## Executive assumption set
- **Horizon:** 6 months core, with a 3-month review window.
- **Risk budget:** max **6% NAV drawdown** on this thesis package.
- **Objective:** express Citrini’s public thesis as a **late-cycle rotation / AI-infra peak-capex unwind**: short yesterday’s crowded AI/high-beta winners; long lower-beta, cash-flow, defensive/real-economy exposures.
- **Caveat:** Citrini’s exact “Citrindex, Reloaded” holdings are behind a paid section, so I’m using the public article’s thesis and not assuming hidden names. The public post says it is “Paid” and cuts off before the reloaded portfolio. (
citriniresearch.com)
---
# Phase A: UNDERSTAND
## 1. Thesis Parser
**Your thesis:** “Citrini is right.”
The public article’s actionable message is:
1. **The broad thematic bull trade is mature.** Citrini says the easy 2023–2025 era of “structural tailwinds + cheap stocks + rising tide” is over, and the portfolio is being rebuilt around “reversals, pairs, and entirely fresh ideas,” rather than broad continuation trades. (
citriniresearch.com)
2. **AI is not dead, but the AI-infrastructure trade has worse asymmetry.** Citrini’s key line is not “AI adoption stops”; it is that current AI capex expectations require extreme revenue growth, and the risk has shifted from upside surprise to downside in capex growth, funding, and execution. (
citriniresearch.com)
3. **Peak AI capex growth / funding stress is the central pressure point.** The article argues hyperscaler capex is approaching a huge scale, revenue must catch up, financing appetite is fading, and data-center bottlenecks/policy are moving from “momentum narrative” to real constraints. (
citriniresearch.com)
4. **Macro is late-cycle, not immediate recession.** Citrini argues the economy still has expansionary and inflationary signs, but asset behavior is tilting stagflationary, with low-beta beginning to outperform high-beta. (
citriniresearch.com)
5. **Trade style changes:** less “long big theme beta,” more **pairs, reversals, low-beta, defensive rotation, and selective overlooked sectors**. Citrini explicitly notes high-beta has gone from a bull-market feature to a bear-market failure point. (
citriniresearch.com)
**Parsed thesis in one line:**
> Own the next regime: **long low-beta / defensive / real-economy / cash-yield exposures, short crowded AI infrastructure, semis, neoclouds, high-multiple AI software, and momentum beta.**
---
## 2. Scenario Formalizer
| Scenario | Probability | Market description | What should work |
|---|---:|---|---|
| **Base case: AI capex peak + rotation** | 55% | AI spending continues but second derivative rolls over; funding costs matter; semis/high-beta derate; defensives and cash-flow sectors outperform | Long XLV/XLP/XLU/XLF/KIE/SGOV; short SMH/SOXX/QQQ/NVDA/PLTR/ORCL/AI beta |
| **Hard AI bust / liquidity break** | 20% | AI funding freezes, capex gets cut, major AI complex drawdown | Short AI basket wins big; long defensive sleeve may fall but less |
| **Wrong-way AI reacceleration** | 25% | AI revenue/capex validates, funding reopens, NVDA/semis squeeze higher | Trade loses; risk controls required |
---
# Phase B: MAP THE BUTTERFLY EFFECT
## 3. Consequence Graph
**Causal chain:**
1. **AI capex growth peaks**
→ hyperscaler / neocloud forward estimates revised down
→ semis, memory, AI equipment, neoclouds, AI power winners lose multiple.
2. **Funding pressure rises**
→ more equity/debt issuance competes for capital
→ crowded AI winners face dilution / credit / liquidity repricing. Axios recently reported that Alphabet, Amazon, Meta, Microsoft, and Oracle had raised $326B this year through stock and bond issuance for AI. (
axios.com)
3. **Data-center bottlenecks and politics bite**
→ permitting, grid, labor, interconnects, ratepayer backlash
→ project timelines stretch
→ capex suppliers see order risk. Voters also just approved data-center moratoriums and elected candidates favoring slower data-center buildout in key states, according to Axios. (
axios.com)
4. **Macro stays inflationary / late-cycle**
→ Fed not free to ease aggressively
→ long-duration growth multiples remain vulnerable. September PCE rose 0.5% m/m and 3.6% y/y, while the Fed funds target range is 3.75%–4.00%. (
bea.gov)
5. **Market rotates from beta/momentum to quality/defense**
→ staples, health care, utilities, insurance/financials, short-duration cash, and select energy outperform on a relative basis.
---
## 4. Instrument Mapper
| Consequence | Best public-market expression | Crypto/perp venue expression | Prediction-market expression |
|---|---|---|---|
| AI infra derating | Short **SMH/SOXX**, **NVDA**, **AMD**, **AVGO**, **ORCL**, **PLTR**, **IGV** | Short Hyperliquid **NVDA**, **SMH**, **SOXL**, **DRAM**, **AMD**, **ORCL**, **PLTR**, **CRWV**, **NBIS** where legally accessible | Buy YES on “AI bubble burst by June 30, 2027” if rules fit |
| Nasdaq / high-beta unwind | Short **QQQ/NQ** or QQQ put spreads | Short **XYZ100** / Nasdaq-100 perp | Buy lower NVDA market-cap buckets; NO on NVDA largest-company outcomes |
| Low-beta rotation | Long **XLP**, **XLV**, **XLU**, **KIE/XLF**, **SGOV/TBIL** | Longs are less complete on Hyperliquid; possible proxies include **COST**, **LLY**, **MRNA**, **XBI**, **XLE**, **NATGAS**, **TLT** | Not well represented |
| Inflation / late-cycle | Long cash bills, selective energy/nat gas, financials | Long **NATGAS**, **XLE**, maybe **CL/BRENTOIL** tactically | YES on Fed hike only if you think odds are too low |
| Hard bust hedge | SMH/QQQ put spreads, VIX calls, small TLT calls | Use low leverage; avoid liquidation | “AI bubble burst” market gives convex payoff |
---
# Phase C: ESTABLISH THE EDGE
## 5. Market Baseline / Pricing
Current tape confirms the internal rotation is already visible but not fully complete:
| Instrument | Latest observed price / move | Read-through |
|---|---:|---|
| **SPY** | $766.81, +1.63% | broad market still resilient |
| **QQQ** | $691.95, +0.96% | megacap tech not broken yet |
| **SMH** | $504.27, -2.72% | semis under pressure despite green index tape |
| **SOXX** | $450.41, -3.00% | confirms semi weakness |
| **NVDA** | $200.13, -3.10% | core AI leader is trading weak |
| **XLP** | $91.94, +3.37% | staples rotation already active |
| **XLV** | $189.66, +3.06% | health care rotation active |
| **XLF** | $59.20, +3.47% | financials participating |
| **UNG** | $9.27, +1.64% | nat gas can be a small inflation/power-demand sleeve |
**Pricing conclusion:**
The market is already sniffing the trade, but the **index has not yet de-risked enough** to make this “too late.” The edge is not “semis are down one day”; the edge is **a regime shift from high-beta thematic continuation to lower-beta relative value.**
---
## 6. Historical Analog Finder
Useful analogs:
1. **Dot-com / telecom capex overbuild:** Nasdaq peaked on March 10, 2000, and fell 77% by October 2002. Goldman’s historical summary says Nasdaq rose 86% in 1999 alone before the 2000 peak. (
goldmansachs.com)
2. **2022 duration-growth reset:** QQQ fell 36% peak-to-trough while SPY fell about 25% during the Fed hiking cycle. (
kiplinger.com)
3. **Unprofitable / speculative growth unwind:** ARKK’s drawdown from February 2021 to December 2022 was about -80.9%. (
medgecapital.com)
4. **Railway-style infrastructure overbuild:** historical railway mania research emphasizes that the real loss often came not only from the stock bubble but from huge capital poured into projects that later failed to earn expected returns. (
tandfonline.com)
**Analog lesson:**
The winning trade is usually not “short technology forever.” It is **short the over-owned beneficiaries of capex extrapolation** and **long the parts of the market that survive when discount rates and funding discipline return.**
---
## 7. Response / Beta Estimator
My estimated factor sensitivities:
| Sleeve | Estimated beta to SPY | Estimated sensitivity to “AI capex shock” | Comment |
|---|---:|---:|---|
| Short SMH/SOXX/NVDA/AI basket | 1.3–1.7 | Very high | best direct expression |
| Short QQQ / XYZ100 | 1.1–1.3 | Medium-high | cleaner and liquid but less pure |
| Long XLV/XLP/XLU | 0.5–0.8 | Low / inverse relative | lower beta and defensive |
| Long XLF/KIE | 0.9–1.1 | Mixed | benefits from rates but watch credit |
| Long UNG/XLE | 0.7–1.0 | Mixed | inflation hedge; avoid over-sizing |
| Long SGOV/TBIL | ~0 | none | dry powder + yield |
---
## 8. Trade Backtester / Scenario Model
I ran a simple scenario model using a **$100 NAV portfolio** with **$80 long notional**, **$55 short notional**, and cash collateral. This is not a tick-level backtest; it is a deterministic stress test calibrated to the analogs above.
| Scenario | Long sleeve P&L | Short sleeve P&L | Total P&L |
|---|---:|---:|---:|
| Base: AI capex peak + rotation | +5.1% | +10.6% | **+15.6%** |
| Hard AI bust / liquidity break | -4.7% | +19.8% | **+15.0%** |
| Wrong: AI reaccelerates / short squeeze | +2.3% | -13.3% | **-11.1%** |
| Broadening rally, AI flat-up | +6.6% | -3.8% | **+2.8%** |
| Stagflation / rate shock | +2.6% | +9.2% | **+11.8%** |
**Key risk:** the trade loses most when **AI reaccelerates and defensives lag**. Therefore the expression should be **paired, staged, and stop-managed**, not a max-leverage outright short.
---
# Phase D: BUILD THE TRADE
## 9. Expression Selector
Best expression hierarchy:
### Best core expression: liquid equity long/short basket
Use listed ETFs and single-name shorts.
### Cleaner risk-defined version
Use **put spreads** on SMH/QQQ/NVDA rather than outright shorts.
### Crypto/perp expression
Use Hyperliquid only for the **short AI/high-beta sleeve**, because the long defensive sleeve is less complete there and U.S. access is legally restricted.
### Prediction-market expression
Use Polymarket only for **convex event exposure**, not as the whole trade.
---
## 10. Basket Optimizer
### Model basket per $100 NAV
**Long sleeve: $80 notional**
| Long | Weight |
|---|---:|
| XLV / health care | $18 |
| XLP / staples | $18 |
| XLU / utilities | $12 |
| XLF or KIE / financials-insurance | $12 |
| UNG or XLE / nat gas-energy | $8 |
| SGOV/TBIL / cash bills | $12 |
**Short sleeve: $55 notional**
| Short | Weight |
|---|---:|
| SMH or SOXX / semis | $20 |
| QQQ / Nasdaq beta | $12 |
| IGV / software | $8 |
| NVDA | $8 |
| PLTR / ORCL / AMD basket | $7 |
**Portfolio stats:**
- Gross notional: **135% NAV**
- Net notional: **+25% NAV**
- Estimated beta: roughly **-0.20 to -0.35**
- Expected payoff: positive in rotation, hard AI bust, and stagflation; negative in AI melt-up.
---
## 11. Risk & Sizing
For a **$1,000,000 account**, implement as:
| Sleeve | Dollar notional |
|---|---:|
| Long XLV | $180,000 |
| Long XLP | $180,000 |
| Long XLU | $120,000 |
| Long XLF/KIE | $120,000 |
| Long UNG/XLE | $80,000 |
| Long SGOV/TBIL | $120,000 |
| Short SMH/SOXX | $200,000 |
| Short QQQ | $120,000 |
| Short IGV | $80,000 |
| Short NVDA | $80,000 |
| Short PLTR/ORCL/AMD | $70,000 |
**Risk controls:**
- **Initial max thesis loss:** 6% NAV.
- **Review level:** -3% NAV.
- **Cut level:** -6% NAV.
- **Take-profit 1:** +8% to +10% NAV, cover 25%–35% of short sleeve.
- **Take-profit 2:** if SMH falls 20% or QQQ falls 12% from entry, cover another 25%.
- **Invalidation triggers:**
- NVDA and SMH reclaim momentum leadership together.
- Hyperscaler funding markets reopen strongly.
- AI capex guidance rises without bond/equity issuance stress.
- Polymarket AI-bubble odds collapse while NVDA market-cap upside odds expand.
---
# Phase E: ACT
## 12. Trade Proposal / Execution Layer
### Primary execution plan
1. **Enter 50% now.**
2. **Add 25% on an AI relief rally**: SMH +4% to +7% or NVDA +5% to +8% from entry.
3. **Add final 25% on confirmation**: SMH makes a lower high or QQQ underperforms SPY for 5–10 trading days.
4. **Use limit orders, not market orders.**
5. **Prefer put spreads for single-name shorts** if borrow, squeeze, or earnings risk is high.
### Options overlay
If using options instead of outright shorts:
- Buy 4–6 month **SMH 10% OTM / 25% OTM put spread**.
- Buy 4–6 month **QQQ 8% OTM / 18% OTM put spread**.
- Add a smaller **NVDA put spread** only if you accept single-name event risk.
- Fund part of premium by trimming gross short notional, not by selling naked upside calls.
---
# Final: Hyperliquid and Polymarket
## Hyperliquid: what to trade
**Important U.S. caveat:** Hyperliquid’s interface terms are described in an SEC filing as not available to U.S. persons; using a VPN to circumvent restrictions can create regulatory/enforcement risk. (
sec.gov) If you are in the U.S., do not assume you can legally use it.
If legally accessible, Hyperliquid’s HIP-3 design allows builder-deployed non-crypto perps, including traditional assets, commodities, and indices; HIP-3 markets are isolated-margin initially and deployers set oracle specs, leverage limits, and liquidity. (
hyperliquid-co.gitbook.io)
**Most direct Hyperliquid expression:**
| Action | Instrument | Why |
|---|---|---|
| Short | **NVDA PERP** | direct AI leader exposure; large volume/OI listed on Hyperliquid frontends (
app.syndicate.trade) |
| Short | **SMH PERP** | cleaner semi basket; Perpys lists SMH as a Hyperliquid ETF perp (
perpys.com) |
| Short | **SOXL PERP** | aggressive semi-beta; use small size because it tracks a leveraged ETF/perp and can be path-dependent (
app.syndicate.trade) |
| Short | **DRAM PERP** | memory-cycle / AI supply-chain exposure; DRAM is listed with high volume in Hyperliquid market trackers (
perpys.com) |
| Short | **XYZ100 PERP** | Nasdaq-100 beta / broad AI-duration short (
app.syndicate.trade) |
| Short | **ORCL, AMD, PLTR, CRWV, NBIS** | AI funding / neocloud / high-multiple AI complex; several are listed in Hyperliquid market dashboards (
perpys.com) |
| Long small | **NATGAS PERP** | inflation / power-demand hedge; NATGAS is listed on Hyperliquid dashboards (
hyprswarm.com) |
| Long small | **XLE PERP** | energy sleeve if you want inflation protection on-venue (
usekeel.io) |
| Long small / hedge | **TLT PERP** | only as a hard-bust hedge, not core stagflation exposure (
perpys.com) |
**Hyperliquid implementation I would use if eligible:**
- Short 30% NVDA
- Short 25% SMH or SOXL risk-adjusted equivalent
- Short 20% XYZ100
- Short 15% DRAM
- Short 10% ORCL/PLTR/CRWV/NBIS basket
- Optional hedge: long 10% NATGAS or XLE, funded by reducing XYZ100 short
Use **≤2x effective portfolio leverage** even if the interface shows much higher max leverage.
---
## Polymarket: what to trade
**Important U.S. caveat:** Polymarket.com currently shows U.S. users that trading is blocked and directs them to Polymarket US; the international platform says it is not CFTC-regulated and operates independently. (
polymarket.com)
### Best direct Polymarket trade
**Buy YES: “AI bubble burst by June 30, 2027”**
- Recent odds are around **23%–24%**. (
polymarket.com)
- This is the cleanest event-market expression of the Citrini thesis, but the rules are strict: the market requires an “AI industry downturn” defined by at least **three qualifying events within 90 days**, including conditions such as NVDA down 50% from ATH, SOXX down 40% from ATH, OpenAI/Anthropic bankruptcy or acquisition, H100 rental price collapse, or major AI hardware supplier collapse. (
predictdesks.com)
**Sizing:** max **0.50%–1.00% NAV premium**. Treat it as a convex lottery, not a substitute for the equity long/short trade.
### Secondary Polymarket trades
| Trade | Why | Issue |
|---|---|---|
| Buy YES on **AI bubble burst by Dec. 31, 2026** | Higher convexity / near-term crash bet | Timing is too tight; small only |
| Buy lower buckets in **NVDA market cap end-2026** | Directly expresses NVDA derating | Liquidity is limited; end-date risk |
| Buy NO on high NVDA market-cap buckets | Cleaner if available with tight spread | Order-book dependent |
| Buy NO on **NVIDIA as largest company** outcomes | Captures AI leadership reversal | Resolution can be affected by MSFT/AAPL moves too |
| YES on **Fed December 25 bps increase** | Aligns with late-cycle inflation thesis; Polymarket showed 78% YES and $13M volume | Already crowded; not much edge unless you think true odds are >85% (
polymarket.com) |
**My Polymarket pick:**
- **Primary:** YES “AI bubble burst by June 30, 2027” at ~23%–24%, small size.
- **Avoid making Polymarket the main trade** because the resolution rules are too discontinuous versus the smoother equity thesis.
---
## Final model trade
If I had to compress everything into one implementable package:
> **Long:** XLV, XLP, XLU, XLF/KIE, SGOV/TBIL, small UNG/XLE.
> **Short:** SMH/SOXX, QQQ, IGV, NVDA, PLTR/ORCL/AMD.
> **Optional convex add-on:** small YES on Polymarket “AI bubble burst by June 30, 2027.”
> **Optional perps if legally accessible:** short Hyperliquid NVDA, SMH/SOXL, DRAM, XYZ100; long small NATGAS/XLE as hedge.
That is the cleanest way to express “Citrini is right” without needing the hidden paid Citrindex holdings.