Most AI Companies Won’t Survive (Tech Investor Explains)

Advice for AI Founders to Exit in 12-18 Months00:00

Founders running successful AI companies should take a cold hard look at exiting in the next 12 to 18 months, which might be a value maximizing moment for outcomes. This draws from the dotcom bust and survival rates.

Historical Precedents: 90-99% of Tech Cycle Companies Fail00:30

In every technology cycle, 90-95% or 99% of companies go bust, dating back to the automotive industry in Detroit with dozens of car companies and hundreds of suppliers collapsing into a small number. During the internet bubble of the '90s, 450 companies went public in 1999, 450 in the first few months of 2000, and another 500,000 in the couple years before, totaling 1,500-2,000 that went public. Of those, only a dozen or two dozen survived; 1,980 or so went under or got bought for a little bit. Every cycle is like that: SaaS, mobile, crypto. Most AI companies won't make it; a handful will.

Characteristics of Durable AI Companies

AI founders should ask: What is the nature of your company's durability? Are you one of the dozen or two that will be really important 10 years from now, or is now a good moment to sell because your work will get commoditized, competed by a lab, or become obsolete due to market or technology shifts? A handful of companies should never sell or exit; they should keep going. For every company, there's a value maximizing moment, usually a 6-12 month window where what you're doing is important enough and scaling before a headwind hits. Sometimes the headwind is predictable, visible in the second derivative of growth plateauing.

I'm incredibly bullish around AI... it's less about the transformation that's happening overall because of the technology and more that only a handful of companies are going to continue to be really important. Are you one of them or not? If you're one of them, you should never ever ever sell.

Handful with Durable Advantage: Core Labs and Characteristics03:32

Core labs will be around: OpenAI, Anthropic, Google (barring accident or disaster). Market structure is an oligopoly aligned with cloud providers (predicted in a Substack post 3 years ago). Meta, xAI, others may change it, but short run it's oligopoly, not monopoly unless one pulls ahead massively (compute constraints may prevent). No monopoly so far.

Application Layer Criteria for Durability

• If underlying model gets better, does your product/service get dramatically better so customers keep using you?
• How deep and broad: multiple integrated products built directly into company processes, hard to pull out? Adoption issue is often change management and workflows, not AI quality.
• Embedding into workflows, how people work, ties everything together is durable.
• Capturing, storing, using proprietary data (data moats overstated but sometimes useful, e.g., system of record view).

Examples: Harvey (legal), Brevian (health? [context: bridge for health]), Decagon and Sierra (customer success).

Exit Options for AI Companies06:02

Multi-billion dollar AI companies with limited lifespan: labs can probably build what they sell. Options:

• Acquired by a lab (build vs. buy decision).
• Acquired by players getting skin in the game like Amazon.

Enormous buying power now: biggest market caps were $300B 10-15 years ago (e.g., Exxon); now multi-trillion, unprecedented. 1% of $3T is $30B. Many companies with $100B to few trillion market caps mean big acquisitions possible.

Potential Buyers

• Big labs, hyperscalers, giant tech: Apple, Amazon, Google (both lab and giant), Oracle, Samsung, Tesla, SpaceX.
• Snowflake, Databricks, Stripe, Coinbase (financial).
• Vertical-specific: Thomson Reuters (legal/accounting).
• Mergers of competitors, especially private: better to merge if neck-and-neck, destroying pricing. Examples: X.com and PayPal (Elon Musk companies merged because "why fight?"); Uber-Lyft rumor (Uber walked away, but fighting cost money).

If the primary purpose is to win the market, you're already fighting big incumbents. Why make it even harder?

Even well-known companies get zero allocation from successful tech investors in small group threads when allocating 10 bets.