Two Sigma
Co-founded by David Siegel and John Overdeck (both ex-D.E. Shaw; Overdeck a former math olympiad medalist). Data-science-first systematic firm.
A data-science-and-ML-driven systematic manager, peer to RenTec and D.E. Shaw. The 13F long book is dominated by exactly the AI trade — NVDA is the single largest reported position, followed by Apple, Amazon, Alphabet and Tesla (Mag 7 across the top). The reported portfolio value roughly doubled QoQ (to ~$124B), reflecting gross/leveraged positioning and option overlays rather than a doubling of conviction. Read the holdings as factor/ML output.
- 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.
Top holdings
Largest reported 13F position Q1 2026 — the AI-trade anchor of the long book.
Mag-7.
Mag-7; AWS AI/compute exposure.
Mag-7; Gemini/TPU AI exposure.
Mag-7.
AI-cluster connectivity (active electrical cables / SerDes)
Recent moves
Q1 2026 13F (period ending Mar 31, 2026): NVDA the top position; broad Mag-7 long exposure; reported gross book ~$124B vs ~$71B prior quarter. Source: Holdings Channel / HedgeFollow.
Loading disclosed positions…
Our take
Two Sigma's visible long book IS the AI trade — NVDA #1, Mag-7 across the top — but that is a systematic ML/factor model expressing crowded large-cap exposure, not a discretionary thesis, so don't over-read it. The QoQ near-doubling of reported value is gross/leverage optics, not a 2x conviction increase. Honest limit: 13F is long-only US equity; Two Sigma's actual book is heavily hedged with shorts and derivatives invisible here, so net AI exposure is far smaller than the gross long suggests.
The 13F is a trap. Two Sigma reported $123.86B of long US equity at 03/31/2026 — against ~$70B firm AUM — but the same filing showed $71.08B a quarter earlier.
A 74% jump in 90 days looks more like a filer-aggregation change than a leverage change, and Two Sigma Securities files separately, blurring market-making inventory with fund positioning. Regardless, reading NVDA-at-#1 and a Mag-7 top five as an AI conviction bet is a category error: these are longs in a hedged machine whose shorts, futures and options never appear. Governance is the sharper critique, and it is now on the record. The founders' arbitration closed around April 2026 with all claims and counterclaims dismissed — but the panel found their disputes "caused management dysfunction." A second arbitration is live: Siegel's appointee Seth Platt moved to remove Carter Lyons; Overdeck blocked it. Co-CEO Scott Hoffman resigned March 31, 2026, citing "ongoing governance challenges." A systematic firm is meant to engineer key-person risk away.
Thesis
Two Sigma sells process — a data-science factory where no human discretion touches a trade — and the most instructive fact about it is that the process broke. The SEC found employees identified vulnerabilities in live investment models by March 2019 and the firm did not address them until August 2023 (settled Jan 16, 2025: $165M repaid to affected funds, $90M in penalties, plus a whistleblower-rule violation). For a firm whose product IS model governance, a four-year unremediated flaw is not a compliance footnote; it is a hit on the thing being sold.
Systematic, ML- and alternative-data-driven multi-strategy trading across equities, futures and other liquid assets, run as a research pipeline: thousands of signals combined by models, executed at scale with leverage and hedges. Founded by David Siegel and John Overdeck (both ex-D.E. Shaw). ~1,700 staff and more than two dozen funds as of late 2025; a separate arm, Two Sigma Securities, is a market maker and files its own 13F. Capital allocation is model output, not thesis.
Assessment
- Real research infrastructure — signal generation, backtesting and execution at a scale few can staff, spreading risk across many uncorrelated bets rather than a few calls.
- Diversification by construction: no single-name blowup should matter, which is precisely why the 13F top names say nothing about actual risk.
- Demonstrated fundraising capacity — Titan, the 2025 multistrategy launch, took in more than $1B on its own.
- Repaid $165M to affected funds during the investigation rather than litigating — the better of the available responses once the flaw surfaced.
- A known model flaw left live from March 2019 to Aug 2023 undercuts the core claim that systematic beats discretionary because the process is auditable.
- Impeding employee communication with the SEC (Rule 21F-17) is a culture signal, not a paperwork slip — it is how flaws stay unfixed for four years.
- An arbitration panel, not a journalist, found the co-chairmen's disputes "caused management dysfunction" — and a second arbitration over committee seats is live.
- The 2025 launch tally flatters: ~88% of the >$1.1B went to Titan; Aurora (~$70M) and Beacon ($54M) are immaterial at a ~$70B firm.
- The reported long book ($123.86B) exceeds firm AUM and near-doubled in one quarter — the 13F is unusable as a read on positioning.
Record
Reported figures are self-disclosed and unaudited: Spectrum +3% and Absolute Return +3.7% in Q1 2026; Absolute Return Enhanced ~+13% through Nov 2025; for 2024, Spectrum +10.9% and ARE +14.3%. Respectable, not extraordinary, and earned in years when equity beta and momentum were paid — skill vs leverage vs factor exposure cannot be settled from outside, because Two Sigma publishes no net-of-fee series, exposures or risk decomposition. The SEC facts sharpen this: between Nov 2021 and Aug 2023 an employee made unauthorised changes to model parameters; some funds consequently overperformed by ~$400M while others underperformed by ~$165M, and only the losers were made whole. That asymmetry means reported fund-level returns in that window are partly an artefact of an uncontrolled process. The $165M is ~25bp against a ~$65B base in aggregate, but concentrated in specific vehicles.
Risks & fit
- Alpha decay: ML signals mined from widely available data are increasingly crowded, and capacity at ~$70B works against per-dollar returns.
- Leverage: gross notional well above net exposure means a factor shock or deleveraging cascade transmits far more than headline positioning implies.
- Governance unresolved — a live second arbitration over management-committee seats, a co-CEO gone March 2026, and the firm's own filing says resolution may never occur.
- Regulatory overhang: a settled fiduciary-duty and whistleblower case raises the bar on any future supervisory lapse.
- Model risk from insiders — the underlying episode involved unauthorised employee modifications to live trading models over ~21 months.
Our durability skepticism is wrong if Two Sigma posts consistent returns with low correlation to equity and momentum factors while AUM keeps growing — that would show capacity was not binding and the pipeline regenerates edge faster than it decays. The governance critique is falsified if the second arbitration resolves, the committee stabilises, and the firm publishes verified risk decompositions showing returns are not levered factor beta. Both require disclosure Two Sigma has not historically provided — which is itself the point: from outside, the thesis is currently unfalsifiable on available evidence.
An analytical critique for readers studying how large systematic managers work and fail — why 13F data is near-worthless for a quant firm, how model-governance risk differs from market risk, what founder instability means at a process-driven shop. Not investment advice and no view on allocating. Two Sigma's funds are private vehicles closed to the general public; nothing here suggests the reader can or should invest.
Fee terms are not publicly disclosed — Two Sigma publishes no management/incentive rates or pass-through terms for its fund series, so any figure would be an assumption and we make none. Treat reported returns as net only where the source says so.