Global X Artificial Intelligence & Technology ETF
Global X (Mirae Asset)
Tracks the Indxx Artificial Intelligence & Big Data Index — developed-market companies that develop or use AI/big-data tech. In practice it is heavily memory/semiconductor-weighted with large Korean and Taiwanese chip exposure, not a US-megacap basket (~85-90 holdings).
Top holdings
Korean memory/HBM, the largest position
US DRAM/HBM
Memory plus foundry
AI data-center GPUs and CPUs
CPUs and foundry
AI data-center networking; NVIDIA only ~2.6% (#10)
Performance
July 2026: 10.24% (close $65.61 Jun 30 → $58.89 Jul 31) — July was a real drawdown, not a basis artifact: -10.24%, confirmed two ways (lazyportfolioetf monthly figure and Jun 30 → Jul 31 close-to-close). The issuer's standardized NAV set is still Jun 30 quarter-end (1yr +49.74%, 3yr +32.96% ann., 5yr +16.69% ann.) and now sits a full drawdown behind live trailing figures. Price $58.89, ~12.5% below the 52wk high of $70.26.
Recent moves
Strong inflows over the past year roughly doubled AUM to ~$10B.
Our take
Despite the 'AI' label, AIQ is really a global memory-and-chip basket — SK hynix, Micron, and Samsung lead while NVIDIA sits at ~2.6%, so its 2026 return tracked the HBM/memory cycle more than the GPU trade. The non-US tilt is a genuine diversifier but also a currency- and Asia-policy exposure most US investors don't expect.
The interesting failure is definitional. An index that admits AI *users* has no natural stopping point — nearly every large-cap now "uses AI" — so the screen degrades into a generic tech basket, which is exactly what the 79.2% IT / 8.0% Comm-Services split shows.
Apple at 3.54% and Cisco at 4.00% are not there because they lead AI; they are there because the definition is porous. The consequence is a portfolio you cannot attribute: when AIQ moves, you cannot say whether the AI thesis worked, because the largest sleeve (memory) and the ballast sleeve (mature US tech) run on different cycles. Second, the diversification is real but asymmetric — a ~43% top-10 caps single-name blowup risk, yet the 16.4% memory cluster is one economic bet, not three, since SK hynix, Micron and Samsung price off the same DRAM/HBM contract cycle. A holder gets breadth against idiosyncratic risk and concentration against cycle risk, which is the reverse of what most thematic buyers assume they are getting.
Thesis
AIQ sells itself as an AI fund but its index selects companies that develop OR USE AI and big data — and the "use" leg is what you actually own. As of Jul 20, 2026 NVIDIA is 2.96%, the 10th-largest position, below nine holdings including AMD (5.34%), Intel (4.03%) and Cisco (4.00%). The real engine is memory: SK hynix 6.25% + Micron 5.85% + Samsung 4.28% = 16.4% in three DRAM/HBM names. So AIQ is a global memory-and-legacy-tech basket wearing an AI label.
Passive, tracks the Indxx Artificial Intelligence & Big Data Index. 84 holdings per the issuer (91 per stockanalysis.com), top 10 = 42.68% per issuer (43.66% on stockanalysis's weights) — a deliberately broad, low-concentration book. 79.2% Information Technology, 8.0% Communication Services (Jun 30, 2026), primary country exposure US / South Korea / Taiwan. Inception May 11, 2018; net assets $9.40B (Jul 20, 2026); 0.68% net expense ratio.
- 2026-07The index reconstituted 31 Jul 2026. Top 10 is now MSFT 3.57%, AMZN 3.40%, GOOGL 3.28%, ORCL 3.17%; SK hynix, Micron, AMD, Intel and NVDA all dropped out. Memory is at most ~8.8% of assets, not 16.4%.
Assessment
- Genuinely low single-name risk: 84 holdings per issuer (91 per stockanalysis), top 10 ~43%, no position above ~6–7% depending on source — nothing here can halve the fund alone.
- Real non-US access. SK hynix, Samsung and TSMC ADR (~13.8% combined) are exposures most US-listed AI ETFs underweight or skip entirely.
- An 8-year live record (inception May 11, 2018) spanning a full drawdown — rare for AI-themed funds, most of which launched after 2023.
- Scale and liquidity: $9.40B net assets (Jul 20, 2026) makes it one of the few AI-labeled funds with institutional-grade tradability.
- The label misdescribes the book. Buying "AI" and receiving 2.96% NVIDIA — the 10th position, below AMD, Intel and Cisco — is a mismatch most buyers discover late.
- The three memory names (16.4%) are one correlated bet on the DRAM/HBM contract cycle, not three diversifying positions. Breadth elsewhere does not offset it.
- 0.68% is a thematic fee on a book whose largest sleeve duplicates cheaper, purer semiconductor index funds, and whose ballast sleeve duplicates cheap broad-tech funds.
- An "AI users" screen has no principled boundary — as AI adoption becomes universal the index's differentiation from a plain tech index mechanically decays.
- Korea/Taiwan exposure is unhedged. FX and Asia policy risk are embedded in the return but absent from the fund's name and marketing framing.
- 2026-07Post-reconstitution the book is flatter still: the top 10 is ~31% and the largest single position is MSFT at 3.57%. Memory falls to ~8.8%, so it is no longer the dominant economic sleeve.
Record
The record is strong and highly basis-sensitive. Issuer standardized NAV returns as of Jun 30, 2026: 1yr +49.74%, 3yr +32.96% ann., 5yr +16.69% ann., since-inception +20.22% ann. stockanalysis.com showed a trailing 1-year of +33.44% on Jul 21, 2026 — the gap is a measurement-basis artifact (quarter-end standardized NAV vs a rolling trailing window), not evidence of a collapse; memory did not unwind, Micron was +314% YTD as of Jul 10, 2026. Two attributions matter. First, the 5-year (+16.69% ann.) is the honest number: it spans the 2022 drawdown and is roughly half the 3-year, meaning the 3-year is measured from a trough. Second, the 1-year owes to memory, not AI compute — and even so it trailed the pure semis badly (SOXX +112.8% in H1 2026 alone). AIQ captured the cycle at a heavy dilution ratio. That is the cost of ~84 holdings, and it is the trade being made whether or not the buyer knows it.
- 2026-07Memory did unwind. Micron fell from $1,213.56 (25 Jun) to $739.00 (29 Jul), roughly 30%, on CXMT's $8.5B DRAM IPO and HBM export controls; AIQ returned -10.24% for July.
- 2026-07The basis gap is now largely real: standardized 1yr +49.74% (30 Jun) against live trailing +31.5–32.2% (31 Jul). The ~18pp widening tracks July's actual -10.24%, not measurement alone.
Risks & fit
- Memory is the most cyclical semi product. A DRAM/NAND price roll would hit 16.4% of the book at once, on the same catalyst.
- Headline returns vary sharply by basis and window (+49.74% Jun 30 standardized vs +33.44% Jul 21 trailing), so any figure quoted from marketing is likely stale or non-comparable.
- Taiwan-strait and Korea-policy tail risk sits in ~14% of assets with no hedge and no disclosure in the fund's name.
- Index-definition drift: an "uses AI" screen may admit ever more generic large-caps, quietly converting the fund into tech beta at 0.68%.
- 3-year and since-inception figures both start from favorable points; neither reflects a full memory boom-bust, which historically runs 3-4 years.
The differentiated-AI-exposure claim fails if, over a full cycle, AIQ's return sits within ~1%/yr of a plain low-cost tech-sector index fund — then the 0.68% fee and the thematic screen bought a label, nothing more. It fails the other way if the memory sleeve decouples: a DRAM downturn mauls SK hynix/Micron/Samsung while the remaining ~84% holds up, proving the breadth is load-bearing rather than cosmetic. The next memory price roll resolves it.
Suits someone who explicitly wants a broad, non-US-inclusive technology and memory-semiconductor basket and knows they are NOT buying concentrated AI-compute exposure. Poorly matched to anyone seeking leadership-name AI exposure (NVIDIA at 2.96%, the 10th position, will not deliver it), to anyone needing US-only or currency-hedged exposure, or to anyone treating it as a cheap core — 0.68% and 79.2% IT make it a sector sleeve.
0.68% net expense ratio (issuer, Jul 20, 2026) — ~$68/yr per $10,000. Charged on a book whose largest sleeve overlaps broad semiconductor index funds and whose remainder overlaps broad tech index funds, both available materially cheaper.