AI is no longer a single trade — it is a capex super-cycle spanning chips, memory, datacenters, power, and software. Owning a single thematic ETF concentrates the portfolio in 30–80 names that overlap heavily with one another. We propose a four-ETF framework that captures the full AI value chain with ~165 unique underlying companies, no single-stock exposure above ~8%, and minimal duplication between sleeves.
The AI build-out has four economically distinct beneficiaries: (1) compute silicon (NVIDIA, AMD, Broadcom, ASML); (2) memory (SK Hynix, Samsung, Micron — the bottleneck of HBM and high-density DRAM); (3) datacenter physical layer (power, cooling, networking, contractors — Vertiv, Constellation Energy, Quanta, Dell); and (4) the application layer (Microsoft, Google, Meta, Palantir — companies monetising AI). Each segment has its own margin profile, capex cycle, and competitive dynamic. A single ETF cannot capture all four without becoming a megacap proxy.
The four ETFs below were selected because each owns one segment of the value chain with limited contamination from the others. SOXX and DRAM tilt the chip backbone; AIQ is the broad anchor (megacap software + Asian semis + China internet); TCAI is the picks-and-shovels diversifier with virtually no overlap with conventional semi indices.
The table below shows the weight of each shared name within each ETF (% of that ETF's NAV). Cells are colour-shaded by intensity. Micron is the only name held in all four sleeves; SK Hynix and Samsung are concentrated in DRAM with light AIQ exposure; the megacap chip names (NVDA, AVGO, AMD) are confined to SOXX + AIQ.
| Stock | SOXX | DRAM | AIQ | TCAI | Theme |
|---|---|---|---|---|---|
| MU — Micron | 7.6 | 25.4* | 3.7 | 5.0 | Memory |
| SK Hynix | — | 23.7 | 4.8 | — | Memory (HBM) |
| Samsung Elec. | — | 19.8 | 4.1 | — | Memory + foundry |
| AMD | 8.0 | — | 4.0 | — | GPU/CPU |
| AVGO — Broadcom | 8.0 | — | 3.7 | — | AI ASICs |
| NVDA — NVIDIA | 6.9 | — | 3.1 | — | GPU |
| INTC — Intel | 6.3 | — | 4.2 | — | CPU/foundry |
| TSM | 3.0 | — | 3.5 | — | Foundry |
| STX — Seagate | — | 5.6 | — | 5.6 | Storage HDD |
| WDC — Western Digital | — | 4.9 | — | 4.9 | Storage HDD |
| CRDO — Credo | 1.8 | — | — | 2.8 | AI connectivity |
*DRAM holds Micron via three lines (a direct equity stake plus two total-return swaps), aggregating to ~25% of the ETF.
| Sleeve | USD | % | Rationale |
|---|---|---|---|
| AIQ | 35,000 | 35 | Largest sleeve — broadest universe (87 names) and the only access route to AI application-layer megacaps and Asian internet. Functions as the diversification anchor that prevents the portfolio from becoming a pure semiconductor bet. |
| TCAI | 30,000 | 30 | Second-largest sleeve precisely because it is the most differentiated. Captures the physical AI build-out — power, networking, contractors — that semis-only ETFs ignore. Provides correlation diversification within the theme. |
| SOXX | 20,000 | 20 | The chip-backbone sleeve. Adds heavy exposure to capital equipment (ASML, Applied Materials, Lam, KLA) and analog semis — segments thinly represented in AI-themed indices but central to the capex cycle. |
| DRAM | 15,000 | 15 | Smallest sleeve due to 14-name concentration. Functions as a high-conviction amplifier on the memory shortage — the bottleneck most directly priced into AI server unit economics. Sized to add tilt without dominating. |
| Total | 100,000 | 100 | ~165 unique stocks · max single-name 8.1% (Micron) · ~72% US, ~10% Korea, ~3% Taiwan, ~2% China |
| Stock | Total % | Source sleeves |
|---|---|---|
| Micron (MU) | 8.1% | All four — largest concentration |
| SK Hynix | 5.3% | DRAM (heavy) + AIQ |
| Samsung Electronics | 4.4% | DRAM (heavy) + AIQ |
| Seagate + Western Digital | 5.6% combined | TCAI + DRAM (storage) |
| AMD / Broadcom / NVIDIA / Intel | 2.5–3.0% each | SOXX + AIQ |
| Risk | Description & mitigation |
|---|---|
| Theme correlation | All four sleeves correlate positively to the AI capex cycle — diversification reduces single-stock and sub-segment risk, not theme risk. Mitigation: size the AI bucket as a satellite (5–15% of total portfolio), not a core. |
| Memory cyclicality & concentration | DRAM/HBM pricing is the most volatile semi sub-segment (2–3-year boom-bust). Look-through Micron exposure is ~8% — the single largest name. Mitigation: review memory spot prices quarterly; if total Micron across the wider portfolio exceeds 10%, cut DRAM by 3–5pp. |
| Currency & geography | ~10% look-through is Korean (KRW) via Hynix/Samsung; Taiwan + China add ~5%. Fund-level exposure is USD via ADRs/swaps but underlying earnings are KRW/TWD/CNY. Mitigation: immaterial at $100k, flag in larger sizings. |
| Liquidity & fund newness | SOXX ($13B), AIQ ($8.8B), DRAM ($1B+) are liquid. TCAI ($93M) — wider spreads, use limit orders; split orders >$25k into 2–3 tranches. DRAM (Apr 2026) and TCAI (May 2024) have limited cycle history — monitor tracking error and mandate drift quarterly. |
| Tax / structure | All four are US-listed (1099 / W-8BEN). DRAM uses total-return swaps for part of Micron exposure — collateralised swap counterparty risk. Mitigation: review prospectuses for client domicile; UCITS alternatives may suit European clients. |
| ETF | Issuer | Expense ratio | AUM (US$) | Inception |
|---|---|---|---|---|
| AIQ | Global X / Mirae Asset | 0.68% | ~8.8B | May 2018 |
| TCAI | Tortoise Capital | 0.65% | ~93M | May 2024 |
| SOXX | BlackRock / iShares | 0.35% | ~13B | Jul 2001 |
| DRAM | Roundhill Investments | 0.65% | ~1B+ | Apr 2026 |