Most AI Companies Won’t Survive (Tech Investor Explains)
A tech investor argues that most AI companies will fail, drawing parallels to the dot-com bust where roughly 1,980 out of 2,000 public companies collapsed. He advises founders to honestly assess whether they're among the durable few or whether the next 12-18 months represents their peak valuation window. He outlines characteristics of durable AI companies and the various exit options available.
Summary
The investor opens by drawing a historical parallel between the current AI boom and past technology cycles, noting that 90-95% of companies in any major tech cycle eventually go bust. He references the dot-com era specifically, where roughly 1,500-2,000 companies went public and only a dozen or two survived in meaningful form. He argues there is no reason to expect the AI cycle to be different, citing SaaS, mobile, and crypto as additional examples of cycles with similarly high failure rates.
Given this historical pattern, the investor advises AI founders to honestly assess whether their company is among the durable few or whether they are approaching a 'value maximizing moment' — typically a 6-12 month window — where selling would yield the best possible outcome. He notes that warning signs include a plateauing second derivative of growth, and that many companies will face commoditization, competition from labs, or technological obsolescence.
On the question of which companies will endure, the investor identifies the core foundation model labs — OpenAI, Anthropic, Google DeepMind — as likely survivors, predicting an 'igopoly' market structure he says he wrote about three years prior. He acknowledges newer entrants like Meta and xAI could shift this dynamic. For application-layer companies, he outlines three key criteria for durability: (1) whether improving underlying models directly enhances the product's value to customers, (2) how deeply and broadly the product is embedded into customer workflows and business processes — noting that change management, not technology quality, is often the real adoption barrier, and (3) whether the company is capturing and storing proprietary data, though he cautions that data moats are generally overstated.
Finally, the investor outlines the exit landscape for founders considering selling. He notes that the emergence of multi-trillion dollar market caps among big tech companies has created unprecedented acquisition buying power — 1% of a $3 trillion company is $30 billion. He identifies four main types of acquirers: large labs and hyperscalers, giant tech incumbents (Apple, Amazon, Oracle, Samsung, Tesla, SpaceX, Snowflake, Databricks, Stripe, Coinbase), vertical-focused strategics (e.g., Thomson Reuters for legal tech), and competitors via mergers. He highlights competitor mergers as an underutilized option, citing the X.com/PayPal merger and the near-miss Uber/Lyft consolidation as examples where combining forces against shared incumbents would have been strategically rational.
Key Insights
- The investor argues that historical tech cycles — from automotive to dot-com to SaaS to crypto — consistently see 90-95% company failure rates, and that out of roughly 2,000 dot-com-era public companies, only a dozen or two survived in meaningful form, with no reason to expect the AI cycle to be different.
- The investor contends that the primary barrier to AI adoption at the application layer is not the quality of the AI itself, but change management — how much companies must alter their existing workflows and how their people work in order to adopt the technology.
- The investor predicted roughly three years prior that the foundation model market would become an igopoly aligned with cloud providers, and argues this has largely materialized, though Meta and xAI were not factors when he made that prediction.
- The investor argues that the rise of multi-trillion dollar tech market caps has created historically unprecedented acquisition buying power, where even a 1% portfolio reallocation by a $3 trillion company represents $30 billion in deal capacity, making large AI acquisitions financially feasible in a way that was not possible 10-15 years ago.
- The investor argues that mergers between competing private AI companies are an underutilized exit strategy, citing the X.com and PayPal merger as a historical precedent where direct competitors combined rather than continuing to destroy each other's pricing, freeing resources to compete against larger incumbents.
Topics
Full transcript available for MurmurCast members
Sign Up to Access