Diamond AI Fund — Strategy Specification

Purpose of this document: This is the strategy thesis for the Diamond AI Fund, a portfolio of AI value-chain companies. It describes how the fund should be constructed and managed — the investment thesis, weighting methodology, allocation targets, rebalancing rules, and known risks.

It deliberately does not list the current constituents, their weights, or their individual theses. That information is in the live state of the fund and is supplied separately at review time. When this document and the live snapshot disagree on specifics, the snapshot is authoritative for what the fund currently holds; this document is authoritative for how it should be run.

---

1. Strategy Overview

1.1 Concept

The Diamond AI Fund is a worldwide portfolio designed to capture the full AI value chain benefiting from the multi-year hyperscaler capital-expenditure wave. Unlike existing thematic AI ETFs (BOTZ, AIQ, CHAT, WTAI, ROBO, IRBO, QTUM), which are dominated by Mag7 mega-caps in their holdings, this fund:

• Uses modified equal-weighting to prevent mega-cap concentration.
• Explicitly overweights mid-cap and small-cap "Diamond in the Rough" names with high fundamental AI exposure relative to their market cap.
• Covers all layers of the AI value chain — chips, semi-cap equipment & EDA, networking/optics, power/cooling/electrification, power generation (incl. nuclear), hyperscalers, neoclouds, AI software, security, and servers/storage/robotics/vertical AI.
• Is worldwide (anything that can be purchased on Robinhood).

The constituent count is not fixed — the fund grows and shrinks as names are added and removed. What is fixed is the methodology.

1.2 Investment Thesis

Supply, not demand, is the binding constraint at every layer of the AI infrastructure stack. Hyperscaler capex commitments are contractually visible in supplier backlogs well before the revenue is recognized. The fund is constructed to capture this multi-year visible revenue while diversifying away from:

• Single-name risk (over-dependence on Nvidia),
• Single-customer risk (suppliers whose revenue concentrates in one or two hyperscalers), and
• Single-architecture risk (GPU vs. custom ASIC vs. CPU).

The thesis holds as long as hyperscaler capex remains in an expansion phase and supply remains tight. The risks section below describes what would break it.

1.3 Construction Methodology

The fund uses modified equal-weighting across two sleeves:

• Core sleeve — broad value-chain representation. Each Core name carries the same base target weight.
• Diamond sleeve — higher-conviction, typically mid-cap and small-cap names with outsized fundamental AI exposure. Each Diamond name carries a higher target weight than a Core name.

The Core-vs-Diamond split (how much of the fund, in aggregate, sits in each sleeve) and the per-sleeve name weights are management decisions, reviewed periodically. They are not hard-coded here — the live snapshot reflects the current weights.

Core-vs-Diamond split (soft target):

• Diamond sleeve target: 45–55% of NAV. The Diamond sleeve is where the fund expresses its highest-conviction, smaller-cap overweights, and it should remain the larger or roughly co-equal half of the fund — but not so dominant that the Core value-chain backbone becomes a minority of the portfolio. Treat a drift outside the 45–55% band as a review signal: above 55%, the fund is leaning too heavily on concentrated small/mid-cap bets; below 45%, the fund is drifting toward a broad-market AI tracker and losing the "diamond in the rough" edge that justifies its active construction. This is a soft constraint — a breach is a trigger to review the sleeve balance, not an automatic rebalance.

Hard constraints (a breach is a rebalance trigger):

• Single-name maximum weight: 4.00%. No constituent may exceed this.
• Sub-category maximum weight: 25.00%. No sub-category may exceed this.

Soft constraints (reviewed, not auto-enforced — a breach is a review signal, not a hard stop):

• Nvidia-circularity exposure: ~12% aggregate. Names that carry a direct Nvidia equity stake, investment, or strategic-partnership tie form a correlated cluster — an Nvidia data-center deceleration would re-rate them together (see risk 3). Keep their combined target weight at roughly 12% or below; treat a breach as a trigger to review concentration.
• Converted-miner concentration within Neoclouds: ~5% aggregate. A growing share of the Neocloud sub-category is made up of former cryptocurrency miners that have pivoted to AI/HPC hosting (e.g. names whose original business was bitcoin mining and whose AI-hosting revenue is contracted but still ramping). These names share a common risk profile: their pivot economics depend on converting mining infrastructure and power access into hyperscaler hosting contracts on a similar timeline, so they are more correlated with each other than their individual theses suggest. Keep the combined target weight of converted-miner names at roughly 5% or below — and no more than ~3 such names — so that the Neocloud sleeve is not effectively a single concentrated bet on the BTC-mining-to-AI-hosting transition. Purpose-built neoclouds (those that never operated as miners) do not count against this sub-limit. Treat a breach as a trigger to review whether the converted-miner names are genuinely differentiated or are tripling up on the same trade.

Individual names may be assigned a target weight different from their sleeve default when conviction warrants — high-conviction overweights or deliberate de-emphasis of a weakening name.

1.4 Sub-Category Allocation Bands

Each constituent rolls up to a sub-category. The fund targets the following aggregate allocation per sub-category. These bands are calibrated to the fund's actual modified-equal-weight construction across a constituent count in the 80–90 range; they describe the realistic target share each sub-category should occupy given that the fund holds many names and each is individually small. If the constituent count changes materially, revisit these bands.

These bands are aspirational targets, not hard constraints (except where they would breach the 25% sub-category cap). The actual sum of constituent target weights in a sub-category may sit slightly outside its band; that is acceptable and is itself a signal worth reviewing. The platform's drift alerts fire against the midpoint of these ranges — keep the bands here and the platform's internal targets in sync, since a mismatch causes the alert system and this document to tell two different stories.

The Hyperscalers & cloud band is deliberately modest (5–7%). The hyperscalers are mega-caps; the fund's anti-concentration design means they sit at the Core base weight rather than being overweighted. The fund still gets full read-through to the hyperscaler capex cycle through the supplier base — that exposure is the thesis, not direct hyperscaler ownership.

Data center REITs (e.g. Equinix, Iron Mountain) are tracked as their own sub-category rather than folded into Power, cooling & electrification — their risk/return profile is real-estate-like, distinct from the equipment and electrification names.

1.5 Maximum Name Count per Sub-Category

Modified equal-weighting means every name added to a sub-category dilutes the others within it. Beyond a certain point, adding names produces diminishing diversification benefit while increasing monitoring overhead and the chance of holding near-duplicate exposures (e.g. two MEP/mission-critical contractors, or three electrical-distribution-equipment vendors occupying the same niche).

Soft guidance: Target no more than ~8 names per sub-category. This is not a hard cap, but a sub-category that grows beyond ~8 constituents should trigger a review with two questions:

1. Are any of these names near-duplicates? If two constituents occupy essentially the same niche and respond to the same catalysts, consolidate into the higher-conviction name and free the slot.
2. Is the marginal name actually adding differentiated exposure, or is it diluting the sub-category's per-name weight below the point where it can matter to fund performance?

When a sub-category exceeds the ~8-name guidance, the preferred response is to prune the weakest-conviction or most-redundant name rather than to keep expanding — consistent with the fund's principle that a healthy fund prunes (§2.3). Sub-categories with genuinely broad, non-overlapping opportunity sets (e.g. Power, cooling & electrification, which spans generation-adjacent EPC, switchgear, cooling, and site-development niches) may justify carrying more names, but the justification should be explicit and the names should be demonstrably non-redundant.

---

2. Rebalancing Rules

2.1 Scheduled Rebalancing

• Frequency: Monthly (or on-demand, as management decides).
• Action: Return each constituent to its target weight within a ±0.25% tolerance.

2.2 Event-Driven Rebalancing

Trigger a review (and optional rebalance) if any of these conditions are met:

2.3 Constituent Substitution Rules

The fund roster is dynamic. Names enter and leave as the AI value chain evolves.

• Periodic re-screening (at least annually): Are there new mid-cap names with stronger fundamental AI exposure that should replace existing names whose conviction has faded — for example, Diamond names that have re-rated into mega-cap territory and no longer fit the "diamond in the rough" profile?
• Removal criteria — flag a constituent for potential removal if:
  1. Its market cap exceeds ~$300B and it was held as a Diamond (it is no longer a "diamond"); re-sleeve to Core or remove (see procedure below).
  2. Its AI revenue exposure has fallen materially.
  3. Its fundamental thesis has broken — major design loss, accounting or governance issue, secular demand shift, or loss of a key customer.
• Additions must fit one of the sub-categories above (or justify a new sub-category), and should respect the ~8-name-per-sub-category guidance (§1.5) — if a sub-category is already at the soft limit, prefer substitution over expansion.

Diamond → Core re-sleeve procedure. When a Diamond name crosses the ~$300B market-cap ceiling the monitoring tool raises a Diamond-outgrowth alert. The response is mechanical, not discretionary:

1. Re-sleeve the name from Diamond to Core — lower its target weight to the Core base weight. If its AI-exposure thesis has also faded, remove it outright instead.
2. Normalize weights so the fund sums to 100%.
3. Optionally promote a screened replacement into the Diamond sleeve to keep the sleeve populated.

Treat the alert as the trigger. A Diamond that has simply succeeded into mega-cap territory is a graduation, not a debate.

Removal is a normal fund action, not a failure. A healthy fund prunes.

---

3. Performance Benchmarks

Track the fund's performance against these comparables:

Compare cumulative return over 1mo / 3mo / YTD / 1yr / since-inception.

---

4. Known Risks & Considerations

These are the structural risks the strategy runs. A quality review should weigh whether any of them is materializing.

1. Hyperscaler capex-digestion risk. If hyperscaler free cash flow compresses sharply on AI cash burn and the market re-rates them on it, the entire downstream supplier ecosystem corrects together. This is the single largest risk to the thesis.
2. Single-customer concentration. Several connectivity, optical, and neocloud names derive disproportionate revenue from one or two hyperscalers. A pullback at one customer hits them hard.
3. Nvidia circularity. Nvidia holds equity stakes and purchase commitments across parts of the neocloud and optical supply chain. An Nvidia data-center deceleration would coordinate-re-rate every name that depends on those flows.
4. Capacity unlock cycle. Today's supply tightness (sold-out HBM, sold-out high-capacity HDDs, sold-out liquid cooling) is the suppliers' pricing power. When new capacity comes online, pricing and multiples normalize.
5. Power & nuclear execution risk. Reactor restarts, SMR first-power dates, and grid uprates are multi-year construction projects. Equity prices in option value; execution timelines are real and can slip.
6. Diamonds outgrow the sleeve. A successful Diamond name re-rates into mega-cap territory and stops being a "diamond in the rough." Periodic re-screening (§2.3) exists to catch this.
7. Trade, tariff, and export-control risk. The fund carries meaningful non-US and supply-chain-concentrated exposure — foundry and lithography chokepoints are concentrated in Taiwan (TSM) and the Netherlands (ASML), several semi-cap, metrology, and optical names are domiciled outside the US, and much of the value chain depends on cross-border movement of equipment, components, and finished silicon. New tariffs, export controls (in either direction), entity-list actions, or retaliatory trade measures can: (a) directly impair a constituent's addressable market or input costs, (b) re-route or strand supply chains, and (c) trigger correlated re-rating across geographically concentrated names even when individual fundamentals are intact. Because the fund is explicitly worldwide, this risk is structural rather than incidental. A quality review should periodically assess aggregate geographic concentration (especially Taiwan/China/Netherlands chokepoint exposure) and whether any pending trade action would coordinate-re-rate a cluster of holdings. Trade-policy shifts are an explicit event-driven rebalance trigger (§2.2).

---

End of strategy specification. Current constituents, weights, and per-name theses are supplied separately at review time.