## Assumptions I will use
- **Horizon:** 12–18 months, through roughly **Q1/Q2 2028**.
- **Capital unit:** I’ll express sizing as a **$100,000 model portfolio**; scale linearly.
- **Risk budget:** tolerate roughly **15–20% mark-to-market drawdown** before forced de-risking.
- **Core thesis:** AI demand keeps growing, but the binding constraint shifts from “chips only” to **chips + firm power + grid equipment + energy infrastructure**.
- **Style:** long-only public-market basket by default; optional hedged/perp/prediction-market versions at the end.
- **Caveat:** This is investment research and trade-structuring, not personalized financial advice.
---
# Phase A: UNDERSTAND
## 1. Thesis Parser
Your thesis, **“Long Energy + Long AI,”** decomposes into four investable sub-theses:
1. **AI compute demand remains durable.**
Beneficiaries: semiconductors, memory, networking, foundry, AI infrastructure.
2. **Electricity becomes the bottleneck for AI scaling.**
Gartner projects worldwide data-center electricity consumption to grow **26% in 2026**, with AI-optimized servers accounting for **31% of data-center power consumption** in 2026. (
gartner.com)
3. **Firm, dispatchable power and grid interconnection become scarce assets.**
EIA’s 2026 outlook says data centers are a major driver of U.S. electricity-demand growth, and its February STEO forecast U.S. electricity load growth of **1.9% in 2026** and **2.5% in 2027**. (
eia.gov)
4. **The biggest winners may be the “picks and shovels” between AI and electrons.**
These include nuclear/gas power owners, grid equipment, electrical infrastructure contractors, cooling/power management, uranium, and selected gas/LNG infrastructure.
**Parsed thesis:**
> Own the companies and instruments that benefit when AI compute demand stays strong **and** power scarcity, grid bottlenecks, nuclear/gas procurement, and utility capex reprice upward.
---
## 2. Scenario Formalizer
I’ll define five 12–18 month scenarios:
| Scenario | Probability | What happens | Thesis impact |
|---|---:|---|---|
| **AI + power bull** | 45% | AI capex stays high; power bottleneck worsens; utilities/generators/grid suppliers rerate | Best case |
| **AI efficiency surprise** | 15% | AI demand grows, but inference efficiency reduces power intensity | AI wins, energy lags |
| **AI capex air pocket** | 20% | Hyperscalers slow capex; AI valuations compress | Worst case |
| **Regulatory / permitting bottleneck** | 10% | Moratoria, grid delays, local pushback slow new data centers | Mixed: existing power assets win, AI hardware weaker |
| **Macro/rates shock** | 10% | Higher rates/recession hit long-duration growth and capex | Broad drawdown |
J.P. Morgan’s AI-capex work frames the same central bottleneck: power availability is now a primary constraint, and data centers are projected to rise from roughly **3–4% of U.S. electricity demand today** to **9–17% by 2030**. (
jpmorgan.com)
---
# Phase B: MAP THE BUTTERFLY EFFECT
## 3. Consequence Graph
**Root cause:** AI model scaling + inference adoption
→ **More GPUs / accelerators**
→ **Higher data-center rack density**
→ **More electricity demand, cooling demand, backup generation, grid interconnection demand**
→ **Scarcity of firm power, gas turbines, transformers, switchgear, transmission, nuclear PPAs**
→ **Hyperscalers sign long-term power deals**
→ **Existing nuclear/gas assets and grid suppliers reprice**
→ **Utility capex rises**
→ **Energy + AI infrastructure basket outperforms broad market if AI demand persists**
Concrete evidence: S&P Global forecasts roughly **$1.3 trillion** of aggregate U.S. energy-utility capex for **2026–2030**, driven by grid modernization and large-load customers, especially AI/data centers. (
spglobal.com)
Recent corporate actions support the thesis: Amazon, X-energy, and AEP announced a proposed **960 MW** SMR project in Indiana; Meta announced nuclear agreements with Vistra, TerraPower, and Oklo supporting up to **6.6 GW** by 2035; and Constellation’s Three Mile Island restart is tied to a 20-year Microsoft PPA. (
aboutamazon.com)
---
## 4. Instrument Mapper
| Consequence | Best instruments |
|---|---|
| AI compute demand | **NVDA, AVGO, AMD, TSM, ASML, MU, SMH** |
| AI networking / memory | **AVGO, MRVL, MU, SK Hynix exposure via Hyperliquid, SMH** |
| Grid/electrical bottlenecks | **GEV, ETN, PWR, VRT, GRID** |
| Existing nuclear / power scarcity | **CEG, VST, NRG, AEP, NEE** |
| Uranium / nuclear supply chain | **URA, NLR, CCJ, UEC, URNM** |
| Gas as marginal power fuel | **EQT, KMI, LNG, XLE, UNG/NATGAS perps** |
| Broad hedge | **SPY/QQQ shorts, puts, or lower gross exposure** |
---
# Phase C: ESTABLISH THE EDGE
## 5. Market Baseline / Pricing
As of the latest available U.S. equity pricing snapshot, Friday **October 23, 2026**, AI is already heavily priced: **NVDA** traded at **$210.96** with a market cap of roughly **$5.12T**, **AVGO** at **$321.68**, **AMD** at **$516.79** with a high P/E, and **SMH** at **$527.20**.
The energy/power side is also partially priced: **CEG** traded at **$253.45**, **VST** at **$140.16**, **GEV** at **$815.00**, **ETN** at **$366.11**, **PWR** at **$546.43**, **VRT** at **$194.32**, **URA** at **$34.11**, and **XLE** at **$69.33**.
**Edge conclusion:** the market already knows “AI needs power.” The edge must be more specific:
> Do not simply buy QQQ + XLE. Own the **scarce bottleneck assets**: existing nuclear/gas generation, grid equipment, power infrastructure contractors, and AI compute suppliers with pricing power.
---
## 6. Historical Analog Finder
The closest analogs:
| Analog | Lesson |
|---|---|
| **Internet buildout, 1995–2000** | Infrastructure suppliers can massively outperform, but capex cycles can end violently. |
| **China commodity supercycle, 2003–2008** | A real demand shock can reprice energy, metals, equipment, and infrastructure for years. |
| **Shale/LNG infrastructure cycle** | Midstream and contracted assets win when volume growth needs infrastructure. |
| **2022 energy shock** | Scarce dispatchable energy can outperform even while long-duration tech derates. |
| **Electrification / rail buildouts** | The bottleneck suppliers often capture returns before the final application layer matures. |
Key takeaway: this thesis should not be expressed only through the most obvious AI stocks. It should be expressed through **compute + power + wires + fuel**, with explicit protection against a capex bust.
---
## 7. Response / Beta Estimator
Estimated factor betas:
| Instrument bucket | AI capex beta | Power-price beta | Grid-capex beta | Rates beta | Main risk |
|---|---:|---:|---:|---:|---|
| AI semis | Very high | Low/negative | Medium | High negative | AI valuation compression |
| Power generators | Medium | Very high | Medium | Medium negative | Regulation / power-price caps |
| Grid equipment | Medium | Medium | Very high | Medium negative | Order-cycle slowdown |
| Uranium/nuclear | Low/medium | Medium | Medium | High volatility | Nuclear policy / commodity risk |
| Gas/LNG | Low | High | Low/medium | Medium | Commodity cyclicality |
| Utilities | Low/medium | Medium | Medium | High negative | Rate-base politics, allowed ROE |
**Best beta mix:**
- AI semis for upside.
- Nuclear/gas generators for power scarcity.
- Grid/electrical equipment for the “bottleneck capex” leg.
- Uranium/gas as convex but smaller satellites.
---
## 8. Trade Backtester
Because the basket contains newer 2026 assets, recent HIP-3/perp markets, and evolving power-market dynamics, I used a **regime/analog backtest** rather than pretending a clean historical series exists.
Modeled sleeve weights:
- AI compute: **25%**
- Power generation / nuclear: **30%**
- Grid/electrical: **30%**
- Uranium/nuclear fuel: **5%**
- Gas/LNG/energy: **5%**
- Cash/T-bills: **5%**
Scenario output:
| Scenario | Portfolio return estimate |
|---|---:|
| AI + power bull | **+28.8%** |
| AI efficiency surprise | **+5.5%** |
| AI capex air pocket | **-19.4%** |
| Regulatory bottleneck | **+6.6%** |
| Macro/rates shock | **-15.3%** |
Probability-weighted modeled return: **~+9.0%** over the 12–18 month horizon. The key downside is a true AI capex bust, not oil weakness.
Analog stress test:
| Analog regime | Modeled basket outcome |
|---|---:|
| Dot-com / capex bust analog | **-24.1%** |
| 2022 rates + energy shock analog | **-2.8%** |
| China/infrastructure supercycle analog | **+28.4%** |
| Soft-landing AI diffusion analog | **+19.2%** |
| Efficiency breakthrough analog | **+5.6%** |
Interpretation: the basket is much more robust than a pure AI basket, but it still loses materially if AI capex breaks.
---
# Phase D: BUILD THE TRADE
## 9. Expression Selector
Best expression:
> **Long AI compute + long power scarcity + long grid/electrical bottleneck.**
Avoid expressing the thesis only as:
- long **NVDA**, because that ignores power scarcity;
- long **XLE**, because crude oil is not the cleanest AI-power exposure;
- long **utilities broadly**, because regulated utilities may absorb capex but not capture scarcity upside;
- long **hyperscalers only**, because they are both AI beneficiaries and power-cost victims.
Preferred hierarchy:
1. **Core equity basket** — best for most investors.
2. **Call-spread version** — best if you want defined downside.
3. **Relative-value version** — long AI-power basket, short SPY/QQQ partial beta.
4. **Hyperliquid perp version** — best for tactical/leverage-aware crypto-native execution.
5. **Polymarket version** — best for event convexity, not full thesis replication.
---
## 10. Basket Optimizer
### Model portfolio: $100,000
Reference prices below are the latest available U.S. equity quotes from October 23, 2026.
| Ticker | Role | Weight | $ per $100k | Ref. price | Shares per $100k |
|---|---|---:|---:|---:|---:|
| **SMH** | AI semi ETF | 8% | $8,000 | $527.20 | 15.17 |
| **NVDA** | AI compute leader | 5% | $5,000 | $210.96 | 23.70 |
| **AVGO** | AI networking/custom silicon | 4% | $4,000 | $321.68 | 12.43 |
| **TSM** | Foundry | 4% | $4,000 | $418.51 | 9.56 |
| **AMD** | AI accelerator challenger | 4% | $4,000 | $516.79 | 7.74 |
| **CEG** | Nuclear / contracted power | 9% | $9,000 | $253.45 | 35.51 |
| **VST** | Power generation / nuclear | 8% | $8,000 | $140.16 | 57.08 |
| **NRG** | Merchant power / retail power | 5% | $5,000 | $92.31 | 54.17 |
| **NEE** | Renewables + utility scale power | 4% | $4,000 | $82.10 | 48.72 |
| **AEP** | Utility/data-center load exposure | 4% | $4,000 | $129.69 | 30.84 |
| **GEV** | Grid/generation equipment | 8% | $8,000 | $815.00 | 9.82 |
| **ETN** | Electrical equipment | 7% | $7,000 | $366.11 | 19.12 |
| **PWR** | Transmission/grid contractor | 6% | $6,000 | $546.43 | 10.98 |
| **VRT** | Data-center power/cooling | 6% | $6,000 | $194.32 | 30.88 |
| **GRID** | Smart-grid ETF | 3% | $3,000 | $162.68 | 18.44 |
| **URA** | Uranium ETF | 3% | $3,000 | $34.11 | 87.95 |
| **CCJ** | Uranium producer | 2% | $2,000 | $78.60 | 25.45 |
| **KMI** | Gas/midstream | 2% | $2,000 | $34.43 | 58.09 |
| **LNG** | LNG infrastructure | 2% | $2,000 | $298.67 | 6.70 |
| **EQT** | U.S. natural gas producer | 1% | $1,000 | $55.63 | 17.98 |
| **Cash/T-bills** | Dry powder | 5% | $5,000 | — | — |
### Sleeve summary
| Sleeve | Weight |
|---|---:|
| AI compute | **25%** |
| Power generation / nuclear | **30%** |
| Grid / electrical infrastructure | **30%** |
| Uranium / nuclear fuel | **5%** |
| Gas / LNG | **5%** |
| Cash | **5%** |
This is intentionally more “powering AI” than “generic AI.”
---
## 11. Risk & Sizing
### Position rules
- Max single name: **9%** initial weight.
- Max AI compute sleeve: **25%**.
- Max power/grid combined: **60%**.
- Max commodity/fuel satellite: **10%**.
- Cash reserve: **5% minimum**.
### Entry plan
Use three tranches:
1. **60% now**: establish core exposure.
2. **25% on pullback**: add if basket falls **7–10%** without thesis impairment.
3. **15% on confirmation**: add after positive catalyst, e.g. new nuclear PPA, hyperscaler capex acceleration, utility load forecast revision, or grid-equipment backlog upside.
### De-risk triggers
Cut 25–35% of exposure if:
- hyperscaler capex guidance falls for two consecutive quarters;
- data-center interconnection queues are materially canceled rather than delayed;
- AI accelerator pricing/margins compress faster than unit growth;
- regulators impose power-price caps that block generator upside;
- the basket falls **15%** while QQQ/SPY are flat or rising.
### Portfolio sizing rule
If your maximum tolerable loss is **$10,000**, and the modeled adverse AI-capex scenario is roughly **-19%**, then size the basket at about:
> $10,000 / 19% ≈ **$52,600**
So a $100k model basket is appropriate only if a ~$19k drawdown is tolerable.
---
# Phase E: ACT
## 12. Trade Proposal / Execution Layer
### Default execution
For each $100k intended allocation:
1. Deploy **$60k immediately** across the model weights.
2. Hold **$5k cash** inside the basket.
3. Put the remaining **$35k** into staged limit orders:
- 25% of basket on 7–10% pullback.
- 15% after positive catalyst.
### Preferred order type
Use **limit orders**, not market orders, especially in:
- CEG, VST, NRG, URA, GRID, CCJ;
- any options version;
- any Hyperliquid perp version with thin books.
### Options version
If using options instead of cash equities:
- Buy **Jan 2028 call spreads** on SMH, CEG, VST, GEV, ETN, PWR, URA.
- Use call spreads rather than naked calls because implied vol in AI/power names may already be elevated.
- Keep premium at risk to **5–8% of portfolio NAV**.
### Relative-value version
If you want cleaner thesis exposure:
- Long the model basket.
- Short **15–25% notional SPY or QQQ**.
- This reduces broad equity beta and leaves more exposure to AI-power-specific repricing.
---
# Final: Hyperliquid and Polymarket
## Hyperliquid: what to trade
Hyperliquid is a better venue than Polymarket for directly expressing this thesis because HIP-3 markets include stock, ETF, commodity, and index perpetuals. HIP-3 enables builder-deployed perpetual markets, including non-crypto perps such as traditional assets, commodities, and indices; docs also note HIP-3 initially uses isolated margin and third-party frontends rather than the main app. (
hyperliquid-co.gitbook.io)
Current Hyperliquid/HIP-3 markets relevant to this thesis include **NVDA, AMD, AVGO, TSM, ASML, MU, SMH, DRAM, GEV, VST, BE, XLE, CVX, NATGAS, URANIUM, URNM, BRENTOIL, CL, COPPER, SP500, and XYZ100**. (
hyperliquidguide.com)
Keel’s live Hyperliquid stock-perp dashboard recently showed active markets/prices such as **NVDA around $211.97**, **AMD around $517.80**, **AVGO around $322.86**, **TSM around $417.38**, **SMH around $528.34**, **URNM around $42.12**, **XLE around $69.14**, **GEV around $814.11**, and **VST around $140.43**. (
usekeel.io)
### Hyperliquid thesis basket
If trading perps, keep total effective leverage **≤1.0x** unless you are deliberately running a tactical book.
| Hyperliquid leg | Suggested weight | Direction |
|---|---:|---|
| NVDA | 8% | Long |
| SMH | 8% | Long |
| AVGO | 5% | Long |
| TSM | 4% | Long |
| AMD or MU/DRAM | 5% | Long |
| GEV | 8% | Long |
| VST | 7% | Long |
| BE | 4% | Long, high volatility |
| XLE | 5% | Long |
| URANIUM | 7% | Long |
| URNM | 5% | Long |
| NATGAS | 8% | Long |
| BRENTOIL or CL | 4% | Long, optional |
| SP500 or XYZ100 | -15% to -25% | Optional hedge |
**Best Hyperliquid expression:**
> Long NVDA/SMH/AVGO/TSM + long GEV/VST/XLE/URNM/URANIUM/NATGAS, optionally short SP500 or XYZ100 to reduce broad-market beta.
**Major risks:** funding costs, liquidation risk, oracle/market-structure risk, no dividends, and venue/regulatory risk.
---
## Polymarket: what to trade
Important: because you appear to be in the **United States**, Polymarket’s international site displays that trading is blocked in the U.S. and directs users to **Polymarket US**; the international platform also states it is not CFTC-regulated, while Polymarket US is described as CFTC-regulated. (
polymarket.com)
### 1. Long AI via Nvidia market-cap outcomes
Polymarket has an active **“Nvidia’s Market Cap end of 2026?”** event. Current visible outcomes showed:
- $5.00–$5.50T: **28%**
- $4.50–$5.00T: **20%**
- $5.50–$6.00T: **20%**
- $6.00–$6.50T: **19%**
- $6.50T+: **8%**
The event had about **$42.3k** volume and resolves around **December 31, 2026**. (
polymarket.com)
**Trade if bullish AI:**
- Buy YES on **$5.50–$6.00T**
- Buy YES on **$6.00–$6.50T**
- Smaller convex tail: YES on **$6.50T+**
I would not focus on the $5.00–$5.50T bin if your thesis is strongly bullish, because that bin is close to spot/current market-cap territory and is less convex.
### 2. Data-center moratorium markets
There are Polymarket markets around data-center moratoria. One active-looking market, **“Will any state enact a data center moratorium by December 31?”**, showed roughly **28% YES** and about **$59.4k** volume in one recent snapshot. (
polymarket.com)
How to use it:
- **Buy YES** as a hedge against AI infrastructure friction: bad for rapid data-center buildout, but supportive of scarcity value for existing power and grid assets.
- **Buy NO** if your version of the thesis is that AI buildout proceeds with limited permitting interruption.
This is more of a **hedge/catalyst trade** than a pure “long energy + long AI” trade.
### 3. Oil / energy event markets
Polymarket has oil markets such as **WTI hit levels in October/November 2026** and **crude oil all-time high by December 2026 / March 2027**. The commodities page showed WTI level markets and crude all-time-high markets with multi-million-dollar volume. (
polymarket.com)
Use these only for the **energy-price** leg, not the AI-power leg. Crude oil is not the same as electricity scarcity.
### Best Polymarket expression
Polymarket cannot fully replicate the basket. Best use:
1. **Bullish AI:** buy higher Nvidia market-cap bins.
2. **Power bottleneck hedge:** buy selected data-center moratorium YES markets if priced attractively.
3. **Energy shock satellite:** buy WTI upside / crude all-time-high event markets if you also believe in geopolitical oil upside.
---
# Bottom line
The best trade is not simply **Long AI + Long XLE**.
The cleaner expression is:
> **Long AI compute, long existing power scarcity, long grid/electrical bottlenecks, with small uranium and gas satellites.**
My default implementation is:
- **25% AI compute**
- **30% power generation / nuclear**
- **30% grid and electrical infrastructure**
- **5% uranium**
- **5% gas/LNG**
- **5% cash**
For Hyperliquid, use **NVDA/SMH/AVGO/TSM + GEV/VST/XLE/URNM/URANIUM/NATGAS**, optionally hedged with SP500/XYZ100 shorts.
For Polymarket, use **Nvidia high market-cap bins** plus **data-center moratorium markets** as event overlays, subject to U.S. access restrictions.