Roundhill Generative AI & Technology ETF
Roundhill Investments — Investment Committee (active)
Actively-managed, billed as 'the world's first generative-AI ETF' (launched May 2023). Global (incl. emerging markets) equities, ≥80% in AI/technology names spanning AI platforms, infrastructure/compute, enterprise software and consumer AI apps. Expense ratio 0.75%.
Top holdings
Top holding per latest fact sheet — Gemini/AI platform + compute (TPUs)
top weight (approx) — Core AI-compute infrastructure name
top weight (approx) — Azure AI + OpenAI partnership
Smaller international/EM AI position disclosed in the fact sheet — exact weight not confirmed; flagged for verification
approx — Chinese generative-AI startup exposure disclosed in the fact sheet — exact weight not confirmed; flagged for verification
Recent moves
Notably includes international/EM generative-AI names (e.g. MiniMax, Knowledge Atlas) alongside the US mega-cap AI leaders — a more globally-diversified AI basket than most US-only AI ETFs. ~$1B of net inflows over the trailing year.
Our take
The most explicitly 'generative-AI' thematic ETF and a genuine pure-play on the GenAI buildout — its differentiator is real international/China AI exposure that the ARK and MAGS funds lack. Solid as a thematic core for the GenAI trade, though concentration in the same mega-cap AI names limits how 'differentiated' it really is.
The teaching point: 'generative AI' branding oversells it — economically CHAT is an AI-compute + Asia-hardware basket, not a pure GenAI-application fund.
The top of the book is NVDA (~7.85%), then GOOGL/AVGO/AMD/SK Hynix/MU/ASML/Samsung — the infrastructure layer every AI ETF owns, so its return is beta to the same chip/memory cycle that lifted SMH/SOXX/AIQ, not demonstrable selection alpha. The US-megacap core (NVDA/GOOGL/AVGO/AMD) is closet-index overlap with those peers; the 0.75% active fee earns its keep only in the Asia/EM sleeve (SK Hynix, Samsung, SoftBank, SK Telecom, and Beijing LLM lab Zhipu AI / Knowledge Atlas), which is the genuine differentiator. Lower top-10 concentration (~42% vs SMH's NVDA ~17%) means it is less single-name levered — it lagged the most NVDA-heavy peers on the way up but should draw down less if NVIDIA de-rates.
Thesis
Active, global 'generative-AI' basket whose real edge over US-only AI ETFs is genuine Asia/EM exposure — SK Hynix, Samsung, SoftBank, and China LLM lab Zhipu AI (its #4 holding, listed as 'Knowledge Atlas', ~4.7%). But the book is dominated by AI-compute semiconductors, so its ~+69% 1-yr (Jul 20 '26) rode the chip/memory supercycle, not GenAI-application stock-picking — the 'generative' label oversells how differentiated it is.
Actively managed, 25-50 global stocks, >=80% in AI/tech across platforms, infrastructure/compute, enterprise + consumer software, explicitly including emerging markets. 47 holdings, top-10 ~42% of assets — diversified for a thematic fund. Launched May 2023 as 'the world's first generative-AI ETF'; 0.75% expense ratio. The only active GenAI fund with a real China/Korea sleeve.
Assessment
- Genuine international/EM AI exposure — SK Hynix, Samsung, SoftBank, and Beijing LLM lab Zhipu AI (Knowledge Atlas, #4 at ~4.7%) — that US-megacap AI ETFs structurally lack
- Diversified for a thematic fund: top-10 ~42%, NVDA only ~7.85% — far lower single-name risk than SMH's ~17% NVDA
- Active mandate can rotate into new GenAI names (Chinese labs, app-layer) an index would add late
- 0.75% ER is mid-pack for active thematic, below peer active fund ARKW's 0.88%
- 'Generative AI' label oversells it — the book is dominated by AI-compute semis (NVDA/AVGO/AMD/SK Hynix/MU/ASML), the same infrastructure trade as SMH/AIQ, not pure GenAI applications
- US core (NVDA/GOOGL/AVGO/AMD) is closet-index overlap with every AI ETF; the active fee is justified only by the Asia/EM sleeve
- 100% single-theme — one AI narrative drives the whole book, so there is no diversification when the trade turns
- Asia/EM sleeve adds FX, China-policy/delisting and Korea-macro risk most US buyers don't price
- Since-inception ~+45.8%/yr spans only ~2 years of a bull supercycle — no full-cycle drawdown observed
Record
~+69.1% 1-yr total return (Jul 20 '26), ~+45.8%/yr since May-2023 inception, YTD ~+39% (Jul 17; sources ranged widely, ~39%-66%, by reporting date). This is beta to the 2024-26 AI-compute/memory supercycle — the same wave that lifted SMH/SOXX/AIQ — not isolated stock-selection alpha. The lower top-10 concentration (~42%) means it trailed the most NVDA-levered peers into the top but carries less single-name downside. No full bear cycle has been tested. The 2.05% yield is incidental to foreign holdings, not an income thesis.
Risks & fit
- AI-capex/compute-cycle stall (hyperscaler capex cut, memory glut) de-rates the whole semis-heavy book at once
- Chip concentration means a semiconductor downcycle hits harder than the 'software' label implies
- China-policy / delisting risk on the China AI sleeve — #4 holding Zhipu AI (Knowledge Atlas) is a Beijing LLM lab exposed to US-listing / export-control risk
- Korea-macro + FX beta via SK Hynix / Samsung / SoftBank / SK Telecom
- Active-manager process/key-person risk with only a ~2-year track record
If the GenAI compute cycle stalls — hyperscaler capex guides down, a memory glut hits, or GenAI monetization disappoints — CHAT's semis-heavy book de-rates in unison and the Asia/FX sleeve amplifies the drawdown, exposing the 'diversification' as one macro trade. Conversely, if the Asia/EM sleeve durably outperforms the US-megacap AI core across a full cycle, the active mandate is validated as real differentiation rather than chip-cycle beta.
Describes the exposure, not advice: it fits someone wanting one-ticker global GenAI participation WITH deliberate Asia/EM AI weighting, who can tolerate a 100% single-theme, chip-cycle-volatile book plus FX and China-policy risk. It is largely redundant for someone already holding a semiconductor or US-megacap AI ETF, given the heavy mega-cap overlap in the top of the book.
0.75% expense ratio — mid-pack for an active thematic AI fund; below peer active fund ARKW (0.88%) but roughly double a passive broad-semiconductor ETF. The premium pays for the active Asia/EM sleeve; the US-megacap core it shares with cheaper funds.