You think your portfolio is diversified, but in reality you're betting on AI? Wall Street is starting to look through AI risks to find cash flow winners.

date
20:51 16/11/2026
avatar
GMT Eight
Until recently, investors regarded artificial intelligence as one of the biggest bets in Wall Street history. Now, many portfolio managers are more cautious about this area, trying to assess just how large AI risk really is within their portfolios. The stakes are high. According to the MSCI Global Think Tank Index, at least $25.3 trillion is invested in the world's largest companies directly related to artificial intelligence. Beyond that, there are other investments that are difficult to quantifycompanies that may benefit or suffer as AI reshapes industries.
Title context: You think your portfolio is diversified, but in reality you're betting on AI? Wall Street is starting to look through AI risks to find cash flow winners. Text: The most important shift in market sentiment surrounding the AI stock market investment frenzy right now is arguably that investors have begun asking: how much of the entire portfolio's potential returns and earnings depends heavily on the same round of trillion-dollar-scale AI compute infrastructure spending expectations continuing to expand? Even more significant, several important survey reports reveal that AI risk allocation exposure has already far exceeded the technology stock classification. The Philadelphia Semiconductor Index, even after rebounding, remains about 31% below its historical closing high. Combined with target price cuts for Nvidia and shrinking forward compute procurement forecasts, the market has begun looking through equities, bonds, and private investments to determine how much they collectively depend on continued AI capital expenditure expansion. GPU, memory, optical interconnect, and power equipment companies depend on customer procurement; compute project loans depend on lease repayments; and some municipal bonds are tied to fiscal revenue generated by data centers. Once financing tightens and campus deliveries are delayed, order fulfillment, profit growth, and debt-servicing capacity could come under pressure simultaneously. It is understood that MSCI statistics show that the largest companies most directly related to AI involve at least US$25.3 trillion in investment; among a group of large pension funds studied by Equable, 52 AI-related listed companies accounted for 38.5% of their public equity holdings as of June 2026, up from 20.7% in early 2010, equivalent to 9%11% of total fund assets, and this does not yet include bonds, venture capital, and other private assets. What truly needs to be looked through is the actual source of investment returnswhether different securities depend on the same AI spending, customer credit, and project delivery assumptions. PGIM's two-layer screening of nearly US$1 trillion in fixed income investments is precisely a reflection of this change: it analyzes both how AI is changing corporate business models and credit quality, and how a slowdown in capital expenditure or a market crisis could transmit to indirect holdings. Municipal bonds may depend on new tax revenue from data centers, industrial companies may depend on campus equipment orders, and loans may depend on compute rental income for debt servicing. Nomura Asset Management strategist Andy Goldberg warned that investors who broadly hold AI-related assets could collectively amplify selling pressure even if they only slightly trim allocations; BlackRock's Jay Jacobs noted that index concentration has already made diversification harder to achieve than it appears. BCA chief economist Peter Berezin emphasized that higher productivity does not necessarily bring higher profits: efficiency gains may be passed on to customers through competition and price cuts, and easing supply bottlenecks may also erode equipment makers' excess profits. Powell Industries' fall from a year-to-date high of US$328 to US$146, a drawdown of about 55.5%, reflects the market's repricing of scarcity premiums. Correspondingly, falling compute costs could also improve the unit economics of application companies. Risk screening can likewise uncover major investment opportunities overlooked by the market. Syzygy Asset Management founder Rob Arnott believes that investors' excessive focus on crowded, highly valued AI assets may cause them to overlook companies that can use AI to create value but have not yet been fully priced. From an operating mechanism perspective, falling compute costs and the spread of AI tools are expected to reduce R&D, customer service, and operating costs for some companies, but whether these efficiency gains remain in the income statement still depends on companies' pricing power, competitive landscape, and organizational execution capability. This also echoes the latest judgment of BCA chief economist Peter Berezinthat higher productivity does not necessarily bring higher profits. Therefore, the investment value of looking through AI exposure lies both in identifying holdings that depend heavily on capital expenditure expansion and in finding companies that can continuously convert AI efficiency gains into profits and free cash flow while valuations have not yet fully reflected such improvement. The allocation value of the AI theme in the next investment phase may depend more on whether companies can preserve cash flow growth after construction growth slows, and whether AI adopters can convert efficiency gains into truly retained profits. Philadelphia Semiconductor Index rebounds but still has not escaped bear market trajectory, market begins recalculating AI risk ledger After the sharp selloff on November 5, global stock market semiconductor and broad technology shares counterattacked on the 6th, 9th, and 11th, but on the 13th the Philadelphia Semiconductor Index and Nasdaq 100 fell 3.43% and 1.22%, respectively. Although they still rose 8.05% and 3.53% for the week, as of the close on the 13th, the Philadelphia Semiconductor Index was still 30.67% below its historical closing high on June 22, and the Nasdaq 100 was about 10.5% below its October 6 peak. The Philadelphia Semiconductor Index can be said to have already fallen into a technical bear market; the drawdown on November 5 once reached 38.43%, and the recent strong rebound has not yet recovered that decline, while the Nasdaq 100 is in a correction range of more than 10%. Nvidia leadership changes further raised the risk compensation demanded by the market. The company disclosed that the board accepted Jensen Huang's resignation on November 11 and appointed Chief Financial Officer Colette Kress as interim CEO. Subsequently, Goldman Sachs cut its target price from US$300 to US$220, Citi from US$285 to US$225, and Morgan Stanley from US$300 to US$267, while respectively maintaining "Buy," "Buy," and "Overweight" ratings. Based on Nvidia's closing stock price of US$162.85 on the 13th, after Nvidia founder Jensen Huang was forced to resign as CEO, these new targets from Wall Street financial giants still imply upside of about 35.1%, 38.2%, and 64.0%. Their disagreement centers on how much earnings existing orders can support and how much valuation should be suppressed by leadership uncertainty: Goldman Sachs believes long-term orders limit the direct hit to earnings over the next 1218 months, but valuation multiples face more obvious pressure. The severe selling pressure on the AI compute theme and semiconductor sector triggered by Jensen Huang's resignation further spread to Asia on November 16: Taiwan Weighted Index closed down 6.03%, with a cumulative drawdown of about 17.63% from its historical closing high on October 6; South Korea's KOSPI fell 3.59%, and Nikkei 225 fell 2.42%. Asia's stock markets are home to a vast array of chip design, wafer manufacturing, and memory chip leaders, as well as the most advanced 3D advanced packaging systems and chip testing systems, which belong to different semiconductor sub-markets yet share procurement budgets and guaranteed financing terms from Nvidia and a host of large AI cloud-related customers. When the market simultaneously reassesses customer spending, management continuity, and cost of capital, geographic diversification can hardly fully offset the shock from a common source of demand. Based on that day's closing prices, Taiwan's benchmark stock index was about 17.6% below its October 6 closing high; SK Hynix and Samsung Electronics were about 62.5% and 47.4% below their respective intraday highs on June 25 and June 19. These declines across different time spans show that AI infrastructure-related hardware trades have already experienced significant valuation contraction brought by the "AI data center construction delays" narrative after the Democratic victory, and Monday's selloff further deepened the correction. Nvidia's leadership change, especially Jensen Huang's resignation as Nvidia CEO due to a series of negative events, has led global investors to reassess the continuity of its customer financing support and industrial chain cooperation arrangements. AI selloff frenzy leaves investors rushing to gauge their true risk exposure Until recently, investors still regarded artificial intelligence as one of the largest investment bets in Wall Street history. Now, many portfolio managers are treating the area with greater caution, trying to assess how much AI risk their portfolios actually contain. This matters greatly. According to the MSCI Global Intelligent Index, at least US$25.3 trillion is invested in the world's largest companies most directly related to AI. In addition, as AI reshapes industries, investors also have additional exposure that is difficult to quantify in companies that may benefit or suffer. The recent violent selloff in AI-related stocks, including the Wednesday decline triggered by Nvidia CEO Jensen Huang's resignation, has put how much AI exposure investors actually have back in focus. For Nomura Asset Management chief investment strategist Andy Goldberg, the scale of the problem hit home last month when he saw wedding dress retailer David's Bridal plan to spin off its AI wedding planning business next year and push for a listing. "Since when did wedding dress stores also become AI investment targets?" Goldberg said. He added that the problem is that AI exposure exists even in the most unexpected corners of the economy. David's Bridal CEO Kelly Cook said the company "has been a pioneer in applying AI to retail for many years." Goldberg said that from freight companies using AI tools to solve logistics problems to municipal bonds whose credit ratings may depend on revenue brought by newly built data centers, AI affects both risk and return. "No one fully knows how much AI exposure they have; at this stage, it has effectively become one massive common trade," Goldberg said. He added that if investors decide to cut AI-related asset allocations by even 1% or 2%, the market reaction could also "look like a rout." A detective-style screening task For investors trying to clarify these exposures, the process is increasingly resembling detective-style screening work. At PGIM, the investment management arm of Prudential, analysts and portfolio managers overseeing nearly US$1 trillion in fixed income investments began this year conducting two AI-focused screenings on every bond, loan, and structured security. One assesses AI's impact on a company's business model and credit condition; the other screens the most unexpected companies for exposure to risks ranging from a slowdown in AI-related capital expenditure to a market crisis. "We know how to assess direct exposure, such as having data centers or chips in the portfolio," said Brian Barnhurst, PGIM's global head of credit research. "But what about indirect effects? Understanding that is crucial." Barnhurst said the approach helps analysts clarify one by one: "What does AI mean for this healthcare company, this industrial company, or this telecom company?" Is the true risk exposure too large? Undoubtedly, the capital global institutions or retail investors have put into the AI theme, and the actual market capitalization scale of the AI theme, are enormous. The nonpartisan pension think tank Equable Institute estimates that as of June 2026, a basket of 52 AI-related listed companies accounted for 38.5% of a group of large pension funds' public equity holdings, up from 20.7% in early 2010. Equable estimates that these funds have 9% to 11% of total assets invested in AI-focused companies. This figure does not include debt holdings, venture capital, and other private equity investments, so it almost certainly understates their broader exposure. Jay Jacobs, head of BlackRock's U.S. equity ETF business, said that at the current stage of the AI cycle, "understanding a portfolio's true exposure is becoming more important." So many core AI holdings now dominate large indices, creating a degree of concentration that makes "achieving diversification harder than investors may realize." A steady stream of news around major AI companies such as Nvidia, Anthropic's upcoming initial public offering, and concerns about CoreWeave have further increased uncertainty. Some fund managers are responding by raising their stock-selection standards. Peter Berezin, chief economist at market, economic, and geopolitical research firm BCA Research, said the market's merciless punishment of companies in the AI trade "can be seen as an early warning signal reminding all of us to be more discerning when examining the AI rollout process." "AI is very likely to greatly increase companies' productivity in the end, but not necessarily make them more profitable." The stock price swings of Powell Industries are an example. The company's power management systems are a key component of data centers. Its stock price was only US$8 four years ago, rose to US$328 at one point this year, and has since fallen to US$146. Powell Industries did not respond to a request for comment. Berezin said bottlenecks in data center construction will eventually ease, eroding the excess returns earned by Powell and other AI beneficiaries. But when that will happen is also unknown. Adding fuel to the fire Wall Street institutional investors and analysts almost unanimously agree that these factors together create conditions for continued market volatility. Janus Henderson technology analyst Sean Baki said that since AI valuations peaked in June, "we have seen a long list of concerns accumulate." He said the U.S. midterm election results "almost seemed like adding fuel to the fire," as Democratic gains raised concerns about AI's future growth. He said this has made investors "increasingly want to distinguish the good from the bad." Liz Ann Sonders, chief investment strategist at the Schwab Center for Financial Research, said that even with careful research, "the quantitative relationships behind all this AI exposure" make such a highly concentrated market especially difficult to navigate. But Syzygy Asset Management founder Rob Arnott said a calm, pragmatic assessment of AI concentration in a portfolio can reveal both risks and opportunities. "There is a kind of tunnel vision here," he said. When investors "take enormous risk in one of the frothier areas of the market, they overlook the opportunity to invest early in other companiescompanies where AI will create value, but that value has so far not been recognized by the market." Market reprices AI risk! SemiAnalysis sharply cuts AI compute forecasts: 2028 capex forecast cut by nearly 30%, silicon shipment area cut by 17% A research report recently published by well-known market research firm SemiAnalysis shows that global AI accelerator capital expenditure forecasts for 20272030 were cut by about 19%, 28%, 41%, and 40%, respectively, to new forecasts of US$627 billion, US$758 billion, US$796 billion, and US$923 billion. These are cuts relative to old forecasts, not year-on-year declines; the new path still implies investment rising from US$392 billion in 2026 to US$923 billion in 2030, a compound annual growth rate of about 24%. The SemiAnalysis analyst team said the main reasons for the adjustment are slowing revenue growth at frontier model companies, tighter financing, and scrutiny of the ability to cover compute costs, with an expected 5.8 gigawatts of U.S. data center capacity delayed delivery further adding pressure. The core change is that after Democrats took control of the U.S. Congress, AI data center construction has slowed, and revenue realization and financing capacity have begun to constrain procurement, rather than a sudden disappearance of computing demand related to AI workloads. On the hardware side, there are also dual adjustments in volume and price: the cumulative forecast for accelerator logic silicon shipment area in 20262030 was cut by 17%, with Nvidia accounting for 66% of the reduced area in particular; 2028 HBM capacity demand was cut from 9.0EB to 6.4EB, a decline of nearly 30%; 2030 accelerator power demand was cut from 122.8 gigawatts to 80.7 gigawatts. However, logic silicon area, memory capacity, and power measure different demands and cannot be converted interchangeably into chip counts. Falling HBM costs and lower prices and markups for new products will also cause nominal procurement amounts to fall faster than physical demand. For the profit outlook of leaders in the AI compute supply chain, it is necessary to observe delivery volumes, product mix, and pricing power simultaneously. The degree of pressure therefore varies significantly among companies. SemiAnalysis believes Nvidia faces financing risk from new cloud customers, competition from alternative computing platforms, and narrowing room for hardware markups; Broadcom's five-year ASIC revenue forecast was cut by 11%, mainly reflecting the pass-through of lower HBM costs into prices; AMD's near-term deployment is relatively solid, but its long-term forecast was still cut, and some custom chip projects at MediaTek were also adjusted; Amazon Trainium's 20272028 shipment forecast remains unchanged. In addition, it is worth noting that this round of model cuts by SemiAnalysis does not include the impact of Jensen Huang's departure. The research firm also said that GPU/TPU/ASIC, CPU, memory, optical interconnect, power equipment, and even project loans may all depend on the same group of customers financing, starting production, and paying on schedule; holding more segments of the industrial chain does not automatically reduce this common risk. Combining Goldman Sachs' and Morgan Stanley's investment strategy outlooks for global and Asian equity markets in 2027, AI-related investment opportunities can be screened along three main lines: infrastructure suppliers able to convert orders into revenue on schedule and maintain profit margins; cloud platforms with solid balance sheets, financing capacity, and commercialization advantages; and companies able to use AI applications to raise productivity and improve profits and cash flow. The investment standard running through these three main lines is whether growth can be delivered and whether current valuations reasonably reflect future returns.