What Is This?
In early February 2026, Anthropic released industry-specific plugins for Claude — and nearly $1 trillion was wiped from software stocks in weeks.^1 FactSet dropped from a $20B peak to under $8B. S&P Global lost 30%. Thomson Reuters shed almost half its market cap. The S&P 500 Software & Services Index fell 20% year to date across 140 companies.
Wall Street called it a panic. Nicolas Bustamante — who spent a decade building Doctrine (Europe's largest legal information platform, competing with LexisNexis and Westlaw) and then Fintool (an AI-powered equity research platform competing with Bloomberg and FactSet) — says the market is right on direction, just wrong on timing.^2
His framework: vertical software was defended by ten distinct moats. LLMs are destroying five of them. The other five survive — but the five being destroyed are precisely the ones that kept competitors out in the first place.
Why Does It Matter?
- The "SaaSpocalypse" reflects a real structural shift, not just panic. The premium multiples vertical SaaS commanded (15-20x revenue) were justified by switching costs and limited competition. Both are dissolving.
- Competition doesn't scale linearly — it explodes. Before LLMs, building a credible Bloomberg competitor required 200 engineers and $50M in data licensing. After LLMs, it takes 6 engineers and a frontier API. Competition goes from 3 players to 300. That alone craters pricing power.
- The interface was secretly most of the value. Bloomberg Terminal users paid $25K/seat not just for data — but because they'd spent a decade learning keyboard shortcuts. LLMs collapse every proprietary interface into chat. That entire moat evaporates.
- Business logic migrates from code to markdown. What used to take engineering teams years to encode — legal research workflows, DCF valuation methodologies — can now be written by a domain expert in a week. Zero engineers required.
- It's a slope, not a cliff. Enterprise contracts (2-year Bloomberg minimums, multi-year FactSet deals) keep revenue stable for 12-24 months. But stocks price multiples, not revenues — and when the market stops believing in pricing power, the stock collapses even if revenue holds.
Key People & Players
- Nicolas Bustamante — Builder of Doctrine (legal SaaS, Europe) and Fintool (AI equity research, US, Anthropic-backed). Uniquely positioned having built the disrupted and the disruptor.^2
- Anthropic — Their Claude "Cowork" plugins for knowledge workers triggered the selloff. Now going deep into vertical territory via a simple stack: agent SDK + MCP data plugins + markdown skills.^1
- Microsoft — Using Copilot inside Excel and Word to do DCF modeling and contract review — horizontal tools becoming vertical through AI.
- Bloomberg, FactSet, S&P Global, LexisNexis — The incumbents absorbing the hit. Their moats are being stress-tested in real time.
- Doctrine — Bustamante's legal SaaS company, which survived by pivoting to proprietary content (annotations, editorial analysis) nobody else has.
- Fintool — Built by 6 people. Serves hedge funds previously running exclusively on Bloomberg/FactSet. No onboarding, no CSMs, no UI to learn.
The Current State
Bustamante identifies ten moats that made vertical software defensible — and maps what LLMs do to each:
Destroyed or Weakened:
- Learned Interfaces → Destroyed. Chat collapses years of keyboard muscle memory into natural language.
- Custom Workflows & Business Logic → Vaporized. Years of engineering → one markdown file.
- Public Data Access → Commoditized. Frontier models already know how to parse 10-Ks. The parser IS the model.
- Talent Scarcity → Inverted. Domain experts can now write software directly, without engineers.
- Bundling → Weakened. AI agents become their own bundle, orchestrating across providers transparently.
Surviving: 6. Private & Proprietary Data → Stronger. Data nobody else can replicate becomes MORE valuable as everything else commoditises. 7. Regulatory Lock-in → Structural. HIPAA, FDA, SOX compliance doesn't care about LLMs. 8. Network Effects → Sticky. Bloomberg's IB messaging network, Veeva's pharma relationships — these are people moats, not interface moats. 9. Transaction Embedding → Durable. If you process payments or settle trades, an LLM sits on top of you, not instead of you. 10. System of Record → Threatened long-term. AI agents are quietly building their own system of record from accumulated context — but this takes years.
The three-question test: Is the data proprietary? Is there regulatory lock-in? Is the software embedded in the transaction? Zero yes = high risk. One = medium. Two or three = probably fine.^2
Best Resources to Learn More
- The original article: Nicolas Bustamante's full 10-moat breakdown on X — mandatory reading for anyone in software, investing, or AI^2
- Bloomberg's SaaSpocalypse coverage: Full reporting on the $1T selloff and Wall Street's split on whether it went too far^1
- Ben Thompson's Aggregation Theory — The framework Bustamante implicitly references: AI agents as aggregators, data vendors as commoditised suppliers
- MCP (Model Context Protocol) — Anthropic's open protocol turning every data provider into a Claude plugin; understand this to understand why the "making data accessible" moat collapses
- Fintool (fintool.com) — Live example of a 6-person team displacing Bloomberg for hedge funds