Widder and Hicks: GenAI Hype Deflates, Leaving Lasting Damage
TL;DR
- Goldman Sachs and Sequoia Capital publicly questioned whether generative AI costs are generating commensurate revenue, signaling the hype peak.
- U.S. coal plant retirements slowed by 40% amid AI energy demand; Microsoft abandoned its 2030 carbon-negative goal as emissions climbed 30% since 2020.
- The authors argue the bubble's deflation will not reverse environmental damage, labor displacement in creative fields, or erosion of information quality.
A pair of researchers has mapped the generative AI hype cycle's landing strip, and the view is not flattering. In an essay for the Ash Center for Democratic Governance and Innovation at Harvard, David Gray Widder and Mar Hicks argue that "Generative AI was sold to us as inevitable by the leadership of AI companies," pushing every business to try to "become an AI business," a transformation now slowly reversing, but not cleanly.
The financial signals of deflation are visible. Sequoia Capital asked "Where is all the revenue?" and Goldman Sachs released a report whose title says most of it: "Gen AI: too much spend, too little benefit?" But the authors' sharper point is that the financial correction is the least durable part of the story. The harms they document operate on slower timescales.
On energy, U.S. coal plant retirement plans reportedly slowed by 40%, with the "AI war" cited as justification. Microsoft's emissions rose 30% since 2020, and the company abandoned its 2030 carbon-negative goal. Google's emissions climbed too, even as the company pushed its net-zero targets further out. As the essay puts it, "carbon can't be put in the ground."
The labor damage centers on creative workers. Illustrator Molly Crabapple is quoted on the threat to working illustrators as editors reach for cheaper AI alternatives, even if high-end commissions survive. On the information side, Google's AI-powered search surfaced suggestions like using glue to keep pizza toppings in place and eating rocks, obvious errors, but the authors' concern extends to the subtler misinformation that will be harder to trace.
What the essay does not resolve is how much of this damage remains reversible, and it predates some of the AI revenue growth reports that emerged after its publication. Those later developments complicate the deflation thesis somewhat. The core claim still holds: "today's hype will have lasting effects that constrain tomorrow's possibilities," and the cloud infrastructure dependencies that built up during the boom do not simply dissolve when the valuations do.
Shared on Bluesky by 2 AI experts
-
Once again, everyone is now worried there's an AI bubble. It is, and it will deflate. But deflating doesn't undo the harmful effects, as @histoftech.bsky.social and I wrote back in 2024… ash.harvard.edu/resources/wa...
View on Bluesky →
Originally reported by ash.harvard.edu
Read the original article →Original headline: Watching the Generative AI Hype Bubble Deflate – Ash Center