Is AI in a Bubble? History Says It's Complicated
The hosts discuss whether AI is in a bubble by comparing it to historical bubbles (railways, dot-com, housing), concluding that while a correction is likely, the technology's fundamental value and companies' proactive planning may prevent catastrophic collapse. They emphasize that societal disruption from mass unemployment poses the greatest risk, requiring governments and companies to be proactive about solutions like universal basic income.
Summary
The discussion begins with research into three historical bubbles: railway mania (1840s), the dot-com bubble (late 1990s-2000), and the 2008 housing crisis. The hosts note that while these bubbles burst, the underlying technologies (railroads, internet) proved genuinely revolutionary. They distinguish between a bubble burst and a market correction, arguing that the dot-com bubble ultimately undervalued the internet's true potential, as today's dominant companies (Google, Amazon, Netflix) emerged from its aftermath.
Mike argues that AI differs from previous bubbles because investors are now more sophisticated and aware that corrections happen. Large AI companies (OpenAI, Anthropic) are positioning themselves for profitability and embedding themselves into economies at infrastructure levels (compared to water or internet providers). He contends that while small AI companies will fail during a correction, the major players will survive because governments and other mega-corporations cannot afford to let them collapse, similar to how Microsoft bailed out Apple in the 1980s.
The host counters that pricing concerns exist: subscriptions appear subsidized (e.g., $200/month for $19,000 in API credits), and companies are spending money on speculative projects (domestic robots, AI pins, AI glasses) without proven market demand. He argues that unlike railroads or the internet, which directly served human needs, AI's main purpose is to replace human labor, creating unprecedented societal risks.
The conversation pivots to the fundamental economic problem: if AI eliminates most jobs, who pays for subscriptions and mortgages? This leads to discussion of potential solutions like universal basic income (UBI) and citizen dividends, which companies like Anthropic and OpenAI are already proposing through token taxes. Mike suggests that AI companies understand they must prevent widespread unrest through proactive social measures, or they cannot maintain their markets.
The hosts debate whether a soft landing is possible. The utopian scenario involves reduced work hours, UBI, and sustained economic growth through increased consumption. However, they acknowledge this requires proactive government planning, which historically hasn't happened—societies are reactive, waiting for crises before acting. The host expresses skepticism that such coordination will occur.
They conclude by emphasizing that regardless of one's stance on AI, understanding these dynamics is essential. Both recommend education over ideology, noting that experts dramatically disagree on AI's trajectory. The hosts advise listeners to prepare themselves based on personal research rather than betting entirely on optimistic or pessimistic scenarios.
About this episode
Is AI really in a bubble? Matt and Mike explore the history of major investment bubbles, from railroads to the dot-com era, and discuss what those events might tell us about today’s AI boom.
Key Insights
- Railway experts couldn't predict where railways should go or how to efficiently operate them despite the technology being revolutionary, demonstrating that even revolutionary technologies experience bubbles due to speculation about applications.
- The dot-com bubble undervalued the internet's true potential; investors who survived the crash and invested in the right companies (Google, Amazon) achieved returns far exceeding even inflated expectations, suggesting that corrections don't invalidate revolutionary technologies.
- AI companies are embedding themselves into economic infrastructure at levels comparable to utilities (internet/water providers), making government bans economically catastrophic and preventing regulatory intervention even if desired.
- The fundamental business model problem for AI companies is selling compute at apparent loss (e.g., $200 subscriptions yielding $19,000 in credits), but the actual profit/loss structure remains unclear because training and inference costs, ROI calculations, and long-term pricing strategies are not publicly known.
- Unlike railroads (which transported goods and people) or the internet (which provided information access), AI's primary function is replacing human labor, creating an unprecedented economic problem: if humans are displaced, who generates income to purchase AI services and maintain the broader economy.
- Companies like Anthropic and OpenAI are proposing pre-emptive measures (token taxes for universal basic income funds) because they understand that widespread poverty and unrest would destroy their markets, suggesting industry awareness of societal collapse risks.
- Small speculative AI projects (domestic robots, AI pins) in early stages will likely fail during a market correction, but this won't cause industry-wide collapse because large tech firms have diversified portfolios and the ability to sustain losses through other revenues.
- Societies historically respond to crises reactively rather than proactively, meaning economic safeguards for potential mass unemployment are unlikely to be implemented before the crisis occurs, increasing the severity of potential disruption.
Topics
Transcript
All righty, everybody. This is another edition of the web news. And today we want to talk about the AI bubble and we've seen tons of people talking about this. Is there an AI bubble? Is there not, when will it pop? Because there has to be one, et cetera, et cetera, et cetera. So I want to be clear that this is not like some hardcore, like lots and lots of research into economy stuff and all this, this is sort of an editorialized discussion about the AI, AI bubble. But also I did do research into previous bubbles and we can kind of compare and contrast those historical tidbits, if you will, into this AI bubble or not…
Full transcript available for MurmurCast members
Sign Up to AccessMore from HTML All The Things - Web Development, AI, and Developer Careers
How to Choose the Right CMS for Your Website
The podcast emphasizes that there is no universally 'best' CMS; instead, the right choice should hinge on critical questions regarding a project's specific requirements. Seven essential questions are outlined to guide CMS selection, addressing aspects such as website management, frequency of updates, and custom functionality.
Do You Actually Own Your Digital Games?
A Brazilian Xbox player won a court case against Microsoft after his account was hacked and permanently suspended, resulting in loss of his entire digital game library and OneDrive files. The hosts discuss this as a potential win for digital ownership rights and explore broader concerns about account security, digital preservation, and the nature of licensing versus ownership in digital storefronts.
AI Coding Makes Development Easier - So Why Are We Burning Out?
Hosts discuss how AI coding tools, while making development easier, are paradoxically causing burnout by eliminating the struggle-and-satisfaction cycle that previously sustained developer motivation. They explore the psychological toll of AI-assisted work where accomplishment feels hollow and recommend finding personal projects and adjusting workflows to maintain job satisfaction.
Kimi K3 Brings Frontier AI Into the Open
Kimi K3, a new 3 trillion parameter open-source AI model, represents a major advancement in frontier AI that rivals closed-source models like GPT-5.6 and Claude Opus. While offering exciting possibilities for corporations to run powerful AI locally without subscription costs, it poses significant safety risks by allowing malicious actors and state entities to deploy unrestricted AI systems without safety guardrails.
JavaScript Isn't Always the Answer: Native Browser Features Every Developer Should Know
This episode discusses how modern browsers have evolved to handle features that previously required JavaScript or Flash, such as form validation, accordions, dialogs, popovers, and responsive images. The hosts argue for using native browser APIs and semantic HTML instead of over-engineering with JavaScript, which reduces code complexity, improves performance, and enhances accessibility.