The numbers should not coexist. Global AI investment surpassed $500 billion in 2025. Enterprise AI adoption rates exceeded 72% across OECD economies. Yet total factor productivity growth — the metric that captures how efficiently economies convert inputs into outputs — has barely moved. This is not a new phenomenon. It is the defining measurement failure of the AI era, and understanding it is essential for anyone allocating capital to AI-enabled businesses.
The original AI productivity paradox explored why trillions in spending are not moving aggregate statistics. This article goes deeper into the measurement problem itself: why conventional economic metrics are structurally incapable of capturing AI-driven value creation, and what alternative approaches reveal.
The productivity paradox is not evidence that AI does not work. It is evidence that our measurement systems were designed for an industrial economy and have not been updated for an intangible one.
Consider what GDP actually measures. It captures the market value of final goods and services produced within a country. When a factory produces more cars per worker, GDP registers the improvement directly. But when an AI system reduces customer churn by 15%, improves decision quality across a management team, or enables a company to enter new markets faster — these gains are distributed, indirect, and often invisible to national accounts.
The Bureau of Economic Analysis acknowledged this gap in its 2024 framework review, noting that "the current GDP framework may systematically undercount output improvements driven by AI and software." The problem is not that AI fails to create value. The problem is that value creation has shifted to dimensions that existing metrics were never designed to capture.
The AI productivity paradox is primarily a measurement failure, not a technology failure. Conventional metrics like TFP and GDP were designed for tangible output and struggle to capture quality improvements, decision accuracy, and intangible value creation driven by AI.
When an AI-powered diagnostic tool reduces medical misdiagnosis rates by 30%, the healthcare system produces the same number of consultations at the same price — but with dramatically better outcomes. GDP registers zero improvement. The patient benefits enormously, but the economic statistics are blind to it.
This quality-adjustment problem is pervasive. AI-driven improvements in fraud detection, customer service resolution, supply chain optimisation, and product recommendations all improve quality without necessarily increasing measured output.
The most widely used AI applications — search, translation, writing assistance, image generation — are either free or nearly so. They create enormous consumer surplus that GDP cannot capture because there is no market transaction to measure. Brynjolfsson and Collis (2019) estimated that the consumer surplus from free digital services exceeds $300 billion annually in the US alone.
When firms invest in AI, much of the spending creates intangible assets — proprietary models, training data, organisational knowledge, process improvements — that compound over time. These assets do not appear on balance sheets under current accounting standards (see IAS 38 recognition criteria), and their productive contribution is attributed to other factors in growth accounting frameworks.
A private equity firm invests £2 million in AI-powered deal sourcing. The system identifies three additional acquisition targets that the team would have missed, leading to £50 million in additional deployed capital. The AI investment is expensed on the income statement. The deal flow improvement is attributed to "management skill." The AI's contribution is invisible in every standard metric.
Historical technology transitions suggest a 10-15 year lag between major technology deployment and measurable productivity gains. Electricity was commercially available in the 1880s but did not drive aggregate productivity growth until the 1920s, when factories were redesigned around electric motors rather than steam-era layouts. Personal computers were ubiquitous by the late 1980s but did not contribute to measured productivity until the late 1990s.
| Technology | First commercial deployment | Measurable productivity impact | Lag (years) |
|---|---|---|---|
| Steam power | 1780s | 1830s-1840s | 50-60 |
| Electrification | 1880s | 1920s | 30-40 |
| Personal computing | 1980s | Late 1990s | 15-20 |
| Internet/e-commerce | Mid-1990s | 2000s | 10-15 |
| AI/ML | 2020s | TBD | Est. 10-15 |
The AI J-curve is likely to be at the shorter end of this range because AI builds on existing digital infrastructure rather than requiring entirely new physical systems. But the complementary investments required — workforce retraining, process redesign, data infrastructure, governance frameworks — take time regardless.
The J-curve is not uniform across firms or sectors. Early adopters with strong data assets and digitally mature operations are already seeing measurable gains. The paradox appears at the macro level because these gains are diluted by the majority of firms still in the investment phase.
If conventional metrics cannot capture AI value creation, what can? Several alternative measurement approaches paint a more nuanced picture.
When researchers examine individual firms rather than aggregate statistics, AI's impact becomes visible. A 2025 Stanford study found that firms with mature AI deployments (3+ years) showed 12-18% higher revenue per employee than industry peers. The National Bureau of Economic Research found that AI adoption in customer service roles increased resolution rates by 14% while reducing handling time by 25%.
The gap between firm-level evidence and macro statistics suggests an aggregation problem: AI creates measurable value at the firm level, but the gains are unevenly distributed and offset by transition costs at firms still adopting.
The Corrado-Hulten-Sichel framework provides a broader view of investment by including spending on software, R&D, design, training, and organisational capital. When AI spending is measured through this lens rather than through conventional capital expenditure accounts, the investment-to-output picture looks less paradoxical.
The measurement gap has practical consequences for capital allocation. If conventional metrics understate AI's contribution, then companies with genuine AI capability may be undervalued by traditional analysis. Conversely, companies that rely on AI hype without substantive capability — AI washing — are overvalued.
For private equity firms and institutional investors, this creates both risk and opportunity. The risk is misallocating capital based on metrics that do not reflect reality. The opportunity is developing proprietary measurement frameworks that capture what public statistics miss.
Three practical steps for investment professionals:
The Opagio Growth Platform provides tools for measuring intangible asset accumulation and benchmarking AI capability across portfolio companies, helping investors see what conventional metrics miss.
The AI productivity paradox is real, but it tells us more about our measurement systems than about AI's actual value. Investors who wait for macro statistics to confirm AI's impact will miss the opportunity. Those who develop alternative measurement frameworks — capturing intangible capital accumulation, firm-level efficiency gains, and quality improvements — will have a structural advantage in capital allocation decisions.
David Stroll is Co-Founder and Chief Scientist at Opagio. His research focuses on productivity measurement, intangible capital, and the economic impact of AI adoption. He holds a PhD in productivity economics and previously led research at Digital Equipment Corporation. Learn more about the Opagio team.
Global AI investment has surpassed $500 billion annually, yet total factor productivity growth remains stubbornly flat. This is Solow's paradox reborn: you can see AI everywhere except in the productivity statistics. History tells us why — and what needs to change.
Read more →
Traditional productivity metrics were designed for an economy dominated by physical capital and tangible output. In a world where the most valuable firms derive their competitive advantage from software, data, brand equity, and organisational know-how, those metrics are increasingly unreliable. Here is why the gap matters — and what we should do about it.
Read more →
The San Francisco Fed's recent research on AI and productivity is among the most rigorous work in the space. But it also reveals critical blindspots in how economists measure the impact of AI on economic growth. Here is what they got right, what they missed, and why the measurement gap itself is the most important finding.
Read more →Get the latest insights on intangible asset growth and productivity delivered to your inbox.
Book a free consultation to see how the Opagio Growth Platform can help your business.