The AI Productivity Dividend: Still a Promise, Not a Paycheck

by policies, tech1 Comment 6 min read

Nearly four years into the generative AI boom, the question has shifted from “will AI transform productivity?” to “where, exactly, is the productivity?” This month, the Bank of England delivered one of the bluntest official answers yet — and the numbers are sobering.

The Bank of England’s verdict

In its November 2026 Monetary Policy Report, the Bank of England found that despite the billions being poured into AI, the technology has delivered little in the way of measurable productivity gains for UK businesses. AI has increased UK productivity by an estimated 0.5% over the past three years — just 0.1 to 0.2 percentage points of additional annual productivity growth since early 2024. As the Bank put it: “AI investment appears to be contributing somewhat to current demand strength, but so far there is only evidence of a modest AI-driven rise in UK productivity growth.”

Even industries supposedly best placed to benefit have little to boast about. Productivity growth has improved in AI-heavy sectors such as information and communications, but has yet to exceed pre-pandemic averages.

That’s not to say AI is doing nothing for the British economy. AI-related activity accounts for just under a fifth of UK GDP growth since the start of 2026 — but much of that comes from companies building and selling AI rather than businesses becoming more productive by using it. “The contribution to growth from AI to date appears to reflect activity in the upstream development and implementation of AI technologies rather than broader gains from their adoption elsewhere in the economy,” the report said. In other words, selling AI is proving rather more lucrative than actually using it.

The Bank also cautioned against giving AI too much credit for Britain’s recent productivity improvements: much of the increase appears to reflect businesses hiring fewer workers while output continues to grow — partly an unwind of post-pandemic over-hiring and a response to higher employment costs — rather than genuine AI-driven efficiency.

The costs arrive before the benefits

There’s a certain irony in the sequencing: while the productivity windfall remains theoretical, the costs of the boom are already flowing through the economy. The Bank noted that surging demand for AI-related products — memory chips in particular — has pushed up export prices in China and other Asian economies, raising import costs for Britain. This matters for monetary policy: AI spending pushes prices up now, while AI only helps bring inflation down later if it genuinely makes workers more productive. The MPC voted 8–1 to raise Bank Rate to 4% at its November meeting.

The gap between expectation and execution

The Bank of England isn’t an outlier. A Gartner survey published in October 2026 found that while three-quarters of corporate board members expected AI to improve productivity and cut costs, just 3% of organizations had actually achieved their targets. Meanwhile, 38% were already over budget on AI spending — yet 85% planned to spend even more next year. Another study of nearly 6,000 executives, published in February, found that more than 80% of firms had seen no discernible impact from AI on either employment or productivity. Apparently, the answer to disappointing returns is to keep writing checks.

The growing disconnect between AI spending and actual returns has caught the attention of financial watchdogs. The International Monetary Fund warned last month that the AI spending spree could come back to bite investors if the promised profits don’t show up, potentially dragging the wider economy down with it. The Bank for International Settlements sounded a similar alarm in June, comparing the current spending spree to the dotcom boom of the 1990s: tech giants are committing enormous sums to infrastructure on the assumption that AI will eventually deliver equally enormous returns, leaving them exposed if reality fails to cooperate.

But the labour-market signal is real

Interestingly, the aggregate employment numbers hide a sharper story at the edges. Stanford Digital Economy Lab data shows employment for workers aged 22–25 in the most AI-exposed occupations fell 16% between October 2022 and September 2025. Firms may not be firing en masse, but they are noticeably less inclined to hire new staff or backfill departures in AI-exposed roles. The productivity revolution may show up first not as output per hour, but as jobs that quietly stop being created.

The forecast: optimism deferred, not abandoned

The Bank isn’t writing off the technology just yet. Businesses surveyed expect AI to add around 0.9 percentage points to annual productivity growth over the next three years, with the gains split roughly equally between faster output growth and slower employment growth. The dividend, in other words, is still coming — someday. It seems the AI industry’s most impressive productivity achievement so far is convincing businesses to keep opening their wallets.

My take: this is what a general-purpose technology transition actually looks like

None of this should be surprising. Electrification and the internet followed the same pattern: massive capital expenditure, inflated expectations, a painful shakeout, and — eventually — real productivity gains that arrived decades after the hype peaked. The “productivity J-curve” is well documented: measurable output often dips or stagnates while firms reorganize workflows, retrain workers, and rebuild processes around the new technology.

Three things seem genuinely different this time, though:

  1. The speed of capital deployment. Never before have so many hundreds of billions been committed so quickly on such speculative returns. The BIS’s dotcom comparison is apt — and the dotcom crash was real, even though the internet’s long-term promise was also real.
  2. The costs are inflationary now. Chip demand is already raising consumer prices in import-dependent economies. In previous technology transitions, costs and benefits arrived together; here, the macro costs are front-loaded.
  3. The entry-level labour effect. The Stanford data suggests AI’s first measurable labour-market impact is compressing the bottom rung of the career ladder — a distributional problem, not just a productivity one.

The honest read of the current state: AI is an extraordinary technology embedded in an ordinary adoption timeline. The 3% of organizations hitting their targets are probably the ones doing the unglamorous work — redesigning processes, measuring outcomes, integrating with legacy systems — rather than buying their way to transformation. For everyone else, the Bank of England’s message is worth taking seriously: the productivity dividend is a forecast, not a fact, and right now the invoices are arriving faster than the returns.

Sources