Outlier Investing in the Age of AI — Arsham Memarzadeh (Meritech)
Key takeaways from an X essay by Arsham MemarzadehDealroom has a profile for this one. Try Dealroom →, a VC at Meritech Capital, on investing at the growth stage in AI (July 2026):
- The future is less knowable than we pretend: in a market where value capture rotates every few months, there are more unknowns than knowns, and there is a credible bear case for every bull case. Skeptics doubted Anthropic's API layer would resist commoditisation, capped OpenEvidence at a Doximity-sized outcome, and bet inference engines and ClickHouse would stall — and left billions on the table.
- Filter for Outlier Components, not old frameworks: for mid- and late-stage growth companies, if a company has at least one outlier component, engage seriously; two or more, lean in hard.
- Outlier growth: top 0.1% of the cohort — paired with a large vision and >100% net dollar retention (leaky buckets excluded). Examples cited: Anthropic, Cursor/Cognition, ClickHouse, OpenEvidence, Kalshi, Mercor, Sierra, Lovable, Harvey/Legora.
- Outlier customer access: a captive or hard-to-penetrate set of relationships — e.g. Anthropic (Amazon GTM), Abridge (Epic), government relations (Anduril, Palantir), or open-source bases (ClickHouse, Vercel).
- Outlier team: not merely "great" or "technical" — a top 0.1% track record. Examples: Anthropic (GPT-3 creators), Nebius (Yandex founder), Sierra (Salesforce co-CEO Bret Taylor), SSI (OpenAI co-founder).
- Scale is the moat: technical moats are near-impossible to find now outside SpaceX, Waymo and Tesla. Companies scale first and build moats after; scale itself compounds via capital availability, cost structure and cross-sell.
- Caveats: the framework does not apply to early growth (<$20M ARR), where outlier components are nascent; and there are counterfactuals in both directions (overfunded A+ teams that never found PMF, and overlooked outliers such as Cerebras). There is no one-size-fits-all framework.
Read more: Arsham Memarzadeh on X