Castle Hook Dumped $1.35B in Bitcoin and Built an $8.4B Gold Fortress: Inside the Soros Alumnus's Macro Bet That Doubled Portfolio Size
Josh Donfeld's Castle Hook Partners exited Bitcoin entirely, built a 35% gold allocation, added $8.3B in equity puts, and grew 13F AUM 74% to $24.1B in Q4 2025 — the most dramatic macro pivot of filing season.
Josh Donfeld — the former Soros Fund Management analyst who co-founded Castle Hook Partners in 2016 — just executed what may be the most dramatic macro pivot of Q4 2025 filing season. He dumped $1.35 billion in Bitcoin exposure, built an $8.4 billion gold position that now dominates 35% of his portfolio, layered on $8.3 billion in equity index puts, and grew his 13F AUM 74% in a single quarter to $24.1 billion.
This isn't a hedge fund tweaking position sizes. This is a wholesale regime bet: long gold, short equities, zero crypto.
TL;DR: The Numbers That Matter
- 13F AUM: $24.1B (up 74% from $13.8B in Q3 2025, up 197% from $8.1B a year ago)
- Top holding: GLD (SPDR Gold Trust) — $8.4B, 35.0% of portfolio (call options)
- Equity hedges: SPY puts ($4.4B, 18.1%) + QQQ puts ($3.9B, 16.3%)
- Bitcoin exit: IBIT (iShares Bitcoin Trust) completely sold — was $1.35B (9.8% of portfolio) in Q3
- AI power play exits: Vistra (VST, $658M), Talen Energy (TLN, $398M), Vertiv (VRT, $185M) — all gone
- New positions: 41 (including QQQ puts, Citigroup, Albemarle, Natera, Eli Lilly)
- Exits: 19 positions eliminated
- Holdings count: 59 (up from 36 in Q3 — portfolio nearly rebuilt from scratch)
Castle Hook Partners Top 10 Holdings — Q4 2025
The Gold Fortress: $8.4 Billion at 35% of Portfolio
Castle Hook's GLD position has been building for over a year, but Q4's scale is unprecedented. At $8.4 billion via call options, gold now represents more than a third of the entire portfolio — a staggering concentration for a multi-strategy hedge fund.
The timing looks prescient. Gold surged 65% in 2025 and hit $5,200/oz in February 2026, driven by central bank buying (1,000+ tonnes annually for three consecutive years), record ETF inflows ($19 billion in January 2026 alone — the strongest month ever), and de-dollarization trends among BRICS+ nations. J.P. Morgan forecasts gold at $5,055/oz by Q4 2026, with Goldman Sachs targeting $5,400.
For Donfeld, this isn't just a trade — it's a thesis. Castle Hook's stated strategy emphasizes "themes based on macroeconomic data and secular trends," and gold in 2025-2026 sits at the intersection of every macro tailwind: tariff chaos, Middle East tensions, a weakening dollar, and a structural shift in central bank reserve allocation away from U.S. Treasuries.
The Bitcoin Exit: From $1.35B to Zero
In Q3 2025, Castle Hook held $1.35 billion in IBIT — the iShares Bitcoin Trust ETF — making it the fund's fourth-largest position at 9.8% of the portfolio. In Q4, every single share was sold.
This is a clean, deliberate swap: institutional "digital gold" replaced by physical gold. While Bitcoin outperformed in raw returns during much of 2025, Donfeld's bet is that in a genuine risk-off macro regime — trade wars, geopolitical escalation, regulatory flux — real gold outperforms digital gold. The $5,200/oz gold price in February 2026 versus Bitcoin's continued volatility suggests the trade is working.
The Equity Put Wall: $8.3 Billion of Downside Protection
Castle Hook didn't just go long gold. The fund built a massive bearish hedge with $4.4 billion in SPY puts and $3.9 billion in QQQ puts. Combined with the gold position, 69.4% of the portfolio is now macro-directional: long commodities, short equities.
The SPY puts grew 198% from Q3 (when the fund already held some). The QQQ puts are entirely new — Castle Hook had no Nasdaq exposure on the short side in Q3. This is a layered bet: not just "equities might dip" but "both broad market and tech specifically face structural headwinds."
In context, the S&P 500 and Nasdaq are both cooling from 2025 highs, and the Supreme Court's recent tariff ruling created what analysts call a "regulatory vacuum" — exactly the kind of policy uncertainty that makes equity puts valuable.
Castle Hook Q4 2025 Portfolio Structure — Macro vs. Stock-Picking
The AI Power Play Exit: Clean Break
Q3 2025 told a very different story. Castle Hook's top holding was Vistra Energy at $658M (3.7%), riding the AI-driven power demand narrative. Talen Energy ($398M), Vertiv ($185M), and Meta ($111M) rounded out a significant tech-infrastructure bet.
In Q4, every single one of these positions was sold — $1.35 billion of AI/power exposure eliminated. This is the mirror image of what funds like Citadel and Coatue were doing in the same quarter, where semiconductor and AI infrastructure remained core holdings.
Donfeld isn't saying AI is over. He's saying the easy money in AI power plays has been made, and the risk-reward has flipped. When Lone Pine exits META and Altimeter slashes Broadcom, it's a signal. When a Soros-trained macro trader exits the entire sector and builds a gold fortress, it's a conviction call.
41 New Positions: The Bottom-Up Sleeve
Beneath the macro overlay, Castle Hook opened 41 new positions — nearly rebuilding the entire portfolio from scratch. The new additions reveal specific sector bets:
| Sector Theme | Tickers | Combined Value |
|---|---|---|
| Banks & Financials | C, KKR, ARES | $638M |
| Healthcare & Biotech | NTRA, LLY, INSM, DXCM | $895M |
| Materials & Commodities | ALB, CLF, FCX, AA, SQM | $1.3B |
| Consumer / Cyclical | CVNA, WMT, DKS, TSCO | $692M |
| Solar & Infrastructure | FSLR, FIX, SHLS | $406M |
| EV / Momentum | TSLA, DASH | $325M |
The materials sleeve is particularly telling. Albemarle ($598M), Cleveland-Cliffs ($240M), Freeport-McMoRan ($180M), and Alcoa ($211M) are all commodity producers that benefit from the same macro regime as gold: inflation, tariffs, supply constraints. This isn't diversification — it's amplification of the macro thesis.
Biggest Q3 → Q4 2025 Portfolio Moves ($M)
AUM Trajectory: From $630M to $24.1B in Three Years
Castle Hook's AUM growth has been explosive. In Q3 2022, the fund held just $630 million in 13F assets. Today it holds $24.1 billion — a 38x increase. The growth accelerated sharply in 2025:
- Q4 2024: $8.1B
- Q1 2025: $9.7B (+19%)
- Q2 2025: $17.6B (+82%) — first major options buildup
- Q3 2025: $13.8B (-21%) — trimmed after summer volatility
- Q4 2025: $24.1B (+74%) — current, all-time high
The Q2-to-Q4 trajectory tells the story: Castle Hook has been building the gold-and-puts macro structure for multiple quarters, but Q4 represents the full expression of the thesis.
Castle Hook Partners 13F AUM History (2021–2025)
Q&A: What Investors Are Asking
Why did Castle Hook sell all its Bitcoin?
Castle Hook held $1.35B in IBIT (iShares Bitcoin Trust) in Q3 2025 and sold every share in Q4. The timing suggests a strategic preference for physical gold over digital gold in a regime of tariff uncertainty, geopolitical risk, and de-dollarization — conditions where gold's 3,000-year track record as a store of value outweighs Bitcoin's volatility premium.
How large is Castle Hook's gold position?
$8.4 billion in GLD call options, representing 35% of the $24.1B portfolio. This makes Castle Hook one of the largest single-fund gold holders in the 13F universe.
What does the SPY and QQQ put position mean?
Castle Hook holds $4.4B in SPY puts and $3.9B in QQQ puts — a combined $8.3B bearish equity bet (34.4% of portfolio). These are options that increase in value when the S&P 500 and Nasdaq decline. Combined with the gold longs, 69.4% of the portfolio is a macro directional bet: inflation up, equities down.
Who is Josh Donfeld?
Josh Donfeld is the co-founder of Castle Hook Partners, established in 2016. He is a Princeton graduate who previously worked as an analyst at Soros Fund Management, and held positions at Canyon Capital Advisors and Morgan Stanley. Castle Hook manages approximately $6.6 billion in discretionary AUM (per Form ADV), with a strategy that combines top-down macro analysis with bottom-up security selection.
Is Castle Hook a Tiger Cub?
No. Castle Hook is a Soros-lineage fund — Donfeld's formative experience was at Soros Fund Management, not Tiger Management. The fund's macro-heavy approach reflects this heritage: few Tiger Cubs run 35% gold positions or build $8.3B put walls. The style is closer to Druckenmiller's Duquesne than to Tiger offspring like Lone Pine.
What sectors did Castle Hook add in Q4?
The 41 new positions span materials (ALB, CLF, FCX, AA — $1.3B), healthcare (LLY, NTRA, INSM, DXCM — $895M), financials (C, KKR, ARES — $638M), and consumer cyclicals (CVNA, WMT, DKS — $692M). The commodity exposure amplifies the gold thesis.
Related Research
Explore all researchWhile most mega-filers lead with NVIDIA, Capital International Investors bets biggest on Broadcom at 7.7% of portfolio. They also increased Netflix shares by 710% in Q4 2025, added $3.4B in TotalEnergies, and exited Disney entirely.
Mar 6, 2026
The second Capital Group division to file a massive Netflix build in Q4 2025. Capital Research Global also doubled its Applied Materials position, holds $26B in Eli Lilly, and opened a $1B position in MicroStrategy — the Bitcoin proxy.
Mar 6, 2026
GROUPAMA ASSET MANAGMENT's Q4 filing shows a dramatic AUM jump versus Q3 and a broad position reset centered on U.S. mega-cap tech.
Mar 6, 2026
BSN Capital Partners reported a compact 24-position portfolio with NVIDIA at 21.8% and top-5 concentration above 61%.
Mar 6, 2026
Franklin Resources — the parent of Franklin Templeton, one of the world's oldest fund families — files $408B with 14,263 positions. Its Q4 2025 13F has Microsoft over NVIDIA at #1, a massive $4B Citigroup bet, Cisco built 18%, and Exxon in the top 10. This is active management that refuses to follow the momentum crowd.
Mar 4, 2026