Companies Spent $443B on AI. The GDP Data Says It Didn't Work.
Will national productivity statistics still show no measurable AI-driven acceleration through end of 2028?
Companies are spending billions on AI tools. The economy-wide numbers say it's not working yet.
Your Prediction
Where do you think this lands?
Join others who've weighed in
Scenarios
Current value: 2.2% average annual productivity growth (2025); Q3 2025 at 5.2% (exceptional quarter); 89% of firms report no measurable AI gains
S-curve position: Pre-inflection — like IT in 1990s, productivity gains may take 10-15 years to appear in macro statistics
No macro signal through 2030 (Solow paradox repeats, gains captured as profits not productivity)
Flat through 2028, first signals in 2029-2030 (matches IT productivity lag)
Visible acceleration by 2027 (AI adoption faster than IT, measurement improves)
How We'll Know
- What we measure
- US Bureau of Labor Statistics quarterly labor productivity growth (nonfarm business sector) compared to 2015-2023 trend
- Confirmed if
- US labor productivity growth through 2028 remains below 2.5% annually sustained for 4+ quarters with no statistically significant AI-attributable acceleration
- Refuted if
- Productivity growth exceeds 3% annually for 2+ consecutive quarters with AI clearly cited as contributor
- Data sources
- BLS Productivity and Costs reports
- NBER working papers on AI and productivity
- OECD productivity statistics
- Federal Reserve economic data (FRED)
Evidence Trail
Evidence For
- Mar 7, 2026
NBER: 89% of firms report no measurable productivity gains. IT productivity paradox lasted 15 years. BLS data shows no sustained break from trend. PwC: 56% of CEOs say they've gotten 'nothing out of' AI investments. Federal Reserve: only 1.9% excess cumulative productivity since ChatGPT launch.→ Probability: 65%
- Jun 13, 2026
BLS revised Q1 2026 nonfarm productivity down to +0.3% (from +0.8%) on June 4, with total factor productivity decelerating from 1.5% (2024) to 0.8% (2025) — even as Dell'Oro put 2026 data-center capex above $1 trillion. Massive AI investment is not yet showing up cleanly in aggregate productivity.→ Probability: 52%
- Aug 2, 2026
The market itself started asking the paradox's question. Big Tech guided 2026 AI capex to ~$725B (+77% YoY), yet Meta and Amazon sold off despite beats and Amazon's trailing-12-month free cash flow swung to -$7.6B — the post-earnings narrative pivoted from "how much are you spending" to "what's the return." Cognizant launched an EMEA unit citing IDC's finding that 88% of agent pilots never reach production, and Microsoft's Copilot is still at ~6.7% paid penetration with 20-30% weekly use. Offset by real scaled revenue (Microsoft's $37B AI run rate, CCC's $120M AI-claims line), and no new BLS Productivity & Costs print landed this week — held at 0.52.→ Probability: 52%
- Sep 3, 2026
BLS Productivity and Costs for Q2 2026 (revised) put nonfarm business labor productivity at +1.4 percent annualized and +2.2 percent year over year, with Q2 unrevised from the August preliminary and Q1 2026 holding at +0.3 percent. This is the third consecutive quarterly print below the 2.5 percent threshold the criterion names (Q4 2025 at 1.8, Q1 2026 at 0.3, Q2 2026 at 1.4); the Q3 2026 release on November 5 would complete the four- quarter requirement. https://www.bls.gov/news.release/prod2.nr0.htm→ Probability: 56%
- Oct 10, 2026
W41 hold and a record correction. The September 3 BLS release shows nonfarm business productivity growth of 1.6% (Q4 2025), 0.8% (Q1 2026) and 1.4% (Q2 2026) annualised; our W38 entry quoted 1.8 and 0.3 for the first two. The count of three consecutive quarters below 2.5% is unchanged, and the November 5 release can complete the four-quarter leg. https://www.bls.gov/news.release/archives/prod2_09032026.htm→ Probability: 56%
Evidence Against
- Mar 7, 2026
2025 annual productivity 2.2% (above decade average). Q3 2025 at 5.2%. Individual studies show 20-40% task-level efficiency gains. Coding productivity up 30-50%. Customer service resolution times down 30-50%. AI adoption is faster than IT adoption was — the paradox may resolve faster.
- Mar 7, 2026
Brynjolfsson (Feb 2026 FT): US productivity grew 2.7% in 2025, nearly double the decade average. BLS revised payroll down 403K while GDP at 3.7% — output-labor decoupling. Furman: 2.5% over 2 years, best since 1973. Block cut 40% workforce. Dev freelance rates -36%. Cursor $2B ARR / 12 employees. 54,694 AI-attributed layoffs. Goldman Sachs: AI completes 95% of IPO prospectus in minutes (was 2 weeks, 6 people), expects 3-4x gains across divisions. Software job postings at 5-year low (-35% from pre-pandemic). These are real economic signals that should show in productivity data.→ Probability: 50%
- Apr 10, 2026
BLS Q4 2025 productivity revised down from preliminary 2.8% to 1.8% annualized. Full year 2025 at 2.1% — above decade average of ~1.4% but below the 2.7% figure Brynjolfsson cited in the FT article that the baseline used. The downward revision weakens the already-modest visible AI signal in macro data. Kansas City Fed finds AI adoption correlates with faster productivity in high-adoption industries — still a micro signal, not macro breakout.→ Probability: 50%
- Aug 8, 2026
W32: Bloomberg (Aug 7) counted ~25 S&P 500 companies that used Q2 calls to attribute margin expansion to AI, averaging ~180bps — its "the AI profit boom has begun" framing, and the first cluster of named non-tech firms (down to a garbage-truck operator) claiming a hard AI margin line. Two reasons it does not move the topic: the figures are management-selected and unaudited (Benchmark Theater applied to the P&L, not model evals), and margin is profit, not the BLS nonfarm-productivity series the resolution measures — "gains captured as profits not productivity" is the bear case, not its refutation. No new BLS Productivity & Costs print this week. Hold.→ Probability: 52%
- Sep 16, 2026
The FOMC statement accompanying its 25bp hike to 3.75-4.00 percent asserts that productivity growth is strong and capital investment robust, but it makes no attribution to AI and is a qualitative characterization rather than the measured BLS series. Against the refuted_if bar of above 3 percent annually for two consecutive quarters with AI clearly cited, it fails both legs - the underlying prints are 1.4 and 0.3 percent. Logged as context, not as qualifying evidence. https://www.federalreserve.gov/newsevents/pressreleases/monetary20260916a.htm
How Our View Evolved
- Oct 10, 202656%→56%
W41 rotation, HOLD at 0.56. No new BLS print until November 5; corrected our record of the quarterly figures (Q4 2025 1.6, Q1 2026 0.8, Q2 2026 1.4). The refuted_if clause has no start date and mid-2025 already meets its numbers, so a criteria fix is flagged.
- Sep 19, 202652%↑56%
W38 rotation, +0.04 to 0.56. The Fed's adjective and the BLS series disagree, and the series is what this topic resolves on. The September 3 Productivity and Costs release put nonfarm business productivity at +1.4 percent annualized and +2.2 percent year over year - the third consecutive quarter below the 2.5 percent threshold, with the November 5 print able to complete the four-quarter leg the criterion requires. The FOMC's September 16 claim that productivity growth is strong carries no AI attribution and is not the measured series, so it was logged as context and did not move the number. Only +0.04 because 2.2 percent clears the bar by just 0.3pp and revisions in this series routinely exceed that - we have watched Q4 2025 go from 2.8 to 1.8 and Q1 2026 from 0.8 to 0.3 - and because 2.2 percent remains well above the decade average, which is the take-off-is-visible case in the other direction.
- Aug 8, 2026Initial assessment: 52%
W32 rotation — HOLD at 0.52. Bloomberg counted ~25 S&P 500 firms attributing ~180bps of Q2 margin expansion to AI ("the AI profit boom has begun"), the first cluster of named non-tech names claiming a hard AI margin line. But the attributions are management-selected and unaudited, and margin is profit not measured productivity — firm-level profit gains that never reach the BLS nonfarm series are the paradox, not its refutation. No new Productivity & Costs print this week. Coin flip holds.
- Aug 2, 2026Initial assessment: 52%
W31 rotation — HOLD at 0.52. Big Tech guided ~$725B of 2026 AI capex (+77% YoY) yet Meta/Amazon sold off on the "what's the return" question and Amazon's TTM FCF went to -$7.6B; Cognizant cited IDC's 88%-of-pilots-fail figure. Offset by real scaled AI revenue (Microsoft $37B run rate, CCC $120M) and no new BLS Productivity & Costs print this week. The spend-vs-measured-output gap is exactly the paradox; coin flip holds.
- Jun 13, 202650%↑52%
W24: BLS revised Q1 2026 productivity to +0.3% (from +0.8%) with TFP decelerating 1.5%→0.8% over 2024-2025, even as 2026 data-center capex crossed $1 trillion. Investment surging, measured productivity lagging — the paradox in hard data. +0.02.
- Apr 10, 2026Initial assessment: 50%
Rotating refresh — HOLD at 0.50. BLS Q4 revised 2.8%→1.8%, FY2025 at 2.1%. Weakens visible AI signal but paradox neither confirmed nor refuted. Coin flip holds. Revisit after Q1 2026 BLS release (May 7).
- Mar 8, 2026Initial assessment: 50%
Baseline — initial published assessment
What Experts Say
Sam Altman
CEO, OpenAI
“30-40% of current economic tasks will be done by AI in the not-very-distant future”
Mustafa Suleyman
CEO of Microsoft AI
“Most professional tasks involving sitting at a computer will be fully automated by AI within 12-18 months”