We Asked Vanguard's Chief Economist Why AI Has Two Huge Tails — And Which One Wins

ELI5 / TLDR

Vanguard’s chief economist Joe Davis built a model that throws four century-old forces — technology, demographics, debt, and trade — into the same pot and lets them fight for control of growth. His conclusion: AI is the only force big enough to win that fight, but only if it does a specific thing. If AI just replaces workers (like the farm tractor), we get a two-year sugar high and then crushing deficits. If it also makes workers better and spawns whole new industries (like electricity did), growth jumps to 3% and everything else gets papered over. He thinks the good outcome is roughly twice as likely as the bad one — but the bad one carries a tail where 10-year Treasury yields top 9%.

The Full Story

Why long-term trends aren’t a “worry about it later” problem

The standard macro playbook looks at near-term stuff — GDP, inflation, what the Fed does next — and treats the big slow forces (aging populations, debt, globalization) as background noise for some distant decade. Davis’s central move is to argue that’s a mistake. When those slow trends change direction, they don’t politely wait fifteen years to matter; they reach into the present and bend the business cycle now.

“When those trends start to change… they themselves affect the near-term, not just some long-term assumption that say, ‘Hey, I’ll worry about that 10 or 15 years from now.’”

So his team built a model that runs four structural drivers — technology, demographics, fiscal deficits, globalization — alongside the usual GDP-and-inflation machinery, all in one “living, breathing system.” Think of it like a horse race where the horses are these forces, and at any given moment you can read off which one is pushing growth up and which is dragging it down.

The one finding that surprised even him

Here’s the eye-opener. Conventional macro assumes that when GDP wobbles in the short term, it’s all about demand — consumers spending more or less. Raise rates when it’s hot, cut when it’s cold. Davis found that’s only true half the time. The other half, the wobble comes from the supply of ideas — new innovation and investment accelerating because productivity is about to shift.

“Changes in mega trends explain roughly half of the S&P 500’s quarterly movements.”

The analogy he reaches for is oil. An oil trader knows the price isn’t just about how much we’re driving (demand) — it’s also about how much oil is being pumped (supply). Macro forecasters, oddly, mostly ignored the supply-of-ideas half. And that half matters differently: when growth rises because ideas are multiplying, inflation eventually falls and you can justify higher valuations — the opposite of the demand story.

Two tails, and why one is twice as likely

This is the heart of it. Davis breaks technology into three jobs it can do. It can automate (replace human work), augment (make humans better — think co-pilot), and become a platform (spawn entirely new industries that didn’t exist before). The magic is in the last two. Electricity didn’t just light factories; it gave us movie theaters and vacuum cleaners. The car didn’t just replace horses; it remade retail.

His baseline (the likely tail): AI does all three. It overcomes the demographic drag and the debt overhang, lifts US growth from 2% to 3%, and the higher tax revenue lets the government keep kicking the fiscal can down the road — just like the late-’90s boom briefly turned the deficit into a surplus.

“Either the trend for growth is going materially higher because of the innovation of AI… that overcomes the demographics and the debt levels we have — that is by far our most likely outcome.”

The disappointing tail (he’d retitle his book’s “deficits dominate” chapter to just “AI only automates”): AI flashes brilliance but only saves time, never makes us meaningfully more productive or creates new fields. It’s the farm tractor — useful, but it doesn’t rewire the economy. You get a two-to-three-year sugar high, then growth fades back to 2%. Now the structural deficit, already near 6% of GDP, marches toward 8% or 10% as the population ages and Social Security and Medicare bills come due. Currency pressure, bond-market pressure, the Fed forced to hold rates high. In this world there’s a 20-25% chance the 10-year Treasury yield clears 9% over the next 5-10 years. He puts this tail at roughly half the odds of the baseline.

What inning are we in, and where the next money goes

Davis dates us — economically, not market-wise — at “1996, 1997”: the full buildout, not the 1999 peak. He has 130 years of investment-rate data across railroads, electricity, and the rest, and by that yardstick we’re nowhere near done. He’s careful to dodge the word “bubble” — tulips have no intrinsic value, but these technologies genuinely rewire the economy. Still, he warns that every great technology has had a significant drawdown, where the economic transformation and the stock market temporarily “divorce.”

“There has never been a great technology that has not had a significant drawdown in stock prices.”

So the buildout phase has maybe another year or two. Then comes a rotation: money stops flowing to the makers of the technology and starts flowing to the users — adopters with unmet needs and high costs to serve. He names healthcare and financial services as the two sectors most likely to benefit in phase two, precisely because they’re labor-constrained, expensive, and full of demand that would surge if costs fell. Note that these sectors mostly sit outside the Magnificent-Seven growth universe — which is why the logic quietly nudges you toward value, and even outside the US toward aging economies (Europe, Japan) that desperately need the automation.

The tariff shrug and the multifactor scorecard

Why did the market shrug off last year’s tariff hikes, which were several standard deviations large even in his long-run data? Because trade is only 5-10% of the US economy, and because AI investment — trillions and growing — kept accelerating and offset the drag. Five years ago, without that AI tailwind, the same tariff and oil shocks could plausibly have meant recession.

His practical takeaway for an investor: keep a mental “multifactor scorecard.” Don’t react to one headline — oil at $120, a geopolitical flare-up, the deficit — in isolation. Put all the factors, good and bad, on one ledger.

“When I do that… it doesn’t lead to drastic changes in one’s portfolio if I had just rather versus if I had just been looking at one of those factors in isolation.”

On the 60/40 portfolio: still perfectly viable. The only real risk is concentrated in the “60” (equities), and only in the disappointing-AI tail. His parting wisdom, borrowed from Franklin and Bogle: the amount you save and let compound will dwarf any cleverness about timing the market.

Key Takeaways

Vanguard’s “mega trends” model integrates four structural forces — technology, demographics, fiscal deficits, globalization — into the near-term forecast, rather than treating them as distant background.
Roughly half of the S&P 500’s quarterly moves trace to changes in these mega trends (supply-of-ideas), not just demand-side business-cycle swings.
Technology works through three channels: automation (replace work), augmentation (improve work), platform (create new industries). The last two are where the real economic transformation lives.
Baseline (most likely): AI is more transformative than the PC — it affects 80% of occupations at twice the PC’s speed (4 years vs 15). Lifts US growth from 2% to 3%, which forestalls the fiscal crunch.
Disappointing tail (~half as likely): AI “only automates” like the farm tractor — a 2-3 year sugar high, then growth fades and the structural deficit climbs from ~6% toward 8-10% of GDP. ~20-25% chance the 10-year Treasury tops 9% over 5-10 years.
Davis dates the cycle economically at “1996-97” — full buildout phase, not the 1999 peak. Probably 1-2 more years of buildout.
Every great technology has had a significant stock-price drawdown where economic transformation and market performance temporarily “divorce.” He avoids calling it a “bubble” because these technologies have real intrinsic value.
Phase-two winners are the users, not the makers — sectors with unmet needs and high cost-to-serve: healthcare and financial services lead. These sit outside the mega-cap tech/growth space, which tilts the logic toward value.
Counterintuitively, in the disappointing-AI tail the optimal hedge is short-duration fixed income — not because rates are low (they’d be high) but to cushion the equity drawdown that hits large-cap growth hardest.
US tariff and oil shocks have had limited impact partly because trade is only 5-10% of GDP and AI investment offset them; the same shocks hurt the UK and Europe more.
The 60/40 portfolio remains viable; its risk concentrates in the equity “60” and only in the disappointing tail. Compounding your savings beats timing.

Claude’s Take

This is a thoughtful, honest interview, and Davis is admirably allergic to narrative — he keeps insisting the conclusions came from the data, not from his gut, and he flags his own surprise repeatedly. The framing of AI as three distinct jobs (automate / augment / platform) is genuinely clarifying, and the historical grounding across 130 years of technology cycles gives the “two tails” story real spine rather than vibes.

A few BS-filter flags. First, the precision is theatrical: “80% of occupations at twice the rate,” “3% growth for 2027,” “20-25% probability of a 9% 10-year.” These are model outputs dressed as findings, and a model that integrates four notoriously hard-to-forecast forces is a model with enormous degrees of freedom. The honest version is “AI either transforms or disappoints, and we lean optimistic” — the decimal points are confidence theater. Second, there’s an unavoidable house-view tint: a Vanguard economist concluding that the 60/40 is fine, that active managers mostly don’t beat it, and that compounding-and-saving beats cleverness is… exactly what you’d expect Vanguard to conclude. None of it is wrong, but it’s not disinterested either. Third, the “it’s not a bubble, it’s a consolidation” distinction is doing a lot of comforting work — both still involve a significant drawdown, and the label won’t matter much to anyone living through one.

Where it earns its keep: the multifactor-scorecard discipline (don’t react to single headlines), the rotation thesis (money flows from technology-makers to technology-users), and the genuinely useful reframe that economy and stock market routinely move out of step. A 7 — substantive and clearly argued, docked for false precision and a faint house-view gravity.