The risk of an artificial intelligence bubble bursting remains relatively low, but investors could be overlooking significant concentration risks in some of the most popular semiconductor ETFs, according to Tal Schwartz, founder of Ai Funds.

The firm's recently unveiled AI Bubble Pop Index currently scores the risk at just 3 out of 20, based on Oct. 2 readings. However, Schwartz cautions that the index measures the broader market, not individual sectors.

"It doesn't score sectors, so a low reading isn't a clean bill of health for any of them," he told MarketDash.

Nvidia Exposure Separates the Riskiest AI ETFs

Among major AI-related ETFs, the VanEck Semiconductor ETF (SMH) appears particularly vulnerable. When asked which, among the major ETFs investors use to play the AI theme, has the greatest vulnerability to an AI spending slowdown, Schwartz opined that it was SMH, due to its concentration.

Nvidia Corp (NVDA) accounts for approximately 21% of its portfolio, while its top 10 holdings represent 69% of assets, according to Morningstar data.

By comparison, the Global X Artificial Intelligence & Technology ETF (AIQ) holds approximately 90 stocks, with its top 10 accounting for around 33% and Nvidia representing less than 3%.

"The more a fund depends on one group of customers' spending, the more a capex slowdown could hurt it," Schwartz said.

The iShares Semiconductor ETF (SOXX) also carries semiconductor exposure but is more evenly distributed, with Nvidia accounting for approximately 8%.

AI Spending Slowdown Could Hit Chipmakers Hard

A hypothetical 20% reduction in hyperscaler AI infrastructure spending over the next 12 months could put chip equipment makers, memory manufacturers, AI accelerators and networking suppliers under pressure.

"The suppliers whose revenue is that spending: chip equipment makers, memory, AI accelerators and networking, plus the optical and power names common in thematic funds. These companies carry high fixed costs, so a revenue dip tends to hit earnings harder, and their valuations assume years of growth," Schwartz explained.

"In the 2022 chip downturn, the Philadelphia Semiconductor Index fell 45% from its January high to its October low. The twist is that the hyperscalers themselves could hold up better, because spending less lifts their free cash flow," he added, illustrating the sector's vulnerability to spending cycles.

QQQ and SPY Also Carry Hidden AI Risks

At September-end, the top 10 stocks represented approximately 47% of the Invesco QQQ Trust (QQQ) and 39% of the SPDR S&P 500 ETF Trust (SPY).

"The risk runs deeper than the weights. Nvidia sells to Microsoft Corp (MSFT), Amazon.com, Inc (AMZN), Alphabet Inc (GOOGL) and Meta Platforms, Inc (META), which sit in the same funds, so one capex decision hits both sides," Schwartz explained.

He added, "A cap-weighted index is static by design. It owns the most of whatever has already risen the most, and it rides every decline."

Investors seeking to reduce concentration could consider the Invesco S&P 500 Equal Weight ETF (RSP), where technology accounts for approximately 16% of assets, compared with roughly 40% in SPY. However, equal weighting introduces greater exposure to banks, cyclicals and rate-sensitive stocks, Schwartz cautioned.

Why a Low Bubble Score Doesn't Eliminate AI Risks

Three of the index's 10 indicators currently carry yellow readings: AI-linked layoffs, China's cheaper AI models and the yield curve. Big Tech capital expenditure already consumes approximately 80% of operating cash flow, with spending above 100% triggering a warning.

Schwartz also sees cheaper Chinese AI models threatening U.S. companies' pricing power.

"In my view, it's the chance that AI monetization disappoints because of Chinese 'AI dumping,' landing on a highly concentrated system," he said.

Drawing parallels with Cisco Systems Inc (CSCO) during the dot-com boom, Schwartz noted that even dominant technology suppliers can experience severe corrections.

"Cisco sold the routers that built the internet, Nvidia sells the chips that power AI, and both became the world's most valuable company," Schwartz said.

"On its way up, Cisco fell 38% in early 1997 on slowing sales and 37% in 1998 during the Russia/LTCM crisis. Yet it gained about 590% in the two years before its March 2000 peak. Nvidia lost $589 billion in one day on the DeepSeek scare in January 2025, then regained nearly all of it within weeks. Scares like these are swings. Cisco fell 89% by October 2002 only after its customers stopped buying," he explained, noting that what ends a boom remained the same — customers reducing spending.

For ETF investors, his warning is particularly relevant: "Segments can swing hard inside a calm market. The Philadelphia Semiconductor Index closed 20% below its June record in July, then fell 6.7% on October 22 alone."