AI portfolio concentration risk is not a hypothetical for index investors: it is already embedded in their accounts. As of June 9, 2026, State Street Global Advisors’ daily holdings data show NVIDIA (NVDA) sitting at 7.97% of SPY, the single largest position in the fund. Microsoft (MSFT) adds another 4.72%. Micron Technology (MU) weighs in at 1.66%. That is before counting Alphabet, Meta, or Broadcom.
| Name | Ticker | SPY Weight (June 9, 2026) |
|---|---|---|
| NVIDIA | NVDA | 7.97% |
| Microsoft | MSFT | 4.72% |
| Micron Technology | MU | 1.66% |
| CHAT ETF expense ratio | CHAT | 0.75% |
| ASML lithography market share | ASML | ~90% |
SPY trades roughly 64 million shares per day and represents more than $790 billion in S&P 500 assets globally under State Street’s management. That scale matters: when a fund this large has nearly 8% in a single AI chip name, every passive dollar is an active AI bet whether the investor intended it or not.
Sizing Additional AI Exposure Around Existing AI Portfolio Concentration Risk
The standard practitioner framework treats dedicated AI positions as satellite holdings around a core index. Most cap the thematic slice at 10% to 15% of total equity. Given SPY’s existing AI weight, an investor who already runs a 60% SPY allocation likely has more direct AI exposure than they realize before buying a single thematic ETF.
For direct infrastructure exposure, NVIDIA’s Blackwell GPU line remains the dominant substrate for model training and inference. Broadcom (AVGO) offers an alternative angle through hyperscaler custom silicon and a more diversified revenue base that spans networking and enterprise software, with somewhat less valuation stretch than NVDA.
Microsoft (MSFT) is the lower-volatility path. Its Azure cloud platform and Copilot integration across the Office suite generate recurring revenue that does not hinge on any single customer’s capex cycle. Enterprises change productivity software slowly. That stickiness insulates MSFT from the kind of single-quarter shock that can crater a pure-play chip name. At a P/E of 26.80 against a consensus price target of $560.88, the valuation is not cheap, but it is not pricing in five years of flawless execution either.
Where AI ETF Diversification Is Real and Where It Is Not
Most AI ETFs replicate the top of the Nasdaq 100. Pull the holdings and you will find the same mega-cap names already in SPY, wrapped in a thematic label and a higher expense ratio. That is not diversification. That is repackaging.
The Roundhill Generative AI & Technology ETF (CHAT) pitches itself as the world’s first Generative AI ETF. It is actively managed at a 0.75% expense ratio. Its SEC annual shareholder report for the period ending April 30, 2025, confirms the fund invests globally in exchange-listed companies involved in generative AI and related technologies, including emerging markets exposure. That last piece is where a thematic fund can genuinely reduce AI portfolio concentration risk rather than replicate it.
The mid-cap and international AI layer is where real diversification lives. ASML (ASML) is the clearest example: the company holds an estimated 90% share of the semiconductor lithography market and is the world’s sole supplier of EUV lithography machines, the equipment that makes advanced chip fabrication possible at all. Per market analysis from PredictStreet, that monopoly position on EUV is structural, not cyclical. It does not appear in SPY at anything close to its strategic importance.
On the momentum end, Micron (MU) has become an AI proxy through surging high-bandwidth memory demand in GPU clusters. Vertiv (VRT), which supplies power and cooling infrastructure for data centers, reflects genuine demand growth, but also carries significant optimism in the valuation.
The practical build: one infrastructure name for direct chip exposure, one quality compounder for recurring revenue stability, and a thematic ETF only if its holdings actually reach into mid-cap and international names that an S&P 500 fund does not cover. AI portfolio concentration risk does not disappear by buying more AI. It compounds. The question is whether the next purchase adds something the index does not already own.
NVDA’s next earnings report is the near-term binary. A guidance miss or any credible signal that hyperscaler capex is decelerating would pressure the entire infrastructure layer simultaneously, and with NVDA at nearly 8% of SPY, the index itself would not be immune.
