AI and Productivity: Can Artificial Intelligence Revive Economic Growth?
Economic growth has slowed across the developed world. Productivity — the engine of rising living standards — has been disappointing for two decades. The 2008 financial crisis, demographic ageing, and slowing technological diffusion have combined to produce what economists call “secular stagnation”: a world of structurally lower growth. Now, proponents of AI argue it could break this pattern entirely — delivering a productivity shock large enough to revive economic growth at a scale not seen since the post-war boom. Sceptics argue the evidence so far doesn’t support the hype. This article examines both sides rigorously, as part of our series on AI and the economy in 2026.
- → Goldman Sachs estimates AI could boost global GDP by 7% — roughly $7 trillion — over the next decade if adoption is broad and sustained
- → Early microeconomic studies show dramatic productivity gains in specific tasks: software engineers code 56% faster with AI assistance; customer service agents handle 14% more cases
- → Macroeconomic productivity data has not yet reflected these gains — mirroring the “productivity paradox” of the early computer era
- → The critical variable is diffusion speed: how quickly firms and workers across all sectors actually adopt and integrate AI tools into their workflows
- → The productivity gains from AI, if realised, would not automatically translate into broadly shared prosperity — distribution depends on policy choices
The Productivity Crisis AI Is Trying to Solve
To understand AI’s potential economic impact, it helps to understand the problem it is being asked to solve. US total factor productivity growth — the broadest measure of economic efficiency — averaged around 1.8% annually from 1948 to 2004. Since 2005, it has averaged less than 0.5%. European figures are similar. This slowdown, compounded over decades, explains much of why living standards have improved more slowly than previous generations expected.
The Microeconomic Evidence: Striking Early Results
The most compelling evidence for AI’s productivity potential comes from controlled studies of specific work tasks. These results are consistent and striking. A 2023 MIT study of software engineers found that those using GitHub Copilot completed coding tasks 56% faster — without any detectable reduction in quality. A Stanford/MIT study of customer service workers found that AI assistance led to 14% more cases resolved per hour, with the biggest gains going to the least experienced workers (who effectively got AI to bring them closer to expert-level performance instantly).
A Harvard Business School study of management consultants at BCG found that those using AI outperformed their peers on analytical tasks by 25%, on creative tasks by 40%, and on writing quality by a significant margin. Critically, this was the average — not just the top performers. AI appeared to compress the performance distribution, raising the floor more than the ceiling.
“The most profound finding from early AI productivity research is not that the best workers get better — it’s that the average workers get dramatically better. AI is a great equaliser of human capability.”
The Macroeconomic Puzzle: Why Don’t We See It in GDP?
If individual workers are becoming dramatically more productive, why hasn’t this shown up in aggregate economic data? This is not a new puzzle. Robert Solow noted in 1987 that computers could be seen everywhere except in the productivity statistics. It took until the mid-1990s — roughly two decades after widespread computer adoption — for the productivity gains to become visible in macroeconomic data.
Economic historians Erik Brynjolfsson and Paul David have documented why general-purpose technologies take 15–25 years to show up in productivity statistics. Realising the gains requires complementary investments: reorganising workflows, retraining workers, redesigning business processes, and building the supporting infrastructure. A factory with an electric motor but organised for steam-era work is not much more productive. The same logic applies to AI — a firm with access to AI tools but unchanged processes and skills captures only a fraction of the potential.
The implication is cautiously optimistic: the productivity gains from AI may be real and large, but concentrated in the late 2020s and 2030s rather than visible today. The trillion-dollar infrastructure investment currently underway is the equivalent of building the electricity grid — a precondition for productivity gains that will arrive years later. This connects directly to the macroeconomic themes in our global economy overview for 2026.
The Conditions Required for the Productivity Dividend
The 7% GDP boost scenarios from Goldman Sachs and others are not predictions — they are conditional projections. They assume several things that are not guaranteed:
| Condition | Current Status | Probability Assessment |
|---|---|---|
| Broad AI adoption across sectors | Early stages — concentrated in tech | High over 10 years; uncertain pace |
| Complementary organisational change | Minimal so far at most firms | Moderate — requires management intent |
| Workforce upskilling at scale | Patchy — dependent on education systems | Moderate — significant policy variable |
| AI models continuing to improve | Strong — scaling laws still operating | High near-term; uncertain long-term |
| No major regulatory disruption | Regulatory pressure building in EU, US | Moderate — EU AI Act already in force |
Distribution: Who Captures the Productivity Gains?
Even if AI delivers a genuine productivity surge at the macroeconomic level, the distribution of those gains is not automatic. In the 1990s productivity boom driven by computers and the internet, the gains were relatively broadly shared — real wages rose across income groups. But the decades since have seen productivity gains increasingly captured by capital rather than labour, widening the wealth gap. Whether AI continues and accelerates this trend is one of the most important policy questions of the coming decade. We examine it in depth in our article on AI and Inequality: Will AI Widen the Wealth Gap?
The productivity case for AI is real and supported by early evidence — but the path from individual task gains to macroeconomic GDP growth is long and contingent on choices that haven’t been made yet. The most likely scenario is a genuine productivity dividend arriving in the late 2020s and 2030s, concentrated first in high-adoption sectors and gradually diffusing more broadly. For investors, this supports a long view on AI infrastructure and the application layer — with the caveat that timing the productivity payoff is genuinely difficult.
Responses