technology already operates at scale
As long as humans are the ones making the adoption decisions, the rollout takes time, whatever the technology's own capabilities already allow
4. Productivity Is Still Flat
Dunne asked why the stronger productivity growth of a couple of years ago seems to have faded.
Productivity is not something that can be measured directly at the economy-wide level. "Productivity itself is not something we can measure, right?" It is derived by dividing total output by total hours worked, which leaves plenty of room for measurement error
The pandemic produced a textbook example of that error. When low-paid restaurant and retail workers lost their jobs in 2020, the remaining, higher-paid workforce made headline productivity look like it had exploded — a composition effect, not a real productivity gain
The genuine trend has been close to flat, with output edging a little ahead of trend by the end of 2024 and running roughly on trend since
The San Francisco Fed strips out effects that aren't real productivity — a more experienced workforce, more capital per worker, and people simply working harder when the economy runs hot — to isolate an "underlying" productivity measure. Its estimate: zero growth in underlying productivity over roughly the past 18 months, possibly slightly negative
The Fed's own model implies people are working harder rather than smarter so far, which is kind of the opposite of what you'd think, naively, based on what we're seeing with the technology, Klein said
5. AI Doesn't Break GDP Math
Dunne asked whether treating AI as a distinct new factor of production, as Warsh has suggested, changes anything about how economists model growth.
Klein doesn't think it does. Existing accounting methods already handle intermediate inputs, and tokens can be treated much like energy or another material input
The agency that produces GDP data, the Bureau of Economic Analysis, is already thinking about this. Klein said its economists are aware of the classification questions and have published papers on how AI spending translates into value across the economy
The reason it isn't slipping through the cracks is that it costs money. When a company buys AI products or services, that purchase shows up in the national accounts as a cost of goods sold, the same as any other input, with someone recording the expense and someone else the income
Klein is generally skeptical of claims that a new technology overturns the fundamentals of economic measurement. In his experience, the framework holds up; what changes is how a new thing gets classified within it
6. Capex Boom's Import Leak
Dunne raised the possibility of an AI capex boom-bust cycle, distinct from the dot-com era's valuation bubble because this one rests on real spending.
The 1990s investment boom was real and broad, not just a dot-com stock story. Real net investment, adjusted for depreciation and inflation, saw its biggest sustained increase since data began in 1929 — and that holds even excluding high-tech sectors like semiconductors and computers. Industrial capacity growth over that period was similarly unprecedented, on Federal Reserve data going back to the 1940s
The bust that followed was more damaging than usually appreciated. Business investment collapsed, and the government and the Fed offset it largely through a housing bubble that Klein does not think was a sustainable substitute — growth stayed weak through the 2000s in part because of it
The key difference this time is where the money actually goes. In the 1990s, most of what companies bought was made domestically. Today, when the GDP accounting is done properly, most of the incremental spending on data centers flows abroad. "The big beneficiaries of that are Taiwan and Korea," to a lesser extent other places
That has an optimistic reading: if AI investment slows, the direct hit to US production could be smaller than it looks, because so much of the spending never stayed in the country. But Klein said "the idea that the US economy is entirely dependent on AI growth I think is a misreading of what's been happening" cuts both ways — a slowdown would still hit equity valuations hard, and he does not see an obvious way to get a transition without some kind of downturn
7. Why Bond Yields Could Rise
Dunne turned to the bond market, noting 10-year Treasury yields near where they stood in 2007, though still below their 2023 level, with 30-year yields running higher still.
Klein's framework for a long bond is a simple opportunity-cost comparison. Locking up money for 10, 20 or 30 years for a fixed nominal return only makes sense against two alternatives: short-term cash that keeps flexibility, or a risk asset with more upside but no guaranteed payout
With the US economy currently growing near 7% a year in nominal dollar terms, a fixed 5% return for three decades is a hard sell unless an investor expects growth to slow sharply. "If you think it's going to 7% now and then it's going to go to 3%, then 5% looks great as a return," Klein said — but that is a bet on a specific slowdown, not the base case
He does not trade this view himself, holding target-date retirement funds rather than positioning around it directly
8. Deficits Don't Set Rates
Dunne asked how much of the move in yields traces to the US budget deficit, contrasting it with the government surplus of the late 1990s.
For a country like the US, the deficit itself is not a default risk, because the central bank can always finance the government if needed. That creates an inflation risk, not a solvency one — true as well for the UK, Japan and New Zealand, but not for countries that borrow in someone else's currency
The relevant question is whether deficits are adding to or subtracting from overall nominal growth, not whether the deficit level mechanically sets the rate. "Households matter. The domestic business sector matters. The rest of the world matters." All of it has to be netted out together, and in the 1990s rates were higher than now even with a government surplus
Households and businesses have been unusually conservative since the financial crisis and the pandemic, spending more than in the 2010s but without borrowing more to do it — leaving the government, and foreign buyers in aggregate, to finance the resulting gap
There is no clean policy lever that shrinks the deficit without also hitting growth, in Klein's view. Tax increases or spending cuts that reduce nominal growth would lower rates, but that is not the same as a costless fix. Small countries — Canada in the early 1990s, Sweden and Norway after their banking crises — escaped partly by letting their currencies depreciate sharply, an option Klein does not think is realistically open to an economy the size of the United States
Dunne noted that, for now, nominal growth (G) still exceeds the government's borrowing rate (R), and Klein agreed that keeps the arithmetic from being urgent yet
9. Higher Rates Aren't Bad News
Dunne asked whether the rise in yields simply reflects stronger growth.
Yields near a 19-year high sound alarming without context, but many of the intervening years were genuinely bad, which is why rates were so low. Klein said it is worth having some perspective rather than reacting to the headline comparison alone
Japan is the clearest illustration of the same point. Its 10-year government bond yield recently hit its highest level in 30 years — not a sign of distress, but a sign the country may have finally escaped decades of near-zero growth and falling prices after its "lost decades." "It seems like they finally managed to make it happen. Good for them."
Klein's overall framing: rising rates are not inherently a problem. "There are worlds where interest rates go up and it's not bad. I think this is one of those times."
10. Avoiding the Brazil Trap
Dunne pressed on debt sustainability, given a debt-to-GDP ratio far above the roughly 40% the US carried around 2000.
The deficit excluding interest is large enough on its own that debt keeps growing relative to the economy, independent of debt-service costs
Klein's framing of the "snowball" concern: as long as the US runs a current-account deficit, some part of the domestic economy has to be a net borrower, and it is better for that borrower to be the government. The government has the lowest cost of capital and the easiest ability to roll debt over, so if someone in the economy must sell more financial assets than they buy, that is the least bad sector to do it
Debt service today is still lighter, relative to the economy, than it was at its early-1990s peak. "As of now, debt service costs relative to the economy are still lower than at the peak in the early 1990s." A lot of existing debt was locked in at the higher rates of the 1980s and early 1990s and has since rolled down; more of that low-rate debt will roll off over time, pushing service costs up somewhat, but Klein said the context matters
The scenario genuinely worth watching is Brazil's. "So it's a really bad equilibrium that they find themselves stuck in." Brazil runs a budget that is roughly balanced excluding interest, but extraordinarily high real rates — a legacy of repeated hyperinflation episodes — keep debt service punishingly high regardless. Klein does not think the US is anywhere near that equilibrium, but flags it as the risk to keep in view
11. China's Record Surplus
Dunne turned to global imbalances, the subject of Klein's book with Michael Pettis, "Trade Wars Are Class Wars."
Imbalances matter, in Klein's framing, for two reasons: when they reflect a badly skewed distribution of income, and when they are financed by debt large enough to threaten financial stability. Both showed up in the run-up to the 2008 crisis and, Klein said, in the 1920s as well
China's version of the problem is consumption that is extraordinarily low relative to what the country produces. Klein said consumer spending in China, relative to the value of what's produced in China, is extraordinarily low, basically lower than what you see in any other society that's comparable ever — with the only real comparisons being corporate tax havens, oil-exporting states, and the Soviet Union under Stalin
Historically, China offset weak consumption with heavy state-directed investment in infrastructure, real estate and industry, plus exports of whatever wasn't consumed domestically. The pandemic changed the balance further: rather than support workers directly the way most of the rest of the world did, China supported production. Official Chinese data imply that the number of people who were migrants from the countryside to cities fell something on the order of 50 to 70 million people during the initial lockdowns in 2020, sending many back to subsistence farming with no unemployment support, since they were not official residents of the cities where they had been working
Domestic demand growth has stayed weak since, offset by exports. China's trade surplus in the aggregate has increased from something on the order of 100 billion to $200 billion a year on the eve of the pandemic to over a trillion dollars a year now
Even China's own officials say publicly that consumption needs to rise, but it hasn't become enough of a priority to change the trend. Klein said he was in Dalian in June when China's premier, Li Qiang, told a conference that there's no China Shock 2.0. It's China Opportunity 2.0 — which, Klein noted, does not change the fact that whoever is on the other side of the trade is losing out
12. Why Plaza Accord Won't Fly
Dunne asked about the G20's recent statement on imbalances and a Wall Street Journal column by Greg Ip proposing a new Plaza Accord.
The undervaluation of China's currency is, in Klein's view, close to uncontroversial at this point, and both a G7 expert commission and the G20 have made statements on it recently
A Plaza Accord-style deal is politically impossible with China, whatever its economic merits. Klein said the Chinese believe that this was essentially a trick that the US played on Japan to destroy them and take them out of the running as being another great power — a reading of the 1985 original Klein does not think is actually correct, but one that shapes how Beijing sees any similar proposal today. Germany's chancellor, Friedrich Merz, drew the same historical comparison around the same time
Klein estimates the yuan is undervalued by roughly 35%, which he treats as a reasonable basis for calibrating a response even if the currency itself isn't the mechanism used
Europe applying tariffs in that range carries its own risk of just being routed around. He described the scale under consideration as tariffs on a bunch of sectors on the order of 35 to 40%, but pointed to the precedent from the first Trump administration's China tariffs: direct US imports from China fell in the data, while imports from Southeast Asian countries rose by a comparable amount, and the net effect was minimal
13. The Wrong Tariff Targets
Klein said the countries the US has actually targeted with tariffs over the past 18 months don't line up well with the countries actually driving global imbalances.
China has been the one largely spared, despite being the biggest source of the imbalance Klein describes. Most of the US tariff and trade-conflict energy has instead gone toward countries that are not meaningfully contributing to global imbalances or damaging US economic prosperity
"I mean, this latest example of the Canada situation is absurd," Klein said, contrasting it with how comparatively lightly China has been treated
A McKinsey executive's own account of the business climate underlines the mismatch. Klein said the head of McKinsey's Greater China office remarked publicly, on the record, that the US-China business climate was better than it had been in a long time — much better than two or three years earlier — which he called bizarre given the tone of the public rhetoric on both sides
14. Why the US Bought Yen
Dunne asked about the Treasury's recent intervention to buy yen, framed publicly as helping an ally.
The reported size and the actual size may be very different. "There was that note that Bessent wrote in very big letters and had people photographing him buy 5 to 10 billion dollars worth of yen." Klein said his former colleagues at FT Alphaville, working from the disclosed data and assuming it was reported correctly, put the real figure at only $500 million — small either way relative to the size of currency markets
Japan has more than enough reserves to move the exchange rate on its own if it wanted to. With roughly a trillion dollars in official foreign-currency reserves, plus more held across other public entities, Japan could sell dollar assets for yen and book an immediate profit, since those assets were bought when the yen was far more expensive
Klein's read on the US motive: Bessent didn't want the Japanese to sell dollar assets because he was afraid of the impact on U.S. Treasury yields from them doing that. That is also why the US floated an alternative — Japan borrowing against its Treasury holdings at the Fed's repo facility rather than selling them outright — though Klein noted that option just costs Japan interest for no clear benefit if it's only a short-term move
Why Japan wants a stronger yen at all is a more tangled question. Japanese bond yields have risen sharply, particularly on longer maturities, but the yen has barely moved in four years, partly because rising inflation and growth expectations offset the nominal yield gain. Domestic conditions do not clearly justify a much more aggressive rate-hiking path, which is part of why currency intervention, rather than higher rates, is the tool in play
15. The Fed's Open Balance Bet
Dunne closed by asking which of Warsh's new Fed task forces is likely to matter most.
The balance-sheet task force stands out to Klein because it is the question where Fed officials themselves are least settled. "The balance sheet task force I think is very interesting potentially just because of the caliber of people they have there and the nature of the question and the fact that it's really unsettled." Its members are Raghuram Rajan, Jeremy Stein and Karen Dynan
Klein's reasoning is about internal division, not the topic's inherent importance. Public statements from Fed officials on the balance sheet suggest genuinely open disagreement about what to do and why, more so than on the task forces addressing other questions
The task forces are advisory rather than binding — decisions are still made by an internal Fed vote — but Klein said that where officials are the most divided, outside recommendations have the best chance of actually shifting the outcome
Bonus Insights
Klein, who lives in San Francisco, said the AI wealth boom has visibly spilled into the local housing market
This was Klein's second appearance on the show; he first came on at the start of last year
A tangent on Japan: non-financial Japanese companies have seen profits inflated relative to their operating profits over the past six years, because they hold large portfolios of interest-bearing assets (including outside Japan) and global rates have risen relative to Japan's
Klein writes The Overshoot on Substack, where listeners can follow his ongoing work
Klein's bottom line is that the AI story, taken on its own logic and on the 1990s precedent he knows in detail, argues for higher interest rates rather than the lower ones a deflationary AI narrative would predict — and that the more urgent imbalance to watch is not the US federal deficit but China's, now running a trade surplus above $1 trillion a year with no clear political fix in sight.
Products, Companies & Tools Mentioned
Apple (Raised MacBook and iPad prices by around 20%, Klein's evidence that AI-driven input costs are already reaching consumers)
Waymo (Driverless cars operating at scale in San Francisco — Klein's example of how slowly a working technology diffuses once adoption depends on human decisions and local rules)
Federal Reserve Bank of San Francisco (Its productivity decomposition is the source of the "zero growth in underlying productivity" finding Klein cites)
Bureau of Economic Analysis (The agency producing US GDP statistics; Klein says its economists already have a framework for classifying AI spending)
McKinsey & Company (Its Greater China office head told a public conference the US-China business climate was better than it had been in years, which Klein cited as evidence against the tariff rhetoric)
FT Alphaville (Klein's former employer; its analysts estimated the Treasury's actual yen purchases at $500 million against a reported $5 to $10 billion)
Books & Resources Mentioned
Trade Wars Are Class Wars – Matthew C. Klein and Michael Pettis (Klein's own book on global imbalances, the framework behind the China section of the conversation)
The Overshoot – Matt Klein (His subscription newsletter on economics, finance and public policy)
The Man Who Knew: The Life and Times of Alan Greenspan – Sebastian Mallaby (The Greenspan biography Klein helped research, source of his detailed 1990s Fed timeline)
Greg Ip's Wall Street Journal column proposing a new Plaza Accord for the yuan (The piece that prompted the G20 and Plaza Accord discussion; Klein said the idea is a political non-starter with China)
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: