Will AI make anyone rich? A deep dive into the 'containerization' bear case for generative AI investing

Summary

Jerry Newman's bear case on AI investing draws a historical parallel. Containerization revolutionized global commerce but created almost no wealth for investors, while the microprocessor sparked a 30-year wave of venture creation and founder fortunes. Newman questions which pattern AI will follow.

The containerization comparison turns on a structural difference. Malcolm McLean's Sealand solved a real logistics problem by moving goods seamlessly between trucks, trains, and ships. The technology was economically transformative, but because existing businesses captured most of the innovation's benefits, almost no venture-scale winners emerged. The PC revolution created thousands of new companies, venture rounds, and paper millionaires because microprocessor costs collapsed from $360 in 1971 to $25 within years. That cost cliff made it possible for tinkerers like Steve Wozniak to build in garages. Decentralized experimentation with low barriers to entry sparked the entire wave.

Newman argues generative AI resembles containerization more closely. It concentrates in a few foundation model companies like OpenAI and Anthropic, owned and controlled by incumbents, with limited room for permissionless experimentation. Application companies built on top of these models face a structural squeeze. If Perplexity, Inflection, or others succeed, the foundation model owners can raise prices, acquire them, or launch competitors. Success for the app layer becomes a defeat.

For investors, the implication is stark. Large venture funds deploying $200 million at high multiples into pre-revenue foundation model labs will struggle. Application-layer investing "as a whole will lose money," though individual companies may become acquisition targets or benefit from FOMO-driven consolidation.

Newman doesn't claim AI creates no wealth. Instead, he argues the 50-year playbook for tech investing no longer applies. The old strategy was to bet on what the new thing is. That worked for PCs. Now investors should "bet on the opportunities it opens up." Think through what markets efficiency unlocks, not just the models themselves.

The hosts push back substantively. In venture, even if you lead early rounds in successful companies, you're often locked in. A $200 million check deployed early can't easily be exited on a 3x markup two rounds later. Among app-layer companies, venture's model still works: most fail, a few create enormous value. FileVine, an existing enterprise software player for law firms with 100,000 daily users, is now well-positioned to integrate AI, a classic incumbent advantage play.

A deeper tension emerges. Saying the app layer will lose money differs from saying spray-and-pray investing at high multiples will lose money. Category definition matters enormously. Oligopolies in customer service will look different from monopolies in search. Winners in defined categories with first-mover advantages and customer lock-in still exist. They just won't be as numerous as in the PC era.

What remains unresolved is whether the permissionless experimentation phase Newman describes has already begun. Open-source models like Llama are accessible locally. Either eruption is already happening, or incumbent gatekeeping remains in place. If the latter, the containerization pattern holds and real AI wealth accrues to model builders and acquirers of breakout apps, not to venture as a whole.