The AI Bubble Is Widely Misunderstood | Steve Hou

Episode Summary

Executive Summary: Steve Ho argues AI is unquestionably a bubble, but one with unusually large and lasting macro effects because adoption is immediate and agentic AI can multiply compute demand. He sees the current impact mostly in capex, GDP, and markets, while labor-productivity gains are still hard to isolate. He’s skeptical the Fed should preemptively cut on AI disinflation hopes, but bullish that AI will improve economic modeling and policy analysis.

Main Topics: Why AI is a bubble but still economically meaningful (Priority: 5/5): Ho says the AI boom qualifies as a bubble, yet that label should not dismiss its power: bubbles can last a long time and have large real-economy consequences. He contrasts AI with dotcom by emphasizing immediate adoption and direct demand for compute. The macro impact of AI capex and data-center buildout (Priority: 5/5): He argues the biggest near-term GDP effect comes from the buildout itself—data centers, chips, memory, and related infrastructure—rather than from broad productivity gains. This spending has supported U.S. growth while also boosting chip-exporting economies like Korea and Taiwan. Agentic AI as a compute-demand accelerant (Priority: 5/5): Ho’s core thesis is that recursive AI usage—AI calling AI—can raise compute demand by orders of magnitude. He says coding agents are a key reason markets underestimated the latest acceleration, since many observers do not code and therefore miss the workload explosion. Labor productivity: real gains or composition effects? (Priority: 4/5): He is skeptical that current aggregate productivity data cleanly captures AI productivity. Instead, he sees composition bias, overhiring corrections, and capital-intensive investment effects as likely drivers, though he expects cleaner evidence to emerge over time. Fed policy and the limits of preemptive AI disinflation bets (Priority: 4/5): Ho doubts the Fed should justify rate cuts on anticipated AI-driven disinflation. He says inflation effects would be slow, sector-specific, and visible mainly through wages and services; direct inflation from AI-related capex may be more immediate than any productivity relief. Future of AI pricing and token efficiency (Priority: 4/5): He thinks AI pricing is too cheap and subsidies are distorting usage. He expects a more segmented pricing structure to emerge, with auctions or tiering for powerful models and greater emphasis on token efficiency and compute allocation. AI’s role in economics research and policy modeling (Priority: 3/5): Ho is optimistic that AI will improve economic analysis through richer models, agentic simulations, and faster adaptation of ideas across fields. He also thinks AI can improve communication of policy and help economists work more creatively with tools they only partially understood before.

Key Arguments: AI is a bubble, but the better question is not whether it exists; it is how large and how long-lasting it will be, because bubbles can still produce major structural change. Unlike the dotcom era, AI is being adopted almost immediately, so investment is tied to real usage rather than unused capacity. The main near-term macro effect is capex: data centers, chips, memory, and infrastructure are already materially contributing to GDP and supporting growth. Agentic AI and recursive tool use can dramatically increase compute demand, potentially by 100x or more, especially in coding workflows. Most people underestimate AI’s demand because they do not code and therefore do not appreciate how much work agentic systems create. Observed productivity gains are likely mixed with composition effects, reduced hiring, and post-COVID labor normalization, so aggregate data are not clean evidence of AI-driven efficiency. The Fed should not preemptively cut rates on the assumption that AI will soon disinflate the economy; any effect would be delayed and should show up first in wages and services. AI is currently too cheap, and pricing will likely evolve toward tiered or auction-based models to match scarce compute with demand. AI will likely improve economics and finance by enabling richer modeling, more realistic simulations, and better policy communication.

Data Points: ChatGPT launch timing: Late 2022 - Used as the starting point for the AI capex and adoption cycle PhD completion year: 2018 - Steve Ho said he earned his PhD in macro and financial econometrics at the University of Michigan in 2018 AQR tenure: A couple of years - He worked at AQR Capital after his PhD before joining Bloomberg in 2020 Bloomberg start year: 2020 - He joined Bloomberg in 2020 AI compute demand increase: ~100x or more - Ho said agentic AI and AI calling AI could increase compute demand by roughly a hundredfold or potentially more GDP framework: C + I + G + (net exports) - He referenced standard GDP accounting while explaining AI’s investment contribution AI stock/bubble comparison: At least as big as the crypto bubble, possibly the biggest bubble of all time - Ho recalled his May 2023 view on the AI investment cycle Productivity change claimed by firms: 20% more efficient - He cited anecdotal reports from companies adopting AI, though he noted measurement is unclear AI pricing regime expectation: Multi-tier / auction / pay-as-you-go - He predicted AI pricing will shift away from all-you-can-eat subscriptions Major macro disruption reference: Liberation Day and the Iran conflict - He suggested macro bearish events may have helped prolong the AI cycle

Pivotal Quotes: "I mean, of course it's a bubble. I mean, it has been a bubble. It was always going to be a bubble, right?" — Steve Ho: On whether the AI boom fits the definition of a bubble "The AI bubble is very much adopted and used by everybody almost right away. Unlike the dotcom bubble where you actually had a lot of unused capacity." — Steve Ho: Explaining why AI differs from the dotcom bubble "AI is too cheap, right? People are just out there consuming token because it's heavily subsidized." — Steve Ho: On current AI pricing and demand distortions

Implications: Expect AI to keep driving capex, market leadership, and tool adoption even if it’s a bubble. Near-term macro effects may be more inflationary than disinflationary, while labor impacts remain hard to measure. Long term, AI should improve modeling, pricing, and policy analysis.