Global X U.S. Infrastructure Development ETF
Global X (Mirae Asset)
Tracks the Indxx U.S. Infrastructure Development Index — US industrials, materials, construction/engineering, electrical equipment, and rail that benefit from domestic infrastructure spend.
Top holdings
Electric-grid / infrastructure construction; top holding
Class I freight rail
Engineered metal/aerospace components
HVAC / climate — datacenter cooling
Electrical equipment / power management
Equipment rental for buildout
Performance
July 2026: 4.45% (adj close $58.92 Jun 30 -> $56.30 Jul 31; no distribution in window) — Mild down month, not a rout: -4.45% vs SOX -20.6%, SMH -17.6%. The AI-power sleeve is small and capped (ETN 3.06%, PWR 3.12%), so the memory/WFE/neocloud epicentre barely touched it — the non-AI ballast the critique calls dead weight is what limited the drawdown. The 1yr fall from 36.30% to 23.38% is window roll-off, not a July loss, confirming
Recent moves
Re-rated as a 'picks-and-shovels' AI-buildout play — grid buildout, datacenter construction, and electrical/cooling names (PWR, ETN, TT) drove ~+21% YTD vs ~+7.6% for SPY, with strong inflows lifting AUM past $14B.
Our take
The broadest, most diversified way to own the physical AI buildout — it captures grid, cooling, and electrical-equipment exposure (PWR/ETN/TT) without nuclear or commodity-price risk, but the rail and machinery sleeve dilutes the pure AI-power signal vs GRID.
Two mechanical features work against the story buyers are told.
First, the 3% cap plus semi-annual rebalance systematically trims the winners — every time an ETN or PWR re-rates on datacenter demand, the next reconstitution sells it back toward 3%. The methodology structurally suppresses the exact signal the AI narrative depends on. Second, the freight sleeve is large and thesis-irrelevant: CSX 3.92% + UNP 3.58% + NSC 3.24% is ~10.7% in three railroads (Jul 20, 2026), a bigger position than any single electrical-equipment name, plus Deere at 3.14% in ag machinery. Those are late-cycle freight and farm cyclicals with essentially no datacenter linkage. Net, at 73.3% industrials and 22.2% materials this is a high-beta U.S. industrials sector bet with an AI tilt bolted on by coincidence — the honest benchmark question is how much of the return is unavailable from a plain industrials fund at a fraction of the fee. Not leveraged, no daily reset: volatility decay does not apply. The structural drags are the 0.47% fee and the cap-driven winner-trimming, not compounding math.
Thesis
PAVE is sold today as the diversified way to own the physical AI buildout, but the index was designed in 2017 to track U.S. infrastructure development — >50% U.S.-revenue companies in construction/engineering, materials, heavy equipment and materials transport. The AI-power exposure (ETN, PWR, TT, ROK) looks like an accident of that overlap, not a design choice. Indxx did publish an amended methodology effective Nov 28, 2025; its contents could not be read at check time, so treat the 2017 design intent — not confirmed present-day stasis — as the basis for this claim.
Tracks the Indxx U.S. Infrastructure Development Index: top ~100 U.S. infrastructure-development names by market cap, >50% U.S. revenue required, modified cap-weighting with a 3% single-security cap and 0.3% floor applied at reconstitution, rebalanced semi-annually with buffer rules to damp turnover. 99 holdings (Jul 17, 2026); 73.3% industrials, 22.2% materials, 3.3% utilities, 1.1% info tech, 0.1% other (Jun 30, 2026).
Assessment
- ~$13.75B net assets (Jul 20, 2026) with ~100 large-cap holdings — deep liquidity, no capacity constraint on the underlying names
- 3% cap / 0.3% floor bound single-name concentration at each reconstitution, though live weights drift above it (CSX 3.92%, PWR 3.78%) between rebalances
- Pure U.S.-revenue screen avoids the FX and sovereign-policy noise embedded in global listed-infrastructure funds
- No commodity-price or nuclear-fuel exposure — the equipment and services layer, not the input layer
- The 3% cap plus semi-annual rebalance mechanically sells down the AI-power winners the fund is now marketed on — the methodology fights the thesis
- ~10.7% across CSX, UNP and NSC plus 3.14% Deere is dead weight against a datacenter thesis; freight and ag are separate cycles
- 73.3% industrials means most of the variance is sector beta, available far cheaper in a broad industrials fund
- 0.47% is 17bp above IFRA's 0.30% and several times a plain broad-industrials fund for exposure that is largely the same names
- The datacenter-linked demand is a second-derivative bet on hyperscaler capex guidance, which is revisable in a single earnings call
Record
The record is heavily regime-dependent. Issuer NAV (Jun 30, 2026): 36.30% 1yr, 24.16% 3yr ann., 18.86% 5yr ann., 16.57% since Mar 6, 2017 inception. Read those in order — since-inception sits well below both the 3yr and 5yr, which means the 2017–2020 stretch was unremarkable and essentially the entire excess arrived after IIJA passage in late 2021 and again on the 2024–26 AI-capex wave. Two fiscal impulses, not a durable structural edge. The window-sensitivity is stark: trailing 1yr was 23.65% as of Jul 21, 2026 (stockanalysis) against the issuer's 36.30% as of Jun 30, 2026 — a ~12.6pt swing from moving the window three weeks. That gap is the single most useful number here: a large share of the headline record is one concentrated hot stretch, and anyone quoting the 36% figure is quoting a window, not a run-rate.
Risks & fit
- Hyperscaler capex deceleration hits the electrical/cooling sleeve fastest and hardest — backlog-driven names de-rate on guidance, not on results
- Rate sensitivity in construction/engineering and equipment rental (URI 2.97%) — financing cost drives project starts
- Federal infrastructure funding is appropriation-dependent; the IIJA tailwind is finite and already partly spent
- Semi-annual rebalance means the portfolio lags a fast thematic rotation by up to six months
- High correlation to broad industrials means little diversification benefit alongside an existing industrials or cyclicals allocation
Watch two things together. If hyperscaler capex guidance is cut and electrical-equipment book-to-bill rolls below 1.0 while PAVE keeps tracking a broad industrials benchmark closely, the AI picks-and-shovels premise is falsified — what remains is a 0.47% industrials fund whose recent record was a capex cycle. The converse test is equally sharp: if datacenter demand accelerates and PAVE still fails to outrun broad industrials, the 3% cap and the ~11% rail sleeve are proven to have diluted the signal to nothing. Either outcome argues the thesis lives in the underlying cycle, not in this wrapper.
Suits an allocator who wants broad, diversified U.S. industrials-and-materials cyclical exposure with a modest structural tilt toward electrification and construction, and who is comfortable that the AI linkage is incidental rather than engineered. Poor fit for anyone seeking concentrated grid or datacenter-power exposure — the cap and the freight sleeve both dilute it. Also a poor fit as an add-on to an existing industrials sleeve, given the overlap.
0.47% expense ratio, above IFRA's 0.30% and below GRID's 0.56%, but several times a plain broad-industrials fund for a portfolio that is 73.3% industrials. Dividend yield ~0.77% — a capital-appreciation vehicle, not an income one.