When I want to know what's going to happen, I look at where people are putting money, because they pay for being wrong.
So I asked the markets a question: is AI ending the Great Stagnation?
That's Tyler Cowen's name, from his 2011 book, for the long slowdown in American productivity after 1973. From 1947 to 1973, output per hour grew about 2.7% a year. Through the 1970s and 1980s it grew about 1.5%, and from 2007 to 2019 it was back at 1.5% (BLS). The one clear break was the computer boom: 2.1% a year in the 1990s and 2.8% from 2001 to 2007.
| Period | Yearly growth |
|---|---|
| 1947–73 | 2.7% |
| 1973–80 | 1.4% |
| 1980–90 | 1.6% |
| 1990–2001 | 2.1% |
| 2001–07 | 2.8% |
| 2007–19 | 1.5% |
| 2019–26 | 2.1% |
On September 29, Cowen linked a new National Bureau of Economic Research paper that gives one answer. I went looking for others.
The stock market is betting on software
The paper, by Alex Blumenfeld, Jonathon Hazell, Chen Lian and Andreas Schaab, looks at companies that employ a lot of programmers but don't sell software: eBay, Airbnb, Expedia, Sonos. When AI stocks rise, these companies rise more than similar companies with fewer programmers. The authors read that as investors expecting programmers to get more productive, which cuts those companies' costs.
| Share of payroll to programmers | Extra gain when AI stocks rise 10% |
|---|---|
| 0% | 0.0 points |
| 10% | 1.2 points |
| 20% | 2.5 points |
| 30% | 3.7 points |
| 40% | 5.0 points |
Working backward from the stock moves, they estimate that by the end of 2025 investors expected software engineers to be about 33% more productive. That works out to a one-time 3.6% increase in GDP, or 6.5% if cheaper software also speeds up research. Then the estimate more than doubled in the first half of 2026. The authors point to coding agents like Claude Code and Codex.
| Period | Length | Expected productivity gain added |
|---|---|---|
| Nov 2022 to Dec 2025 | 37 months | 32.6% |
| Jan to Jun 2026 | 6 months | more than 32.6% |
The paper measures a one-time jump in the size of the economy. To compare it with a growth forecast, I did some rough arithmetic of my own: take the doubled 2026 figures, spread them over ten years, and you get something like 0.7 to 1.3 extra points of growth a year from software alone. Cowen's own forecast is that AI lifts growth from 2% to 2.5%, half a point for all of AI.
The bond market is harder to read
If investors expected much faster growth, long-term real interest rates should rise, because people would borrow against a richer future. They have risen. The 10-year inflation-protected Treasury bond auctioned at a 2.65% real yield on September 17, the highest for that term since October 2008.
AI doesn't look like the reason. Tipswatch, a blog that follows these auctions, blames the Iran war and government borrowing. And an August Brookings working paper by Jens Christensen and Glenn Rudebusch checked what estimates of the long-run "neutral" interest rate (the rate that neither speeds up nor slows the economy) did in the three days around each major AI model release. They fell, by a total of 0.23 to 0.35 percentage points depending on the measure. The authors found "no evidence" that AI news has pushed the long-run rate up this decade.
The bond market isn't pricing a growth explosion. If anything, the evidence around AI releases points slightly the other way.
Prediction markets expect normal growth, with room for surprises
Kalshi runs markets on US GDP growth for every year through 2036. For 2026, traders' most likely range is 2.1% to 2.5%, at about 40%. For 2036, they give roughly even odds to growth between 1.6% and 2.5%. They also put about a 20% chance on growth above 3% that year, and about one in six on growth of 1% or less. These are thin markets. Only about 156,000 contracts have traded on the 2036 market, so treat them as a sketch.
| GDP growth in 2036 | Implied chance |
|---|---|
| 0% or below | 9% |
| 0.1–0.5% | 5% |
| 0.6–1.0% | 4% |
| 1.1–1.5% | 6% |
| 1.6–2.0% | 27% |
| 2.1–2.5% | 23% |
| 2.6–3.0% | 7% |
| 3.1–3.5% | 2% |
| 3.6–4.0% | 6% |
| 4.1–4.5% | 4% |
| 4.6–5.0% | 3% |
| 5.1–5.5% | 1% |
| 5.6–6.0% | 1% |
| 6.1% or above | 2% |
Polymarket has a market on whether the "AI bubble" bursts, defined as any three of a list of events within 90 days, such as Nvidia falling 50% from its high or OpenAI or Anthropic going bankrupt. Traders give it about 9% for 2026, on $2.4 million traded.
Read together, prediction markets center on about 2%, close to what the Fed assumes, and give better than one chance in three that 2036 lands well outside that range in one direction or the other.
Forecasters have moved up a little
The Philadelphia Fed's Survey of Professional Forecasters now expects productivity to grow 1.8% a year over the next decade. In 2023 the same survey said 1.3%. The Fed raised its estimate of long-run GDP growth from 1.8% to 2.0% in March and kept it there in September.
The Forecasting Research Institute surveyed economists, AI company employees, policy researchers, top forecasters and the general public. The median in every group expects about 2.5% GDP growth in 2030, above the roughly 2% that government agencies use. Economists gave a 14% chance of "rapid" AI progress by 2030.
The official data is moving the same way. Productivity is up 2.2% over the past year and has grown 2.1% a year since late 2019. That's better than the 2010s and well short of the postwar years.
| Measure | Earlier | Latest |
|---|---|---|
| Professional forecasters: 10-year productivity growth forecast | 1.3% (2023) | 1.8% (2026) |
| Federal Reserve: Long-run GDP growth estimate | 1.8% (2025) | 2% (2026) |
| Actual productivity (BLS): Yearly growth, 2007–19 cycle vs. current cycle | 1.5% (2007–19) | 2.1% (2019–26) |
What I take from all this
The forecasts I can track over time have all moved toward faster growth since 2023, by a few tenths of a point. The stock market's read on software moved much more. The bond market isn't buying a boom, and prediction markets center on the old normal. So is the Great Stagnation ending? The stock market, read through one paper, says software is a real break. Forecasters see a small improvement. The rest of the money hasn't moved much yet.
| Source | What it says | Reading |
|---|---|---|
| Stock market, read through the NBER paper | Software engineers about 33% more productive by end of 2025, more than doubled by mid-2026 | Big step up |
| Professional forecasters | 10-year productivity forecast up from 1.3% to 1.8% | Modest step up |
| Expert survey (Forecasting Research Institute) | About 2.5% GDP growth in 2030, above the 2% agencies use | Modest step up |
| Federal Reserve | Long-run growth estimate up from 1.8% to 2.0% | Small step up |
| Prediction markets (Kalshi) | 2036 growth centered near 2%, with wide odds on either side | About the old normal |
| Bond market | Long-run rate estimates dipped around AI releases | No sign of a boom |
I lean toward the upside, for two reasons. First, forecasters were too cautious last time. The professional forecasters held their ten-year productivity forecast at 1.5% from 1992 to 1998, and actual growth over the following decade came in at 2.1% to 2.7%. As the economist N. Kundan Kishor put it this spring, "structural breaks are genuinely hard to see in real time." Second, the most detailed market signal we have is also the most bullish, and it doubled in six months.
Stock prices can run ahead of reality, and that paper leans on them. But it also points to the version of this story that would matter most. Cowen's original argument was that new ideas were getting harder to find. In the paper's research scenario, cheaper software makes research cheaper, and the estimated effect on GDP nearly doubles. That's the part that would go straight at the cause of the stagnation. It's also the part with the least evidence behind it so far.
If you run a company, I'd plan for an economy that grows a bit faster than the last one, with more chance of surprise than most plans allow for. And notice where the paper found investors placing their bets. It looked only at companies that use software to run their business, and among those, the ones employing more programmers got the bigger lift.