Skip to navigationSkip to main contentSkip to right columnADVERTISEMENTDavid BerenSat, June 27, 2026 at 3:23 PM GMT+2 5 min readQuick ReadSMH owns zero physical infrastructure; PAVE fills that gap with 119 positions in power, steel, cooling, and rail companies building AI data centers.PAVE has returned 132% over five years versus 73% for SPY, outpacing the broader market while capturing reshoring and infrastructure tailwinds.Rate sensitivity is the real risk: the 10-year Treasury at 4.51% already pressures PAVE's construction and capex-heavy holdings.Don't wait: the analyst who called NVIDIA in 2010 just revealed his top 10 AI stocks. See the full list FREE now.If you own VanEck Semiconductor ETF (NASDAQ:SMH) for AI exposure, you are betting on the same handful of chip designers everyone else holds. SMH is the default vehicle for the trade: a concentrated basket led by NVIDIA, Taiwan Semiconductor, and Broadcom that has produced the headline gains of the AI cycle. The case for owning it is straightforward and still mostly intact. The problem is what it leaves out. The hyperscalers' spending on those chips also needs land, steel, switchgear, transformers, cooling systems, and rail capacity, and SMH owns none of that. A different fund does, at a fee that is competitive with most thematic products on the market.24/7 Wall St.That fund is Global X U.S. Infrastructure Development ETF (NYSEARCA:PAVE), which charges 0.47% and holds engineering, electrical, materials, and rail companies that are physically building the AI buildout.Where SMH Falls Short for an AI Infrastructure ThesisThe design of SMH really concentrates risk in chip pricing and foundry cycles. When hyperscaler orders slow down, or an inventory correction sets in, the same concentration that powered the rally can reverse course just as fast. The fund also provides almost no exposure to the actual gating constraints of the current cycle, namely, power and physical capacity. Data center developers are out there signing multi-year contracts for transformers, switchgear, and grid interconnects that have nothing to do with chip ASPs at all. A semiconductor-only allocation like SMH captures the demand signal just fine, but it completely misses the broader supply chain that is feeding into that demand.Don't wait: the analyst who called NVIDIA in 2010 just revealed his top 10 AI stocks. See the full list FREE now.What PAVE Actually OwnsThe fund holds 119 equity positions with $12.43 billion in net assets. The top weights map directly onto AI infrastructure spend: Quanta Services at 3.37%, which builds electrical transmission lines and substations for hyperscaler campuses; Eaton at 3.16%; Trane Technologies at 3.33%, which supplies power management and liquid cooling; and Sempra at 3.15%, which delivers gas and electricity. Rails carry the steel and transformers: Union Pacific at 3.22% and Norfolk Southern at 3.06%. None of these names sit in the semiconductor fund, SMH.Terms and Privacy PolicyEU DSA contactPrivacy & Cookie SettingsMore Info