68% Markdown in Three Years. The Pre-ChatGPT Unicorns Are Learning What Moats Are Actually Made Of.

More than 220 companies that once held billion-dollar valuations have now fallen below $1B. The cohort that last raised in 2021 is worth 68% less on average. Those that last raised in 2022 are down 52%. The largest single category of fallen unicorns is enterprise SaaS — 75 companies, double the number of fintech firms on the same list. These are not obscure startups that couldn't find product-market fit. Many were well-run companies with genuine enterprise contracts, predictable revenue, and credible management teams. What they shared was timing: they reached billion-dollar valuations by solving problems that AI can now solve differently, and sometimes for free.

The pattern in enterprise SaaS is specific. Companies that built revenue on per-seat pricing, workflow middleware, and manual process automation are the hardest hit. When the same workflow that required specialized software can be approximated by a general-purpose AI agent with a well-written prompt, the pricing power of the specialized software collapses. The customer doesn't churn immediately — switching costs are real. But they don't expand. They don't add seats. They wait for their contract to expire and then ask whether the same output can come from a different stack. That conversation is now happening in every enterprise IT department globally, and the sales cycles of companies that haven't answered it credibly are getting longer.

While 220+ companies lost unicorn status, AI-native startups raised $255.5 billion globally in Q1 2026 alone. Three deals — OpenAI's $122 billion round, Anthropic's $30.6 billion raise, and xAI's absorption into SpaceX — accounted for 67% of that capital. The capital is not being spread across hundreds of startups. It is concentrating rapidly in the companies that own the model capability itself, and in a smaller number of application-layer businesses that have demonstrated data moats or distribution that general-purpose models cannot replicate. Everything in between — the AI-enhanced products that added a wrapper to a pre-existing workflow tool — is under compression.

For investors managing LatAm portfolios, the fallen unicorn list is a stress test for every position built before 2023. Brazil's unicorn cohort was constructed primarily in 2019-2022 on fintech, payments, and enterprise SaaS logic. Some of those businesses are durable: the ones built on proprietary transaction data, embedded financial infrastructure, or distribution networks that took years to build. Others were valued on growth multiples that assumed continued SaaS expansion economics. The companies that built genuine data moats — payment processors with years of transaction history, credit platforms with proprietary default data, neobanks with behavioral data across millions of customers — are not the same companies as the ones that built workflow software with AI in the feature set.

The harder question is which of today's AI-native startups will be the fallen unicorns of 2029. The same mechanism operates with a three-year delay: a company reaches a billion-dollar valuation on a capability that is eventually commoditized or absorbed by a foundation model, and the valuation mark never updates until the fundraising process forces the issue. The difference between the current AI wave and the 2021 SaaS wave is that AI capability is compounding faster. The moat has to be data, distribution, or both — not the model itself. The companies that raised in 2024 on application-layer theses that assumed a capability advantage the foundation models will eliminate by 2027 face the same reckoning their SaaS predecessors are having now. The calendar is just shorter.