About
Paul Kedrosky is an investor and writer who has become one of the most cited skeptics of the AI spending boom. He started as an engineer, then worked in sales, then as a Wall Street technology analyst, founding the technology equity research practice at HSBC James Capel and becoming one of the first analysts to cover Internet companies. He holds a Ph.D. in the economics of technology, with research on risk, adoption and path dependence, and was a senior fellow at the Kauffman Foundation. He is a partner at the early-stage firm SK Ventures and a research fellow at MIT studying the digital economy. In 2025 his analysis comparing AI data-center spending, as a share of GDP, to the railroad boom of the nineteenth century became a reference point in the debate over whether AI is a bubble.
Key Contributions
- Compared AI data-center capital spending to the nineteenth-century railroad boom as a share of GDP, arguing that, unlike rails, the chips wear out within years
- Co-wrote 'Society's Technical Debt and Software's Gutenberg Moment' with Eric Norlin, arguing that collapsing software costs will release a backlog of unbuilt software
- Founded the technology equity research practice at HSBC James Capel, among the first analysts to cover Internet companies
- Senior fellow at the Kauffman Foundation, researching entrepreneurship, innovation and risk capital
- Edited the long-running business blog Infectious Greed, and has been an on-air contributor for CNBC and Bloomberg
Close Listening
Videos & Interviews
Connections
Azeem Azhar
DebatedFounder, Exponential View
On The Ezra Klein Show in October 2026, on this site with a Close Listening, they took opposite sides of whether AI is a bubble, and the disagreement turned out to be about what to count. Kedrosky counts capital and fragility: 'a bubble is whenever asset prices detach from any semblance of economic reality,' and every general-purpose buildout from canals to fiber has overbuilt. Azhar counts customers, calling it a productive boom 'with the hints of maybe getting… a little bit stinky.' Even their closing tests split the same way: enterprise spending for Azhar, productivity and GDP for Kedrosky.
youtube.com · nytimes.com
Daron Acemoglu
KindredInstitute Professor of Economics, MIT
An interpretive pairing of two skeptics who reach the same number by different roads. Acemoglu's 'Simple Macroeconomics of AI' estimates no more than a 0.66% gain in total factor productivity over ten years; Kedrosky, an investor watching the national accounts, says of the recent productivity rebound 'Could it be AI…? Almost certainly not—too soon,' and told Ezra Klein there is 'no sign of AI improving productivity.' Both make productivity the test the AI economy has yet to pass, which is why Kedrosky named it as the thing that would prove him wrong.
nber.org · paulkedrosky.com · youtube.com
Dario Amodei
In contrastCEO & Co-founder, Anthropic
An interpretive contrast over one phrase. Amodei's essay 'We Must Pace the Frontier' asks that the rate of capability advance slow, verifiably, and insists pacing is not halting; Kedrosky, on The Ezra Klein Show, calls it 'a ridiculous expression' that 'sounds vaguely like we're training for the Boston Marathon,' and answers with the prisoner's dilemma: everyone says 'you slow down. I'm not slowing down.' The pairing is the hinge of this site's thread on pacing the frontier: whether restraint can survive a race that capital keeps funding.
darioamodei.com · youtube.com
