Had You Bought This Vanguard ETF at the Start of January, You'd be Obliterating the S&P 500 in 2026

Vanguard Information Technology ETF (VGT) returned 32.7% in 2026 through Sept. 28, materially outperforming the S&P 500's 12.2% gain, driven by AI-linked mega-cap holdings. Nvidia, Apple, Microsoft, Broadcom, Micron and AMD represent 56.6% of the ETF, while Nvidia and AMD executives expect their data-center businesses to double next year. The outlook remains favorable for AI infrastructure spending, but proposed state-level data-center restrictions and a UBS survey showing 60% of businesses shifting some AI workloads to cheaper models present concentration and semiconductor-demand risks.

Analysis

The relevant exposure is not broad "technology" but a highly correlated AI-capex and memory-cycle basket. VGT’s concentration creates a reflexive flow risk: continued ETF inflows reinforce leadership, but any disappointment in hyperscaler capex guidance or AI-chip lead times can force simultaneous de-risking in NVDA, AVGO, AMD and MU. A 5-10% drawdown in the largest semiconductor constituents would likely overwhelm software/security gains elsewhere in the fund, making VGT a less efficient vehicle for expressing a selective AI view over the next 1-3 months.

Cheaper-model adoption is not unambiguously bearish for infrastructure. Lower inference cost expands the addressable set of enterprise use cases and can increase aggregate token volume; the near-term loser is premium training compute intensity, while networking, storage and enterprise security can benefit as deployments broaden. This favors AVGO’s custom silicon/networking exposure and PANW/CRWD’s AI-enabled security monetization over a pure long-duration GPU-beta position, though the benefit likely emerges over 6-18 months rather than in the next earnings print.

The key near-term catalyst is not state-level data-center rhetoric but disclosed hyperscaler capex, power procurement and deployment schedules. Construction restrictions can shift capacity geographically and raise project costs rather than erase demand; the sharper downside case is a measurable fall in accelerator utilization, cloud pricing, or memory contract prices. Consensus appears to treat all AI beneficiaries as one trade, leaving relative-value opportunity between capital-intensive component suppliers and recurring-revenue software/security names.

Key Decisions for Investors

• Do not add broad VGT exposure after its relative-strength move; use it only as a hedgeable beta sleeve. For a 1-3 month tactical long, wait for a 7-10% sector pullback or post-earnings confirmation that hyperscaler capex guidance remains intact.
• Initiate a 6-12 month pair: long PANW and CRWD, short an equal-dollar basket of MU and AMD. The thesis is that enterprise deployment broadening supports recurring security budgets while memory and second-source accelerator expectations retain greater utilization and pricing sensitivity; exit if MU contract pricing accelerates for two consecutive months or PANW/CRWD net retention and billings decelerate materially.
• Prefer AVGO to NVDA for incremental AI infrastructure exposure over the next two quarters, sized modestly. AVGO offers custom-ASIC and networking participation if workloads migrate toward cheaper, optimized models; thesis fails if hyperscaler custom-chip programs are delayed or networking revenue guidance weakens.
• Set a risk trigger around the next MSFT cloud-growth and capex disclosure: a capex reduction or cloud-growth deceleration without offsetting margin expansion would warrant cutting semiconductor longs first, as it would challenge the assumed conversion of AI demand into hardware orders.
• For portfolios already long AI semis, buy 3-6 month VGT put spreads rather than selling core winners outright. This targets concentration-driven downside while preserving upside from continued infrastructure orders; finance only partially, since upside gaps remain possible around earnings.