Keynesian Economics
The Ghost in the Machine: Why AI Wealth Isn't Showing Up in the Economy
Trillions are being poured into artificial intelligence, and by almost every measure that matters to an economy, it's as if the money vanished.
The Idea
Here is the puzzle in plain numbers. By 2026, global investment in AI infrastructure had surpassed a trillion dollars over six years, with the five biggest technology companies alone projected to spend roughly 600 billion in 2026 — a 36 percent jump from the year before. Yet a March 2026 survey of around 6,000 executives found that roughly 90 percent of firms reported zero measurable impact on productivity or employment from AI over the preceding three years. At the individual level, meanwhile, studies consistently find AI lifting task-level performance by anywhere from 14 to 55 percent, depending on the work. The macro data and the micro experience seem to describe two different worlds. This is what economists call the productivity paradox, and it has a distinguished ancestor. In 1987, Robert Solow made what became the most-quoted observation in technology economics: 'You can see the computer age everywhere but in the productivity statistics.' The same mismatch haunts AI in 2026. Apollo's chief economist Torsten Slok put it bluntly: AI is nowhere in the employment data, the productivity data, or the inflation data. The Keynesian relevance here is easy to miss. Keynes was not just interested in demand shortfalls; he understood that investment creates growth only when it flows through the economy and lifts what people can produce. If a transformative technology eats capital without raising output, the multiplier effect stalls — and with it the broader growth story that justifies the spend in the first place. The question of whether AI will show up in growth is therefore not merely a technology question. It is a question about how economies convert investment into lasting improvement.
In the World
The clearest way to see the paradox is through the history of electricity. Edison's Pearl Street Station — the first commercial electrical distribution plant in the United States — opened in New York in 1882. By 1900, the technical case for electricity was settled, but electric motors still supplied less than 5 percent of the mechanical drive power in American factories. Manufacturers had simply swapped steam engines for electric motors without changing the fundamental layout of their buildings. Productivity gains were negligible. It took until roughly the 1920s — four decades after the first power stations — before electrification began lifting manufacturing output in a measurable way. The reason, as economic historian Paul David argued in his landmark 1990 paper 'The Dynamo and the Computer,' was organizational: factories eventually realised that electricity's true advantage was distributed power, which allowed the entire production floor to be redesigned around the flow of work rather than around the location of the engine. The technology had not changed. The thinking around it had. Stanford economist Erik Brynjolfsson, who has spent his career studying exactly this kind of lag, calls the AI version the 'productivity J-curve': investment rises sharply, disruption follows, measured output initially dips, and then — if firms do the organizational work — productivity climbs. In February 2026, Brynjolfsson wrote that the AI productivity take-off was finally becoming visible in aggregate data, citing a US productivity growth figure of around 2.7 percent for 2025. But the picture remains contested: a McKinsey survey published in August 2026 found that 80 percent of workers say AI has improved their personal productivity, while only 37 percent report any positive contribution to their organisation's bottom line. The gap between feeling faster and being measurably more productive is itself a form of the paradox.
Why It Matters
Understanding this lag reframes almost every conversation you will hear about AI and the economy. When someone tells you AI is revolutionising growth, ask them to point to the productivity statistics. When someone tells you AI is overhyped and useless, remind them that electricity was dismissed the same way in 1900. Both positions miss the actual dynamic: transformative technologies require complementary reorganisation before they show up in the numbers, and that reorganisation is slow, expensive, and invisible in quarterly reports. For anyone making decisions — about careers, about where to invest savings, about which industries are genuinely being transformed — this matters enormously. The productivity payoff from AI, if it arrives, will not be evenly distributed. The Atlanta Fed found in 2026 that expected gains vary sharply across sectors, and earlier research by Brynjolfsson found AI disproportionately lifts the output of less experienced workers while doing little for top performers. That asymmetry is a signal about where value will migrate. The deeper Keynesian lesson is that investment alone is never enough. What counts is whether that investment reshapes how work is done — not just what tools sit on the desk.
A Question to Ponder
If the electricity paradox took four decades to resolve because factories were redesigned around power rather than just electrified, what is the equivalent redesign that would need to happen in your industry before AI shows up in the numbers?
One reader's Thinkable for Finance Friday — a random draw from the library.
Start learning with Thinkable