What follows (see below) was written by Paul Kedrosky, a fellow blogger back in the days when people blogged on Blogger. Paul is an investor in private and public companies, as well as a writer and researcher. D.
What follows (see below) was written by Paul Kedrosky, a fellow blogger back in the days when people blogged on Blogger.
Paul is an investor in private and public companies, as well as a writer and researcher. Originally trained as an engineer, he went on to do a Ph.D. at the University of Western Ontario in Canada, where he researched aspects of the economics of technology—specifically, the role of path dependency and network effects in risk & complexity.
Paul regularly speaks at private and public events across the U.S. and around the world, usually on topics related to risk finance, economics, the future of work, and artificial intelligence. He is currently a research fellow at MIT’s Institute for the Digital Economy, where he is studying artificial intelligence, economic disruption, and the future of work.
You can find his daily note at paulkedrosky.com. We urge you to subscribe. It is always worth reading. Always.
By Paul Kedrosky:
As long-time capital markets people, we think in terms of “house views”: What we believe, and what thus drives our interpretation of events. These can and do evolve, but it is also important to know them so one can be clear when one is drifting away, basking in the warm seas of confirmation bias, or seeing actual falsification.
And, to be clear, we delight in discovering when we are wrong. Our greatest fear is being wrong and never finding out, so we aggressively look for falsifying data. As the writer Kathryn Schulz wrote in Being Wrong, her classic book on the topic, being wrong feels exactly like being right—until you find out you are wrong
So, with thousands of new subscribers, heading into Q4 2026 feels like a good time to lay out our house views—subject to evolution, revision, falsification, and the odd moment of tub-thumping triumphalism.
Here is the house view, in plain language, bulleted for easy scanning. We will revisit in the future as events evolve.
**AI is real and consequential.**Adoption is extraordinarily fast, many productivity gains are real, and the technology will matter enormously.**The current AI boom is wildly distorting the economy.**Massive spending on chips, data centers, power and networking is now large enough to essentially be GDP growth, to distort profits, make trade data useless, break electricity grids, and distort capital spending.**Headline economic data is misleading.**Without AI capex and its growing externalities—and recently energy shocks—the U.S. economy would look much weaker, plausibly even contracting.**The boom is debt-financed.**AI infrastructure is no the largest borrower in investment-grade and high-yield markets, which is causing spiking funding pressure in other markets, from corporate to sovereigns.**The basic financial mismatch is ugly:**companies are borrowing long-term to build infrastructure producing a commodity—compute/tokens—whose performance-adjusted price is falling 70-80% year-over-year. Usage therefore has to rise hundreds of percent into perpetuity to stand still, let alone produce Wall Street growth while funding towering debt.**This can be true even if AI succeeds technologically.**The internet was transformative; telecom investors still destroyed vast amounts of capital. Great technology does not imply good infrastructure economics. It generally implies the opposite: back to fiber, railroads, canals, etc.**AI capex is crowding out other investment.**Corporate cash flow can look strong partly because firms are delaying ordinary capex, having been squeezed out, while an unusually concentrated group pours money into AI.
**Near term, AI can be inflationary.**Data centers bid up electricity, equipment, construction, financing and scarce infrastructure. Those inflationary pressures are poorly addressed by higher interest rates.**Underneath that, AI is profoundly deflationary.**Software, cognitive work, and many intermediary services become cheaper; labor demand in many high-wage occupations will weaken; competitive pressure rises as what was once hard to create becomes easy, from wars to apps; and falling inference costs propagate through the economy.**The labor effect probably arrives more through hiring than layoffs.**Entry-level pipelines shrink, firms hire fewer people, and productivity gains allow output to rise without comparable growth in employment. This is analogous to what happened in prior automation waves, from office automation, to factory automation, to the Industrial Revolution.**That creates “ghost GDP.”**GDP can rise because enormous amounts of capital are being spent while wage income, labor share, and household demand weaken. The economy looks healthier in aggregate than it feels underneath, as gains flow increasingly to capital vs. labor.**The dangerous phase comes when the capex boom slows.**If AI investment has been masking underlying weakness, a slowdown removes the stimulus just as overcapacity, falling prices, refinancing needs and weaker labor income hit together. For example, a refinancing wall sits waiting in 2028.Failure is over-determined. This moment sits at the intersection of the four forces that have produced the most consequential boom-bust cycles in Western economic history: loose credit, government policy, technology, and real estate. There are, as a result, myriad catalysts, each with a low probability of happening, but which collectively guarantee collapse.**That is why the principal risk is deflationary rather than inflationary.**An AI bust could transmit through capex, employment, consumption, credit, housing, and asset prices simultaneously. The other side will plausibly see a decade of Japan-style balance recession (see Richard Koo’s work) just as Western economies can’t further stimulate via fiscal policy, given massive deficits.**AI safety has a systems problem.**These are stochastic systems that are grown through training rather than engineered like chips. Treating safety as conventional deterministic testing misses emergent behavior, collective-action problems, and correlated failures among interacting agents. Regulation will likely come after material losses, not before.
The short version:
AI may be the most important technology in decades, while the financial structure being built around it is the largest and most unstable capital spending boom in modern economic history. Its spending is currently holding up the economy; its eventual productivity effects are deflationary; and the inevitable transition between those two regimes is where the real macro risk lies.
**The September 25, 2026 edition of Mr. Kedrosky’s newsletter is reposted above, in its entirety, with his expressed permission. We do this from time to time because the quality of his insights and analyses are consistently first-rate. Again, if you’re not a subscriber, we urge you to become one. **
**See you tomorrow. **
Send this story to anyone — or drop the embed into a blog post, Substack, Notion page. Every play sends rev-share back to News Items.
We’ve simplified responses to 👍 / 👎. Past comments are archived but no longer visible.