Greenoaks Capital
Neil Mehta (founder/managing partner); registered investment adviser
Concentrated, research-driven, cross-over (private + public) investing — fewer, larger, long-hold bets. Punches far above its AUM by co-leading mega-rounds in the very largest private AI/data names. The boutique with marquee-round access.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Self-reported — Headline returns are manager- or firm-reported and unaudited.
- Partial 13F — The disclosed book is longs-only and ~45 days lagged — a slice, not the strategy.
Key positions
Co-lead of the $65B Series H at the ~$965B valuation (May 2026) — alongside Altimeter, Dragoneer, Sequoia, Coatue
Long-held AI-data infrastructure position
Data-labeling leader (amid its major Meta transaction)
Cloud-security marquee win (recent large exit)
Pattern of concentrated generational bets
Recent moves
Co-led Anthropic's record $65B Series H at ~$965B (May 2026) — an unusually large check for a ~$6B-AUM firm — after raising a $2.5B sixth fund, doubling down on a concentrated frontier-AI + data book.
Loading disclosed positions…
Our take
The boutique that gets marquee-round allocations despite a fraction of a16z/Sequoia's AUM — appealing for concentrated, high-conviction exposure, but the small fund base means each bet is existential and there's little diversification cushion. Lowest AUM, highest per-bet conviction of the group.
The marketing line is 'concentrated conviction.' The mechanical truth is that the headline IRR is largely a Fund I artifact now amortized across a far larger asset base.
Roughly 40% of a $50M first fund into Coupang returned ~$8B — one position plausibly accounting for most of the $13B lifetime gross profit, and unrepeatable at $2.5B fund size. A check that moves a $2.5B fund the way Coupang moved a $50M fund must be a $1B+ position that then compounds 100x. So the firm has structurally shifted from 'find the next Coupang early' to 'buy a large late-primary slice of an already-consensus winner.' Co-leading Anthropic's $65B Series H at $965B post-money (closed May 28, 2026) is a late-stage crossover trade, not a venture bet — a different edge (access, capital, speed) with a different return profile: 3-5x, not 160x. Judge the 2024-26 vintage on that basis, not on the 33%.
Thesis
Greenoaks runs a hedge-fund concentration budget inside a venture wrapper: ~15 companies per fund, with a willingness to put 40% of a fund into one name (Coupang, Fund I). The disclosed record — 33% net IRR, >$13B gross profits over 13 years — is manager-narrated on a podcast, not audited, and appears dominated by a single 2014 decision. The live question is whether the 2024-26 frontier-AI vintage is the same underwriting discipline, or the same discipline applied at 50x the entry valuation.
Crossover, research-first, long-hold. ~$25-75M initial checks into roughly 15 companies per fund, held for decades and followed hard into later rounds; the same names reappear in the public book after IPO (Coupang, Carvana, DoorDash, ServiceTitan). Nine investment professionals cover ~55 core companies — deliberately understaffed relative to AUM, which is the design: no deal-flow machine, no seed spray, few decisions made per year.
Assessment
- Position sizing is genuinely differentiated: ~15 names per fund, and 40% of a fund in one company, is a real deviation from the 30-name index-VC default.
- Nine investors covering ~55 companies forces a low decision rate — a structural defense against the deal-flow-volume trap that degrades most VC firms as they scale.
- The public 13F sleeve holds the same names post-IPO (Coupang, Carvana, ServiceTitan), evidence the 'hold for decades' claim is executed rather than just stated.
- Real distributions, not only marks: Wiz closed into Google at $32B on March 11, 2026, and Coupang/Carvana are liquid listed positions.
- The 33% net IRR and >$13B gross-profit figures are manager-narrated on a 2025 podcast. No audited, vintage-by-vintage, LP-verified breakdown is public — treat as marketing arithmetic until one exists.
- Almost certainly a single-asset record. If Coupang alone returned ~$8B of a ~$13B lifetime gross profit, the other twelve years of selection are unproven at the headline rate.
- Concentration at $2.5B fund size cannot replicate the Fund I math. Scale forces later entry at higher prices, compressing returns even if picking quality is identical.
- The frontier-AI marks are the fragile part of the book: Anthropic, Databricks and Scale are all unrealized, priced by the last round, and correlated to one capex cycle.
- The listed sleeve fell from ~$2.98B (12/31/2025) to ~$2.30B (3/31/2026) even as a new $75M DoorDash stake was added — the 13F cannot separate price drawdown from selling.
- 2026-07Update, Jul 2026: the capex shock came (SOX -20.6%) and the AI marks rose — Databricks took a $188B term sheet on 16 Jul vs $134B in Feb; Anthropic secondaries printed ~$1.2T on 9 Jul vs the $965B May primary.
- 2026-07Update, 30 Jul 2026: the liquid half is now the drag — Coupang is -31.8% YTD at $16.09, and Carvana fell double digits on its 29 Jul guidance miss, even as the private AI marks moved higher.
Record
Manager-reported: 33% net IRR and over $13B gross profits across 13 years with roughly one point of principal impairment (Colossus / Invest Like the Best, Apr 2025) — manager-narrated, unaudited, not decomposed by vintage. Attribution is the whole story: ~40% of the $50M Fund I into Coupang returned ~$8B, so the composite is one 2014 decision plus a good-but-ordinary tail (Figma, Wiz, Carvana, Stripe, Toast, Rippling). Realized evidence is real — Wiz closed into Google at $32B on March 11, 2026 — but the 2023-26 AI book (Anthropic, Databricks, Scale) is entirely paper. The 13F is a poor proxy: $2.297B across 10 names as of 3/31/2026 (filed 5/15/2026), ~15% of the ~$15B firm AUM secondary sources report, longs-only, US-listed, ~45 days lagged.
- 2026-06Update, Jun 2026: Anthropic filed confidentially for an IPO on 1 Jun, targeting a debut as soon as this fall — still unrealised, but with a dated liquidity path and secondaries pricing above the round.
Risks & fit
- Key-person concentration: Neil Mehta is the firm. No public succession plan, and a nine-person investment team is a thin bench.
- Power-law dependence with only ~15 shots per fund — two dead names plus a flat winner produces a mediocre vintage with no diversification cushion.
- AI valuation reset: an Anthropic or Databricks markdown hits several positions at once because the exposures are correlated, not independent.
- Illiquidity and J-curve: Fund VI (2025) will print negative net IRR for years, and interim TVPI is a marked estimate rather than cash.
- Late-stage crossover entry offers less protection — a ~$965B entry price leaves little room for multiple expansion.
Two things would settle it. (1) An LP-grade, vintage-by-vintage, net-of-fee DPI/TVPI table showing Funds III-V clearing ~25% net IRR excluding Coupang — that would show a repeatable process rather than one bet. (2) Cash realizations from the AI book: if Anthropic or Databricks lists and Greenoaks distributes at 3x or better on its 2024-26 entry price, the crossover pivot is validated. Conversely, a flat-or-down AI round in 2027 forcing a written-down mark, or Fund VI still below 1.0x DPI at year seven, would falsify the claim that this concentration style compounds at scale.
Analytically interesting to anyone studying how concentration behaves as fund size grows, or benchmarking the crossover private-plus-public model against pure venture. Greenoaks is a closed, LP-only private partnership — not accessible to outside individuals, and nothing here suggests seeking access. The only public read-through is the 13F sleeve (Carvana, Coupang, Navan, ServiceTitan, DoorDash), a lagged and partial signal that should not be read as a portfolio.
Terms are not disclosed. Concentrated long-hold venture of this profile typically carries roughly 2% management and 20-30% carry with long recycling. The 33% is stated net, but with no fee schedule public the gross-to-net bridge is unverifiable.