The $400 Bet Two Economists Made on AI Jobs — And Why Washington Is Losing Either Way

Two of America's most respected productivity economists made a $400 wager seven years ago about whether AI would supercharge or stall the labor market through 2030. According to Bloomberg, Erik Brynjolfsson of Stanford's Digital Economy Lab is pulling ahead of Northwestern's Robert Gordon — but the real story is what both men now warn about together. The country is sprinting toward an AI-driven workforce transition with almost no plan for the workers who will be displaced along the way. For job seekers reading this on Metaintro, the bet is no longer hypothetical. It is starting to show up in payrolls.

What is the $400 bet really about?

The wager itself is small. Brynjolfsson and Gordon registered it on the prediction platform Long Bets with stakes of $400, payable to GiveWell charities. The terms are precise: US private nonfarm business productivity must average more than 1.8 percent per year from the first quarter of 2020 through the last quarter of 2029. If it does, Brynjolfsson wins. If it does not, Gordon does.

That number sounds dry. It is not. Productivity growth at or above 1.8 percent annually is the dividing line between a roaring twenties scenario where AI lifts wages and output, and a stagnation scenario where AI cuts costs without producing new growth. Brynjolfsson argues AI is a general-purpose technology — the same category as electricity or the internal combustion engine — that will eventually rewire every industry. Gordon counters that AI is essentially nothing new, pointing out that for more than a decade voice-recognition systems have replaced human customer service workers without reviving aggregate growth. Both economists are now visibly worried about the same thing. The transition between those two outcomes is going to be brutal regardless of who collects the $400.

How much money is being bet on AI by everyone else?

The $400 wager is the smallest AI bet currently being placed in the United States. According to Bloomberg Intelligence projections covered earlier this year, hyperscaler capital expenditure is on track to hit roughly $3.7 trillion over the next five years — about $740 billion a year. 2026 AI capex alone is now estimated near $660 billion. That figure is bigger than the entire annual GDP of Switzerland.

What that money buys is data centers, custom silicon, power contracts, and model training runs. What it does not buy is workforce transition. The federal government's annual budget for displaced-worker retraining programs is measured in single-digit billions. The proportional gap between what is being spent on AI infrastructure and what is being spent on the workers AI will reshape is now larger than the gap between US defense spending and the defense spending of every other country combined. Read our earlier analysis of how the AI layoff wave caught Washington unprepared for more on this asymmetry.

Which workers are most exposed right now?

The Brookings Institution recently estimated that roughly 37.1 million US workers are in jobs with high AI exposure. Of those, about 70 percent have enough adaptive capacity — transferable skills, education, or savings — to navigate a role shift if their work changes. The remaining 30 percent, roughly 10.6 million workers, do not. They sit in lower-wage roles with fewer resources for retraining and less geographic mobility.

Goldman Sachs economists added a sharper near-term number to that long-term outlook last month. Their research found that AI has already erased roughly 16,000 net US jobs per month over the past year, with the impact falling hardest on Gen Z and entry-level workers — the same group that traditionally uses early-career roles as their on-ramp into the workforce. Our coverage of how cutting junior talent backfires for employers walks through what is already breaking inside corporate pipelines. The Brookings 10.6 million and the Goldman 16,000-a-month figures describe two ends of the same trajectory. One is the long-term vulnerable population. The other is the monthly attrition rate that is already running while policymakers debate definitions.

Why isn't anyone planning for this displacement?

Economists who used to dismiss AI as a labor threat have started changing their minds in the past 12 months. What they have not been able to change is the federal workforce-policy apparatus, which was designed in an era when displacement looked like a single auto plant closing in a single town. The white-collar AI shock pattern looks nothing like that. Whole job categories — junior software engineering, paralegal work, copywriting, customer service, data entry, parts of finance and accounting — are contracting at once, in every metro, with no plant gate to picket and no single CEO to call into a hearing.

Former Commerce Secretary Gina Raimondo has proposed that the federal government invest in training, incentives, and safety nets to move workers into new jobs while requiring employers to define which skills are essential to the AI economy. Other economists have floated more aggressive ideas — a "circuit breaker" plan that would trigger automatic stabilizers, including wage insurance and expanded income support, when AI-related displacement spikes past a certain threshold. None of those proposals has cleared the policy stage. Bloomberg's Brynjolfsson-Gordon piece quotes a growing coalition of economists making the same diagnosis. The disagreement is no longer about whether AI displaces workers. It is about how fast and how concentrated the damage will be — and the safety net was not built for either answer.

What can workers do while Washington catches up?

The most actionable signal coming out of the economist debate is also the one most often buried in policy discussions. Adaptive capacity is what separates the 70 percent of AI-exposed workers who will land somewhere new from the 30 percent who will not. Adaptive capacity is not a single skill. It is the combination of transferable knowledge, a credible professional profile, a network that can be activated quickly, and access to job markets that show what is actually being hired right now versus what was being hired three years ago.

Three concrete moves matter in 2026. First, audit your role against the Brookings AI exposure framework — figure out honestly whether your day-to-day is something a model can do 80 percent of by 2028, and budget your skill-building accordingly. Second, get fluent with AI tools as a coworker rather than a competitor. Workers who use AI to amplify their output are not the ones being cut first. Our piece on AI's biggest impact being minds not jobs covers that distinction. Third, watch where capex is flowing. Companies announcing nine and ten-figure AI infrastructure commitments are signaling where adjacent roles — implementation, governance, training, model evaluation, security — will be created next.

What does the bet actually predict for the next five years?

If Brynjolfsson wins, US productivity averages above 1.8 percent through 2029 and AI delivers a roaring-twenties output boom. That scenario is good for GDP and bad for the bottom 30 percent of AI-exposed workers in the short term, because productivity gains in the early phase of a general-purpose technology almost always concentrate before they spread. If Gordon wins, productivity stays sticky around 1 to 1.5 percent and the AI capex wave does not pay for itself — which means hyperscalers cut headcount aggressively to defend margins, and the displacement still happens, just through a different mechanism.

Either way, the workers in the middle lose unless transition policy catches up to capital deployment. That is the actual finding inside the Bloomberg piece. The $400 is the headline. The trillion-dollar mismatch between AI investment and worker investment is the story. For more on the broader shift, our analysis of why economists are reversing earlier dismissals of AI job loss lays out the academic about-face in detail, and our piece on how to stay employable in constant change covers the personal playbook.

People Also Asked

Q: Who is winning the Brynjolfsson-Gordon AI bet right now?

A: Brynjolfsson is currently leading. US private nonfarm productivity has run above the 1.8 percent threshold for several quarters, though Bloomberg notes it is too early to say whether that pace is durable or a temporary post-pandemic effect.

Q: How much are companies spending on AI in 2026?

A: Hyperscaler AI capital expenditure for 2026 is projected near $660 billion, with five-year cumulative capex tracking around $3.7 trillion according to Bloomberg Intelligence — roughly $740 billion per year on average.

Q: How many US workers are at risk of AI displacement?

A: Brookings estimates 37.1 million US workers are in highly AI-exposed roles. About 10.6 million of those workers lack the adaptive capacity — transferable skills, education, or savings — to navigate a forced role change.

Future-Proof Your Career

The economists betting on AI agree on one thing: workers who move first will land best. Metaintro shows you which jobs are growing, which are contracting, and what the market is actually paying right now — so you can plan your next move before the displacement curve catches up to your role.