Nasdaq-100 (Invesco QQQ Trust)
Invesco (QQQ Trust; QQQM is the cheaper sibling share class). Tracks the Nasdaq-100 — the 100 largest non-financial companies on Nasdaq, cap-weighted, maintained by Nasdaq.
The most direct 'index = the AI trade' baseline. The Nasdaq-100 excludes financials and is heavily tech-tilted, so Information Technology is ~59% of the fund (vs ~33% for the S&P 500) and the same megacap-AI names sit at even higher weights. Buying QQQ is the closest thing to a passive concentrated AI/compute bet, which is exactly why it's the benchmark active AI managers most often quote themselves against.
Top holdings
Top weight; AI-accelerator core (mid-2026)
On-device AI / consumer compute
Azure AI
AWS compute
long — Custom AI silicon / networking; top-10
long — Gemini / TPUs; top-10 combined
long — AI capex + Llama; top-10
long — AI/robotics; top-10
Recent moves
Mechanical, price-driven reweighting plus the Nasdaq-100's annual reconstitution (December) and quarterly rebalances. The June 2026 QQQM rebalance introduced new capping rules to manage single-name concentration as Nvidia/Apple/Microsoft ballooned. No discretionary conviction — these are flows, not picks.
Our take
The toughest baseline on this page: QQQ is up ~30% over the trailing year and is ~59% tech, so an active 'AI fund' that doesn't clearly beat QQQ is just paying fees for index-plus-tracking-error. Its value as a comparator is precisely that it isolates the megacap-AI cohort. Limit: it is a one-factor bet — no financials, no energy/utilities, minimal power/nuclear/space picks-and-shovels — so it captures the compute layer but misses the AI-infrastructure breadth (grid, nuclear, datacenter REITs) that active managers on this page lean into, and its drawdowns are deeper than the S&P 500's.
Two things the marketing won't foreground.
First, structure: QQQ is a UIT, a 1999-vintage wrapper that cannot lend securities, cannot use derivatives to equitize cash, and must hold dividends in cash until distribution — a small, permanent drag its open-end sibling QQQM doesn't carry, on top of a 3bp higher fee (0.18% vs 0.15%). Investors pay for QQQ's options depth and liquidity; that is a trading feature, not a holding feature. Second, selection: the Nasdaq listing screen is arbitrary. It excludes any comparable business that lists on NYSE, and excludes financials by rule — so the index is near-silent on the rest of the AI capex chain, with utilities at 1.14% and energy at 0.45% of the fund (2026-07-20), while over-representing whoever happens to be Nasdaq-domiciled. The "AI purity" is coincidence dressed up as design. The capping rules are the honest tell: the methodology keeps having to intervene against its own weighting, which is an admission the mechanism produces concentration it doesn't want.
Thesis
QQQ is treated as "the passive AI trade," but the Nasdaq-100 has no AI thesis and no tech mandate. Its only rules are: listed on Nasdaq, not a financial, big. The ~61% tech weight is a side-effect of a listing-venue quirk from 1985, not a view. Cap-weighting then converts that accident into a momentum machine that mechanically adds exposure to whatever just re-rated. As of 2026-07-17 that means a semis/memory book — NVDA 8.09%, MU 4.28%, AMD 3.63% — with Micron and AMD each outweighing Meta at 3.17%. The index didn't decide that; price did.
Modified market-cap weighting of the 100 largest non-financial Nasdaq-listed companies, maintained by Nasdaq. Annual December reconstitution, quarterly rebalances, plus a capping regime that trims outsized names when concentration thresholds are breached, and discretionary special rebalances when needed. QQQ itself is a unit investment trust, not an open-end fund. 106 holdings; top 10 = 45.79% (stockanalysis.com, 2026-07-21).
- 2026-07Weights at 24 Jul 2026: NVDA 7.93%, MU 4.68%, AMD 3.72%, META 2.93% — the 3.17% belongs to GOOGL. Micron's weight rose through a month in which it fell 28.7%, so the memory position got heavier.
Assessment
- Rules-based and unfakeable — no manager discretion to blame, no style drift to detect, and the whole book is disclosed daily rather than 45 days late.
- At 0.18%, the fee is a rounding error against active AI mandates charging 100-200bp+, so the hurdle an active manager must clear is almost entirely alpha, not cost.
- Genuinely the right benchmark for the megacap-compute cohort — its narrowness is the feature when you want to isolate that beta.
- Deep liquidity and a mature options/futures ecosystem, making it usable as a hedging and expression instrument rather than only a holding.
- Selection rule is arbitrary. "Listed on Nasdaq, not a financial" is a venue accident from 1985, not an investment criterion — the AI tilt is coincidence, and it structurally excludes NYSE-listed peers.
- Cap weighting is momentum: it buys after the re-rating and sells after the de-rating. MU at 4.28% and AMD at 3.63% now outweigh Meta at 3.17% — the index bought the memory cycle after the move, not before it.
- The UIT wrapper is a legacy drag — no securities lending, cash-drag on dividends — that QQQM's open-end structure avoids at a lower fee. Same index, different plumbing, and the difference compounds.
- Concentration is not actually controlled, only periodically trimmed. Capping and special rebalances are reactive interventions against the methodology's own output, and they crystallize turnover at inconvenient prices.
- Single-narrative exposure. Financials are 0.17%, energy 0.45%, utilities 1.14% (2026-07-20) — if the AI capex cycle disappoints, essentially nothing inside the fund benefits from the rotation out of it.
Record
+26.15% trailing year, 10.75% annualized since inception (stockanalysis.com, 2026-07-21). Attribution is unambiguous: cohort beta, not skill. There is no manager, so the excess over broad indices is explained by (a) a 60.93% technology weight (2026-07-20) and (b) cap-weighting compounding the winners inside it — top 10 at 45.79%. Invesco's own framing — beaten the S&P 500 in 7 of the last 10 years (as of 2026-03-31) — is a window-selection artifact: every 10-year window starting after 2015 excludes the dot-com hole. The full-history record is the honest one, and the since-inception 10.75% embeds an 83% peak-to-trough drawdown in 2000-02 against -49.1% for the S&P 500 over the same span (1,527.46 to 776.76), a hole that took over 14 years to recover. 2022 repeated the shape at smaller scale, roughly 1.4x the S&P 500's decline. The upside and the drawdown asymmetry are the same mechanism.
- 2026-07At 31 Jul 2026 the trailing year is +21.70% and the since-inception rate 10.64%, against +26.15% and 10.75% ten days earlier — July's AI-hardware rout took 4.45pp off the trailing year.
Risks & fit
- Concentration: top 10 at 45.79%, and the semis block (NVDA/MU/AMD/AVGO) is a single correlated bet on one capex cycle, not four positions.
- Drawdown asymmetry is structural, not historical accident — the same cap-weighted tech tilt produced 83% in 2000-02 against the S&P 500's 49.1%, roughly 1.7x.
- Sector exclusion by rule: financials 0.17%, energy 0.45%, utilities 1.14% — a rotation away from tech has almost nothing inside the fund to cushion it.
- Methodology risk — Nasdaq can and does change capping rules and run special rebalances, forcing index funds into large, price-insensitive, publicly telegraphed trades.
- Rate sensitivity: long-duration cash flows dominate the book, so multiple compression hits harder here than in a broad index.
The core claim is that QQQ is one factor — cap-weighted megacap compute — and that its spread across 106 names is cosmetic. Falsified if returns decouple from the semis/hyperscaler complex: specifically, if a >30% drawdown in the semiconductor block produces less than roughly a 12-15% drawdown in QQQ, the other names are doing real diversifying work and the single-factor read is wrong. It also weakens if capping becomes permanent and binding enough to hold top-10 weight below ~35% through a full cycle, which would change the risk shape from what history shows.
Analytically this is the reference case for the megacap-compute factor, and the hardest benchmark on this page: an active AI mandate charging 100-200bp must beat a 0.18% instrument already owning the same top names at higher weights. Its narrowness makes it a clean isolate of that beta and a poor proxy for AI-infrastructure breadth. Anyone using it as a comparator should note the QQQ/QQQM structural gap — citing the wrong share class misstates the hurdle.
0.18% total expense ratio. QQQM — the open-end fund tracking the identical index — is 0.15% with $97.49B AUM (stockanalysis.com, 2026-07-21). The 3bp gap plus the UIT structural drag is the price of QQQ's liquidity and options depth.