Dragoneer Investment Group
Marc Stad (founder, managing partner)
Tech-focused crossover fund with a small, highly concentrated public book and a large private franchise — a major direct backer of OpenAI and Anthropic.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Partial 13F — The disclosed book is longs-only and ~45 days lagged — a slice, not the strategy.
- Concentrated — Top disclosed positions are 50%+ of the disclosed book.
Top holdings
Dominant top position; core AI compute
Korea e-commerce
AI capex
Consumer/auto e-commerce
Cross-border e-commerce
Fintech
Recent moves
Public book is ~22% NVIDIA and concentrated in ~17 names; on the private side Dragoneer put nearly $3B into OpenAI and co-led Anthropic's $65B Series H (May 2026) — making it one of the most AI-private-levered funds here.
Loading disclosed positions…
Our take
Dragoneer is effectively a private-AI fund with a public sleeve — a 13F that's one-fifth NVDA plus direct multi-billion stakes in OpenAI and Anthropic; the positioning is a concentrated, all-in bet on the leading model labs being the generational winners.
The interesting question is not whether Dragoneer picked the right labs — it is who sets the price.
Dragoneer co-led Anthropic's Feb 2026 Series G ($30B at $380B post), then on 28 May 2026 was one of four named LEADS (with Altimeter, Greenoaks, Sequoia) of the $65B Series H at $965B post. An investor leading the round that re-marks its own prior entry is buyer, price-setter and beneficiary of the markup at once. Legal, common in late-stage AI, analytically corrosive: a ~2.54x paper gain in ~3.5 months is an artefact of a syndicate the holder helped assemble, not an arm's-length print. Second, the two legs are one trade. A 21.7% NVDA weight plus multi-billion OpenAI/Anthropic stakes are all long one variable — sustained frontier capex. If lab spend decelerates, public sleeve and private marks impair together, and the private leg can't be sold into the decline. The 35% QoQ shrink in the public book reads as liquidity routed toward the illiquid side — the opposite of the ballast you'd want against that correlation.
- 2026-07Update, Jul 2026: new capital is no longer one factor — Steadfast's retail brokerage keys off insurance-broking cash flows and Helsing ($1.8B at $18B post) off European defence budgets, not frontier capex.
Thesis
Dragoneer is not a hedge fund with an AI tilt — it is a private AI-lab holding vehicle with a shrinking public sleeve attached. The 13F ($3.00B at 03/31/2026, down from $4.59B the prior quarter) is roughly 12% of a ~$22.6–25B firm. The economically decisive positions — ~$2.8B into OpenAI and named-lead checks in both of Anthropic's 2026 rounds — appear nowhere in it. Grading Dragoneer off its disclosed equities is grading a submarine on its periscope.
Growth-oriented crossover: long-only concentrated public tech (16 names at 03/31/2026; top-5 ≈ 77% of the disclosed book) plus a much larger private growth franchise funded with what the firm calls long-duration capital, deployed via primaries, secondaries and convertible structures. Three SPACs (2020–21: CCC, Cvent, and the $400M DGNU III, which liquidated without a combination) were an earlier expression of the same public/private bridge.
- 2026-07Update, 31 Jul 2026: Form ADV now reports $37.0B regulatory AUM across 39 pooled vehicles, after the $4.3B Opportunities Fund VII close in Dec 2025. The $3.00B 13F book is ~8% of firm assets, not ~12%.
- 2026-07Update, Jul 2026: the private book is broadening past AI labs — Dragoneer joined the A$7.7B (US$5.3B) take-private of ASX-listed Steadfast, taking its retail brokerage arm, and backed Helsing's $1.8B round.
Assessment
- Structure matches the asset: long-duration, largely non-US institutional capital suits pre-IPO stakes that may not clear for years.
- Access is the real moat — named-lead status in Anthropic's $65B Series H is earned by relationship and check size, not analysis.
- Genuine conviction, not closet indexing: a 16-name book with top-5 near 77% makes the manager accountable for identifiable calls.
- Concentrating ~10% of firm funds in OpenAI is an honest expression of the stated thesis rather than a hedged view dressed up as one.
- Marks are substantially self-referential: leading the round that re-prices your own entry makes NAV a function of syndicate behaviour, not an independent bid. Legal and common — but interim gains are unrealised and endogenous.
- Correlation is disguised as diversification. NVDA at 21.7% of the public book and the OpenAI/Anthropic stakes are the same frontier-capex factor expressed twice, with only one of them sellable.
- Almost nothing about the return is externally verifiable. No audited net-of-fee series is public, so skill, leverage and beta cannot be separated at all.
- DGNU III raised $400M and liquidated without ever completing a combination — the historical record of the same late-cycle-pricing instinct, and worth weight before extrapolating the private book.
- The public sleeve falling from $4.59B to $3.00B in one quarter shrinks the liquid buffer exactly as illiquid exposure grows — a liquidity profile that only works if no LP needs cash early.
Record
There is no reliable performance figure to attribute. Dragoneer is a private fund; no audited net-of-fee return series, no YTD, no since-inception IRR/DPI is publicly disclosed, and the 13F cannot substitute — it is ~45 days lagged, longs-only, US equities, and here covers only ~12% of firm assets. What can be said: the disclosed Q1 2026 book (NVDA 21.7%, CPNG 17.5%, META 13.6%, CVNA 13.1%, GLBE 11.4%) is a high-beta growth basket whose 2024–26 result would be substantially explained by the AI and consumer-tech beta any index buyer received — the public sleeve is unlikely to be where alpha is claimed. Any headline gain on the private side is a mark, not cash: no OpenAI or Anthropic realisation has occurred, so DPI on the AI book is effectively zero and every reported increase rests on primary-round pricing rather than exit proceeds.
Risks & fit
- Frontier-capex deceleration hits public and private legs simultaneously; the private leg cannot be exited into weakness.
- A down or flat round at OpenAI/Anthropic would reverse marks non-linearly given entry at $300B (OpenAI, Aug 2025) and $380B/$965B (Anthropic).
- Key-person concentration: Dragoneer's access and allocation appear tied to Marc Stad; succession is undisclosed.
- Illiquidity/J-curve mismatch — an ~$2.8B OpenAI position reportedly near 10% of funds has no near-term liquidity path absent an IPO.
- Valuation-governance risk: repeatedly leading rounds that set one's own carrying value invites LP and auditor scrutiny.
The read that Dragoneer's AI gains are endogenous marks rather than skill would be falsified by an arm's-length liquidity event it did not help price — an Anthropic or OpenAI IPO, or a large secondary to unaffiliated buyers, clearing at or above the $965B-era entry with cash proceeds realised. Anthropic is reported to be positioning for an IPO, so this is testable. Equally falsifying: an audited net-of-fee series showing the public sleeve beating a matched high-beta growth benchmark across a cycle including 2022. A flat or down round led entirely by outsiders would confirm the concern.
Analytically relevant to anyone studying how late-stage AI valuations are actually formed — Dragoneer is a clean case study in syndicate-set marks, crossover structure, and single-factor concentration. It is a private fund whose capital is institutional and long-duration; it is not open to retail participation, and nothing here is a view on whether to seek exposure. Useful as a read-through signal on frontier-lab funding conditions, not as a portfolio to mirror.
Fee terms are not publicly disclosed. Private growth vehicles of this type commonly carry management-plus-carry structures, but no Dragoneer-specific rate is verifiable from public filings — treat any assumed 2/20 as unconfirmed.