Ben Horowitz: bubbles burst when nobody believes it's a bubble — and right now, everyone's talking about one

Summary

Ben Horowitz frames the current AI investment environment as categorically different from the dot-com era, and uses that distinction to push back on 2025's pervasive bubble narrative. His core argument is psychological: bubbles burst when consensus forms that prices will never fall. By that logic, widespread bubble talk is actually a contrarian signal of market health.

The historical contrast he draws is sharp. In 1996, Netscape held 90% browser share serving a total internet population of 55 million users, many on dial-up, while companies were receiving $10 billion valuations. ChatGPT, by comparison, went from zero revenue in November 2022 to an estimated $15 to $20 billion in revenue, a trajectory Horowitz describes as unprecedented. The technology is reaching the market in real time rather than running years ahead of it.

He cites the 2007 housing crisis as a parallel model. Mortgage spreads on home loan debt hit historic lows right before mass defaults, precisely because universal belief in land scarcity had eliminated perceived risk. That same psychological convergence, not financial fundamentals, is what defines a true bubble. His baseline view on venture pricing is that assets are routinely mispriced at either half or double fair value, and that condition is simply the normal operating environment.

Andreessen Horowitz raised its first fund at $300 million in 2009 amid criticism it was too large. Fund three, at $1 billion, drew similar skepticism. That fund returned Coinbase, Databricks, Lyft, DigitalOcean, and GitHub. The firm's latest raise is $15 billion. Horowitz attributes the scaling logic to a structural shift: technology is no longer a sector, it is all sectors, which multiplied the number of consequential companies that will be built.

The firm's response has been to subdivide into independent sub-funds covering infrastructure, applications, crypto, bio, early stage, and American Dynamism, each operating with the deal-making structure of the original firm. Horowitz is explicit that large group decision-making degrades investment judgment, so each team functions autonomously. On AI specifically, the firm ran internal training programs and what he describes as exams to ensure all relevant staff were, in his term, AI native before deploying capital, citing the pattern of firms that dominated in 2015 but failed to adapt.

On California's proposed wealth tax, Horowitz points to Norway as a live experiment in the consequences. Norway's unrealized capital gains tax drove out its tech entrepreneurs because founders holding illiquid equity in marked-up private companies had no mechanism to pay the liability without leaving. He argues Norway now has effectively no tech entrepreneurship, and describes a similar policy applied to California as the most credible threat he has seen to Silicon Valley's network effects.

Within the firm, Marc Andreessen handles policy and AI at a depth level while Horowitz runs operations. The two have worked together for roughly 30 years. Horowitz describes the relationship as collaborative and argumentative in equal measure, which he treats as a feature rather than a friction.