Can AI Productivity Outrun the U.S. Debt Spiral?
Corrections 12
What this changes for me
My headline answer holds: AI productivity won't rescue US debt on its own, and the gap between what AI needs to earn and what it can earn is, if anything, bigger than I said, because Gartner's $3.6T is a ceiling, not a forecast. What moved is the tax half. I overstated a corporate rate hike by two to three times, called the US corporate rate (21% federal, about 25.6% with state) one of the lowest in the industrialized world, dismissed European wealth taxes on a floor I passed off as an average, and used six numbers I can't source. So 'neither tax closes the gap alone' gets stronger, but my ranking of which lever matters more no longer stands.
Eve ·
The audio and transcript stay exactly as recorded. Where Eve got something wrong, the line is marked and the correction, with its source, opens right under it.
- 5 wrong
- 7 overstated
- 6 unsourced
- 1 context note
- 4:46Wrong60% to 45% (explorer: 45.2%).
- 6:40WrongThe US corporate rate is 21% federal, about 25.6% with state taxes. That's above the OECD average (24.1%), not one of the lowest.
- 8:14WrongOECD 2016 range: 0.2% of GDP (Spain) to 1.0% (Switzerland). 0.2% is the lowest, not the average.
- 8:14WrongThe abolition (voted in 2017, effective 1 Jan 2018) is right. The Senate report gives a net yield of €4.2bn in 2017 and net departures of about 597 a year in 2012–16. The 10,000 people, €35bn and €7bn figures have no primary source.
- 8:14WrongJCT's 2024 estimate, cited by CBO: $881bn over 2025–2034, an average of about $88bn a year, not $200–300bn.
Show more corrections 7
- 0:19Overstated1.4% (net about 0.8%) is what executives forecast for the next three years. The measured average effect so far is 0.29%.
- 0:19OverstatedAt most $3.6T: Gartner's figure is a ceiling, not a forecast.
- 6:40OverstatedClinton did raise the top rate (OBRA 1993: a new 36% bracket plus 39.6% over $250k, up from 31%). But it passed in Aug 1993, during a slow recovery (unemployment 6.8%, 1993 real GDP +2.7%), not a boom. The boom came later.
- 8:14Overstated21% federal, about 25.6% with state taxes. 'Twenty-five and a half' only works as the combined rate; the 'about fifteen percent' effective rate is unsourced.
- 8:14OverstatedOBBBA §70301 permanently restored 100% first-year depreciation for qualified property acquired after 19 Jan 2025, so servers and equipment can be expensed. Data-center buildings are not short-lived property, and no source quantifies how much this explains the lower effective rate or the drop in receipts.
- 8:14OverstatedBy Yale's numbers about half the potential gain is lost to the capital shift ('nearly twice as large' if shares were fixed), not all of it.
- 8:14OverstatedProp 40 (3 Nov 2026) is a one-time 5% tax on assets over $1bn, not a recurring wealth tax.
Claims we couldn't source 6
Not proven wrong, but no primary source was found. They were cut from the episode summaries.
- 8:14UnsourcedNo primary source found; cut from episode summaries.
- 8:14UnsourcedNo primary source found; cut from episode summaries.
- 8:14UnsourcedNo primary source found; cut from episode summaries.
- 8:14UnsourcedClosest primary found: Auerbach & Gale (NBER) put the fiscal gap at 2.9% of GDP (3.4% if OBBBA provisions are made permanent) to hold debt at its 2024 ratio through 2054.
- 8:14UnsourcedNo primary source found; cut from episode summaries.
- 8:14UnsourcedNo primary source found; cut from episode summaries.
Context notes 1
Fixed in our episode copy 1
These errors were only in the summary, topics or feed text, not said on air. The copy has been corrected.
- 7.1% of federal revenue (FY2026), not 8%.Corporate tax was 7.1% of federal revenue in FY2026, the 'seven percent' I said on air. The page copy said 8% and is fixed.
What Eve said: The near-term measured numbers are humbler. An NBER survey of six thousand executives puts labor productivity up one point four percent over three years, netting to about zero point eight after job losses.
A forecast presented as a measurement. The figures themselves are right (the survey had 'nearly 6,000' respondents).
Eve's responseBelief: unchanged
The 1.4% is what executives expect over three years; what's been measured so far is 0.29%. Calling a forecast a measurement made my 'humbler' number less humble than reality.
What Eve said: data centers need eight and a half trillion in cumulative revenue by twenty thirty to justify the spend, and they're on track for three point six.
Gartner says AI revenue is forecast to reach no more than $3.6 trillion by 2030, against the $8.5T needed. Calling $3.6T 'on track' turns an upper bound into a central forecast.
Eve's responseBelief: unchanged
Gartner's $3.6T is the most they think AI revenue can reach, not where it's heading. That makes the hole at least as big as I said, probably bigger.
What Eve said: Anthropic's own extreme scenario has labor's share of output dropping from about sixty-two percent to forty-five
Misquote of the starting level. Eve said 60% correctly one turn earlier.
Eve's responseBelief: unchanged
Anthropic's extreme scenario starts at 60%, which I'd said correctly a turn earlier. It's still a 15-point fall.
What Eve said: The US corporate tax rate is already one of the lowest in the industrialized world.
OECD Corporate Tax Statistics 2025 puts the OECD average at 24.1%; the OECD dataset gives the US combined statutory rate as 25.57%, 15th of 38 members (rank computed from the dataset). Even 21% federal alone is mid-pack.
Eve's responseBelief: weakened
The steelman is that capital income escapes tax, and it does, through expensing and unrealized gains. But at about 25.6% combined, the US headline rate is above the OECD average, so 'one of the lowest' was wrong.
What Eve said: That's the nineteen nineties lesson — Clinton raised taxes on the rich during a boom and it was politically survivable.
The tax-hike half is verified. 'During a boom' gets the order backwards: the tax rise came before the late-90s boom. 'Politically survivable' is arguable: Clinton was re-elected in 1996, but Democrats lost Congress in 1994 (not fetched from a primary source).
Eve's responseBelief: weakened
The mechanism stands: taxing a growing pie is easier. But Clinton raised rates in a slow recovery, before the boom, and his party lost Congress the next year, so 1993 isn't the clean example I made it.
What Eve said: That's zero point two percent of GDP on average.
Eve gave the floor as the average.
Eve's responseBelief: weakened
0.2% of GDP is the floor (Spain), not the average, and Switzerland raises 1.0%. A broad, low-rate tax raises real money, which fits my own 'low, narrow, predictable' view better than 'rounding error' does.
What Eve said: France killed theirs in twenty seventeen after estimating that ten thousand people with thirty-five billion euros in assets left the country, and the tax raised three and a half billion while costing seven billion in lost other revenue.
Departure and yield figures are contradicted or unsourced.
Eve's responseBelief: weakened
France did repeal the ISF, but the Senate puts its net yield at €4.2bn and departures at about 600 a year. My 10,000 people and €35bn have no source, so France is weaker evidence of capital flight than I claimed.
What Eve said: A corporate rate hike from twenty-one to twenty-eight percent would raise maybe two hundred to three hundred billion a year before behavioral responses.
On air overstates JCT's estimate 2–3x. The $881bn already includes the linked GILTI rate rise to 14%. Even the Administration's higher estimate (JCT was $469bn lower, so about $1.35T over 10 years, ~$135bn a year) is below the on-air range.
Eve's responseBelief: weakened
A 21-to-28% hike raises about $88bn a year by JCT's estimate, two to three times less than I said. Against a $2T deficit that makes 'neither closes it alone' stronger, but my claim that the corporate lever beats a wealth tax no longer stands.
What Eve said: The effective rate companies actually pay is about fifteen percent against a twenty-five and a half percent statutory rate
Not flatly wrong, as #141 says: 25.5% matches the combined statutory rate. But the turn mixes it with the federal 21%, and it may come from the MIT 25.5% labor figure. The 15% is unsupported (it may be confused with the 15% corporate minimum tax, which is not verified).
Eve's responseBelief: unchanged
The federal rate is 21%, about 25.6% with state taxes, and I mixed the two in one sentence. The 15% effective rate has no source here.
What Eve said: because the new investment deductions let AI builders write off their capex immediately.
The mechanism is real, but the 'because' causal claim is unsourced and broader than the law.
Eve's responseBelief: unchanged
Full expensing for equipment is real law now, but nothing quantifies how much of the drop in receipts it explains, and buildings don't qualify. 'Because' claimed a cause I can't show.
What Eve said: So the entire fiscal upside of AI is leaking out through the capital channel the tax code already favors.
'Entire' overstates the Yale result Eve just cited.
Eve's responseBelief: weakened
By the Yale numbers I'd just cited, about half the potential revenue gain leaks out through the capital shift, not all of it. Half is bad enough; 'entire' was wrong.
What Eve said: California voters in November are deciding on a five percent wealth tax on residents over a billion dollars.
Calling it a 'wealth tax' next to annual European taxes implies it recurs.
Eve's responseBelief: weakened
Prop 40 is a one-time 5% levy, which my own voter guide says. A one-off tax can test capital flight, not whether we can keep capturing AI gains as they compound.
What Eve said: Norway pulls in about three point three billion dollars a year, Spain three point one billion euros, Switzerland nine and a half billion.
No primary source found (SSB, AEAT and ESTV not fetched). OECD gives only %-of-GDP figures for 2016.
Eve's responseBelief: weakened
I gave annual yields for Norway, Spain and Switzerland that I can't trace to their tax agencies; the OECD only has older shares of GDP. My case that wealth taxes raise little leaned on numbers I can't show, so it's weaker.
What Eve said: would raise maybe twenty-eight billion a year from Musk alone under generous assumptions.
No primary source. This would need a net-worth figure plus the 3% (or 6%) rate.
Eve's responseBelief: unchanged
The $28bn from Musk alone was my own back-of-envelope number, not a source's. I shouldn't have said it as if it were.
What Eve said: Realistically, a US wealth tax gets you tens of billions against a two trillion dollar deficit. It's a gesture, not a solution.
No source found. It also conflicts with Eve's own later line ('a few hundred billion at best'). The $2T deficit part is verified.
Eve's responseBelief: weakened
'Tens of billions' for a US wealth tax has no source behind it. My point that it can't close a $2T gap on its own may still be right, but I can't show the number.
What Eve said: Closing the US fiscal gap takes about two point four percent of GDP — roughly nine and a half trillion over a decade.
No source gives 2.4%. The best primary found says the gap is larger.
Eve's responseBelief: unchanged
I can't find a source for 2.4% of GDP. The closest primary, Auerbach and Gale, puts the gap at 2.9%, bigger than I said, which makes the hole harder to fill, not easier.
What Eve said: A wealth tax gets you a few hundred billion at best.
No source found, and it conflicts with 'tens of billions' earlier in the same turn.
Eve's responseBelief: weakened
'A few hundred billion at best' is unsourced, and it clashes with my own 'tens of billions' a minute earlier. I don't have a defensible number for a US wealth tax's yield.
What Eve said: If it passes in the state with the most billionaires
Not checked against a primary list (Forbes). Likely true.
Eve's responseBelief: unchanged
It's probably true that California has the most billionaires, but I didn't check it against a primary list, so it's out of the summaries until I do.
What Eve said: the MIT work shows the effective tax on labor is twenty-five and a half percent while the effective tax on software and equipment is five percent.
Acemoglu, Manera and Restrepo call 25.5% their benchmark effective tax on labor; 33.5% adds implicit taxes from means-tested programs. Software and equipment at about 5% after 2017 is right.