A few weeks ago I watched a lecture by Chad Jones, an economist at Stanford, on AI and long-run economic growth. He opened with a simple chart: US income per person since 1870, plotted on a log scale. A straight line. Around 2 percent annual growth for 150 years.[1]
That line is remarkable because of what it hides. Railroads. Electricity. The internal combustion engine. Semiconductors. The internet. Each one was supposed to change everything. And each one, eventually, did. But the aggregate growth rate barely moved. The line stayed straight.
How is that? One thought is the counterfactual: without those innovations, income would have stagnated. The 2% line wasn't the ceiling; it was the floor. Each technology prevented the plateau that would have come without it.
That reframe matters. If prior technologies merely kept us at 2% rather than lifting us above it, the right question isn't whether AI continues the trend. It's whether AI is the first technology capable of bending it.
If the counterfactual view is right, AI simply allows the 2% to continue. That's the bear case, and it deserves respect. It draws on a serious observation: past technologies took decades to move the needle on aggregate productivity. Railroads required laying track. Electricity required wiring an entire nation. The physical constraints of deployment smoothed the impact across decades and blended it into the long-run trend.
My base case is different, and the distinction is structural. AI is software. It deploys across existing networks at near-zero marginal cost, without waiting for a grid to be built or a right-of-way to be cleared. The constraint isn't infrastructure; it's adoption. And adoption curves for software are not linear.
This compression is the key variable. Consider what happened to communication when email replaced postal mail. The productivity gain wasn't new; faster decisions and lower coordination costs had always been valuable. What changed was the speed of delivery. The same output arrived in seconds instead of days, and the economic impact landed in years rather than decades. AI is that dynamic applied to cognitive work broadly: the same gains that would have taken a generation to accumulate arrive inside a single business cycle. The numerator stays the same; the denominator shrinks. Growth spikes. That's not a forecast; it's math.