Citadel
Ken Griffin (founder/CEO since 1990). Multi-strategy giant; the most profitable hedge fund in history (~$90B net gains since inception).
A multi-strategy/multi-PM platform — quantitative and fundamental blended — not a pure systematic shop, but its 13F is one of the largest and most options-heavy filed. The visible book is dominated by index-option overlays (SPY/QQQ puts and calls) and large single-name positions including NVDA, Tesla, Apple, Microsoft and NVDA calls. The ~$618B reported value is gross notional (options counted at notional), wildly overstating directional AI exposure. Treat as platform-wide flow, hedges, and many independent PM books netted together.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Self-reported — Headline returns are manager- or firm-reported and unaudited.
Top holdings
Among largest reported equity positions Q1 2026; also holds NVDA call options — the AI-trade core.
Largest notional lines — index hedges/overlays — flow and hedging, NOT a bearish conviction call.
Top holding — Mag-7; large call-option line.
Top-5 — Mag-7.
Top-5 — Mag-7; Azure/OpenAI AI compute exposure.
NAND/memory — AI-driven storage demand (post-WD spinoff)
Recent moves
Q1 2026 13F (filed May 15, 2026): top lines are SPY/QQQ puts, SPY/TSLA/NVDA calls; large NVDA/TSLA/AAPL/MSFT equity. Returned ~$5B of 2025 profits to investors. Source: 13f.info / CNBC. On or about 30 Jul 2026, reportedly bought the bulk of Situational Awareness LP's broker-financed public equity book in a single overnight block at a discount, after that fund's prime brokers margin-called a position reported levered ~4x; Millennium reportedly bid and lost. Reuters, Bloomberg and CNBC report it on unnamed sources — no filing corroborates it and no price or size is disclosed. This was a purchase of POSITIONS, not of the fund, its management company or any GP/LP interest.
Loading disclosed positions…
Our take
Citadel is multi-strat (quant + fundamental), so its 13F is the noisiest of this set — option notionals inflate the headline to ~$618B and the SPY/QQQ puts are hedges/flow, not directional bets. Real AI exposure is there (NVDA equity + NVDA calls, Mag-7), but it is the sum of many independent PM books plus market-making/hedging, so single-name reads are unreliable. Honest limit: the strong systematic-vs-discretionary caveat applies doubly — this is a platform, not one mind, and the 13F omits all shorts.
The number that teaches most isn't the headline return — it's the wedge between gross P&L and what LPs keep.
In 2025 Citadel generated ~$11.1B of net investment gains while operating costs across the three main multi-strat funds rose ~4% to ~$4.5B, with compensation ~$3.8B roughly flat despite gains falling 14%. Per Bloomberg's 2025 fee analysis, the three largest funds charged ~$12.5B of pass-through expenses across 2022–Sep 2024, ~$11B of it employee comp; over a longer window investors netted ~$30B after ~$7.5B of management/performance fees and ~$17B of pass-through — we could not independently confirm those aggregates outside that single paywalled source. That is roughly half the economics staying inside the building, and the pass-through is uncapped, so a flat year still bills a full cost base. The strategy works; the open question is whether the after-cost share of it is what a reader thinks they are looking at. Figures as reported/press-sourced through Jul 2026; Citadel's funds are not audited public filers.
Thesis
Citadel is best understood as a capacity-managed risk factory, not a strategy. Roughly 200+ semi-autonomous PM pods run tightly risk-budgeted books; the firm's real product is centralised risk management, leverage allocation and a talent market — not any single insight. That makes the edge unusually durable (it doesn't decay when one signal crowds out) but structurally expensive, and it caps at the capital the risk system can absorb. The recurring capital returns are the clearest evidence the constraint is real.
Multi-strategy, multi-PM platform across five broad businesses (equities, commodities, fixed income & macro, credit, quantitative). PMs run market-neutral-ish books under hard drawdown and factor limits; the centre allocates gross leverage (platform-wide gross typically many multiples of equity), nets exposures, and fires underperformers fast. Blended quant + discretionary — the 'tactical trading' sleeve explicitly fuses discretionary stock selection with systematic models.
Assessment
- Pod structure with hard risk limits means no single crowded signal is existential — edge is organisational, not one decaying alpha
- Returns capital rather than gorging on AUM: ~$5B returned entering 2026 (~$32B since 2017), an honest capacity signal most managers won't send
- Tactical trading (+14.3% H1 2026) sidestepped the late-June systematic unwind — evidence the risk centre actually nets and de-crowds, not just aggregates
- Talent market is the moat: pass-through comp lets it outbid rivals for PMs, which is self-reinforcing while performance holds
- Uncapped pass-through expenses: comp held ~$3.8B in 2025 while net gains fell 14% — cost base is sticky when returns aren't
- Scale drag is visible — Wellington +10.2% (2025) and +5.7% (H1 2026) sit well below the ~19% since-1990 lore built at far smaller size
- The ~19% annualized figure blends a tiny-AUM 1990s with today's platform; it is not a forward-looking expectation
- Key-person concentration in Ken Griffin (founder/CEO since 1990) with no publicly articulated succession plan
- 13F barely informs: ~$618B 'value' counts options at notional, longs-only US equities, shorts omitted; the latest filing (Q1 2026) is ~113 days stale as of 22 Jul 2026
Record
Wellington (flagship multi-strat) returned +10.2% in 2025 — its weakest since 2018 — and +5.7% in H1 2026. The dispersion inside the platform is the real story: H1 2026 saw tactical trading +14.3%, equities +11.2%, global fixed income roughly flat. Firmwide AUM (all Citadel funds, not Wellington alone) was ~$67B entering 2026 after returning ~$5B of profits, ~$69B by early June; Wellington's standalone AUM is not public. Attribution: far more skill-plus-leverage than beta — pods run low net exposure, so returns come from many small, levered, weakly-correlated bets, not market direction. Leverage is doing real work, and the blended flagship number is diluted by design. The ~19% since-1990 figure is firm-reported and reflects vastly smaller capital — history, not a run-rate. All returns are manager-reported via press (CNBC/Bloomberg/Hedgeweek), not audited filings.
Risks & fit
- Deleveraging shock: platform gross leverage means a correlated multi-pod drawdown forces fast risk-off, as in Aug 2007 and Mar 2020 analogues
- Crowding across platforms — Citadel, Millennium, Point72 and Balyasny hire from the same pool and often hold the same relative-value trades
- Pod-model talent inflation: comp is the largest cost and is pass-through, so investor economics compress if PM bidding wars continue
- Griffin key-person and succession risk at a firm whose culture and capital allocation still run through the founder
- Closed/constrained access — the funds are not broadly open, so the record is largely unavailable to outside capital regardless of view
The 'edge is organisational and durable' read breaks if: (a) two-plus consecutive years of Wellington below its cash-plus-spread hurdle while pass-through costs stay flat — showing the cost base, not the alpha, is the constant; (b) a correlated multi-pod drawdown exceeding the firm's historical worst, proving the risk centre nets less than claimed; or (c) a resumption of large net capital raising instead of profit returns, signalling the capacity discipline was marketing. Sustained double-digit net returns at $70B+ with continued returns of capital would strengthen it.
Analytical interest only — Citadel's funds are not a vehicle available to general investors. This is a study of how the multi-manager platform model works, what it costs, and where it is fragile. The transferable lesson: judge any pod-platform on after-all-cost share to LPs rather than headline return, on whether it returns capital when capacity binds, and on intra-platform dispersion rather than the blended flagship number. Not a recommendation to buy, sell, or allocate.
Pass-through model, not classic 2/20: investors reimburse essentially all fund operating costs (effective management-fee-equivalent commonly cited at 3–6%, not disclosed) plus ~20% performance. Pass-through is uncapped and ~90% of it is employee comp.