Cursor, Series C

i. Frame

Cursor is the IDE rewrite for the post-Copilot world: an AI-native editor that absorbs the workflow rather than annotating it. The category breaks first when AI scales — engineers spend their day in this surface and pay a higher willingness-to-pay than any horizontal SaaS we underwrite.

ii. Market

~28M professional developers worldwide. Estimated USD 30B addressable across IDE, code review, and AI-assisted coding by 2028. The AI-assisted slice grows from ~USD 1.5B (2024) to USD 12-15B (2028) on Gartner / a16z projections; we discount these but the direction is clear. Cursor sits at the centre of the slice.

iii. Product

Forked from VS Code, so compatible with the world's largest IDE extension library on day one. Differentiates on: AI-native UX (Cmd-K, Tab completion, Agent mode), multi-model routing, codebase-wide context, and a tasteful UI that has earned Hacker News fluency. Sticky workflow — once a team adopts, the switching cost is the muscle memory of every engineer.

iv. Team

Four MIT founders, all under 28, top-quartile competitive-programming pedigree. Two stints at OpenAI / Anthropic between them. Velocity is the team's brand — releases every two weeks, public benchmarks, founder-led customer support on Twitter. Bench depth is the unknown — 90+ headcount is fast for a series-C-scale team; org design is the next test.

v. Traction

Reported USD 100M ARR by Q1 2026, up from USD 35M Q1 2025. Net dollar retention reportedly 140%+. Free→paid conversion in the 8-12% range based on third-party signals. Enterprise tier launched late 2025; early logos include three FAANG-adjacent firms and the major LLM labs (notable signalling).

vi. Unit economics

Gross margin pressured by inference costs — estimated 55-62% blended (model costs ~ USD 8-12 per paying user per month at average usage). Enterprise tier carries higher margin (~78%) due to bring-your-own-key arrangements. CAC payback approximately 4 months on individual plans (product-led), 9-14 months on enterprise.

vii. Competition

GitHub Copilot (incumbent, ~1.5M paid users, Microsoft-owned). Codeium / Windsurf (competing on enterprise self-hosting). Sourcegraph Cody (incumbent in code search, AI add-on). Continue (open source, dev-mindshare). Cline (agentic, open source). Replit (adjacent — IDE + hosting). All except Replit are direct competitors on the IDE surface.

viii. Defensibility

Three moats, in descending order of durability: (1) workflow lock-in — the editor is muscle memory; (2) brand among AI-native engineers — Cursor is the default vote in the most influential developer communities; (3) model-agnostic stance — by routing across providers, Cursor reduces dependency on any single LLM lab. The first moat compounds. The second is fashion. The third is hedge.

ix. Deal structure

USD 100M primary, pre-money USD 9.0B (10.5% dilution). Co-lead with one growth fund; we propose USD 15M check, full participation in our pro-rata. Information rights, no board seat (one observer). Standard 1× non-participating preferred, broad-based weighted-average anti-dilution. No special anti-protective on AI-IP scenarios — we should request.

x. Risks (12-mo)

Inference cost spike if base-model providers reprice (Anthropic / OpenAI inference price increases would compress margins ~400 bps within 90 days). GitHub Copilot price reduction — Microsoft has both motivation (bundle moat) and cash to cut Copilot to USD 5/mo or below. Brand churn — Cursor's mindshare is real but fashion-cycle-vulnerable; one competitor with a better Agent mode could move the needle.

xii. Trip-wires

Three triggers that would cause us to exit early in a follow-on cycle:

(1) GitHub Copilot drops to USD 5/mo or bundles free at any non-trivial GitHub tier. Watch in the next 12 months.

(2) OpenAI or Anthropic ships a native IDE (not a chat surface — a true editor). Watch in the next 18-24 months.

(3) Net new enterprise ACV per quarter declines for two consecutive quarters. Watch from Q3 2026.

xiii. Anti-portfolio mirror

We passed on Replit early (2021) on the thesis that browser-based IDEs would not displace desktop. We were partially right and entirely missed the AI-augmented browser-IDE category. The lesson: AI-native developer tools compound faster than the previous tools cycle, because the model itself adds non-linear capability per release. We should not repeat the Replit pass on Cursor for the same reason.

xiv. Exit realism

Three plausible exit paths: M&A by Microsoft (most likely buyer if they fail to compete via Copilot — USD 8-15B range over 2027-2029). M&A by Google or Adobe (less likely; USD 5-10B). IPO at USD 200M+ ARR if growth holds — possible but not necessary for our return; the IPO scenario requires both rapid growth and stable gross margin expansion, two conditions that don't always co-occur.

xv. The ask

USD 15M co-lead at USD 9.0B pre-money. Pro-rata for the next round. Observer seat (not board). Information rights. We propose conditional on: (a) inclusion of an AI-IP protective covenant; (b) committed 90-day diligence on enterprise pipeline (verified ACV, not booked); (c) reference checks with two engineering VPs at named enterprise customers.

Conviction 8 / 10. The room held together on the bull case; Taleb and Gurley reserved on the 36-month frame. The trip-wires are real and the hedge is the discipline.