Index providers are now breaking the word "AI" into a stack of separately tradable parts. MSCI has launched a series of artificial intelligence (AI) value chain indexes to help investors hedge their exposure to the sector more precisely.

This batch of indexes comprises 14 products, launched at the end of August, covering global companies across the AI supply chain — from physical infrastructure and digital infrastructure all the way to the application end. According to an MSCI document on these products, they enable investors to assess exposure at the "layer" or "component" level, align positions, and act quickly when bottlenecks emerge. "AI is often just that headline, but beneath the headline there are many components," MSCI index business head Jana Haines said in an interview. "What we keep hearing from clients is that their challenge isn't identifying AI exposure, but breaking it apart."

The launch timing: the AI trade is no longer one trade, but dozens

Media highlighted the timing of the product launch: the AI trade has become one of the biggest sources of stock market volatility, with investors swinging back and forth between optimism about the technology's economic prospects and concerns about the sector's heavy spending. Volatility is just a general term — it becomes more intuitive when broken down. According to MSCI index data, in the first half of 2026, the semiconductor and equipment industry within MSCI's index system rose 55%, five times the 11% gain of the MSCI ACWI all-country index, while software and services fell 18% over the same period, making it the worst-performing industry — investors fear that AI agents will break the software industry's seat-based charging business model. The same "AI" delivered a 73 percentage point gap between the two ends.

The rift in the internal structure became explicit in mid-September. According to a report sent by Goldman Sachs strategist Guillaume Soria to top sales and trading clients on September 15, that day Goldman's 3-month momentum factor rose 5% in a single day, while the 12-month momentum factor fell 6.7% the same day, marking the widest single-day performance gap between the two in five years; Goldman's AI thematic basket (GSPUARTI) has cumulatively drawn down nearly 45% from its high, the deepest drawdown since ChatGPT's release. Goldman noted that the correlation between the momentum factor and the AI theme across different horizons remains as high as 90%–96%, and explicitly recommended that clients holding AI exposure buy hedges — the indicated cost of one-month, 95% strike put options on its AI beneficiary basket is about 2.02%, with a term of 27 days.

In other words, clients' question is no longer "do I want AI" but "is the AI in my portfolio sitting in chips or in software, in the physical layer or the application layer, and does it fall the same way as the broad market when things go down." This is exactly what Haines meant: "Investors today often demand very specific types of exposure, whether it's a sector, a market cap segment, or a particular country's exposure. And they want to be able to slice exposure precisely to fit their portfolios."

The valuation backdrop is also amplifying this demand. According to MSCI's forward equity risk premium model, the forward equity risk premium for US stocks has fallen to 0.41%, close to levels around the bursting of the internet bubble in 2001 (0.01% in December 2001, with a historic low of -3.90% in December 1999), far below the 4.39% average since 1984 — the market has left very little margin for error for "high valuations plus high real rates," and the falsification of any single link could be amplified into volatility across the entire chain.

The difficulty of breaking it apart: even an "AI index" itself holds Eli Lilly and GE Vernova

"Deconstructing AI" sounds simple, but the hard part is the methodology of layering. MSCI's own practice illustrates just how inconsistent this is. According to an analysis by wealth management firm deVere Group of an MSCI AI index with 100 constituents (as of August 31), the index screens from MSCI World, MSCI China, and MSCI Korea, uses MKT MediaStats' AI relevance scores, is market-cap weighted with a 10% cap on any single constituent — and among its top ten holdings, after Microsoft (11.67%) and Meta (8.78%), the third-largest weight is pharmaceutical company Eli Lilly (8.59%), with grid equipment maker GE Vernova and two cybersecurity companies also mixed in, while chip leader Nvidia did not make the top ten. MSCI does not disclose constituent-level relevance scores, and the institution speculates that the semantic profile of semiconductor manufacturers may be one reason for the omission.

The same company is also running multiple AI products with completely different methodologies. According to MSCI's website, its "Transatlantic AI Industry Select 40 Index" consists of 20 AI-related industry leaders each from the United States and Europe, covering software, semiconductors, and renewable power, with constituents' combined market capitalization of $17.57 trillion, a price-to-earnings ratio of 33.22 times, and a forward price-to-earnings ratio of 20.79 times (as of August 31). Meanwhile, according to MSCI's "2026 Investment Trends Report," its AI value chain research framework spans 9 layers and includes about 230 companies, with roughly 80% by market capitalization in the United States — these companies' capital expenditure grew 35% year over year, exceeding $700 billion, or about 20% of global capital expenditure; research and development spending is close to $600 billion, more than 40% of global R&D; and earnings growth in 2026 is expected to exceed 20%. The 14 value chain indexes launched this time amount to turning this research framework into tradable benchmarks.

Competitors have not been idle either. According to a Nasdaq trader notice, three new indexes including the "Nasdaq Global Artificial Intelligence and Big Data Index" took effect on September 10; according to a Solactive announcement, its "Dan Ives Wedbush AI Revolution Index" (rebalanced on September 21) and "Solactive United States Artificial Intelligence Index" (reweighted on October 1) also underwent intensive adjustments in September. deVere's analysis states plainly: there is no single benchmark for AI — MSCI alone runs multiple AI indexes with different methodologies, while Nasdaq, S&P, and Solactive each have their own versions — returns vary by currency, and volatility varies by comparison benchmark. For institutions, "which index to use to define AI" has itself become a risk management decision.

Hedging demand is already on the table: investment banks are selling protection, exchanges are building pipelines

If indexes are the shelves, derivatives are the pipelines. According to reports, SGX announced after market close on July 23 an expanded licensing agreement with MSCI to launch as many as 100 new derivatives linked to MSCI indexes; in the first phase, 40 futures and options contracts will be introduced first, covering MSCI flagship developed-market benchmarks, major single-country indexes in Asia-Pacific, and Asian emerging-market sector indexes. The reports said this batch of products has now officially gone live, covering developed and emerging markets as well as major sector indexes such as utilities, industrials, energy, and financials.

This cooperation also carries a historical footnote: six years ago, MSCI moved the derivatives licensing for a series of indexes from Singapore to Hong Kong. SGX CEO Loh Boon Chye said at the time that as portfolio management increasingly spans different regions, themes, and benchmark combinations, a comprehensive MSCI product suite allows investors to "manage global equity risk through one trusted channel"; MSCI Chairman and CEO Henry Fernandez said the agreement reflects MSCI's commitment to ensuring investors can access MSCI's most important benchmark indexes wherever they conduct risk management.

On the buy side, hedging has shifted from "whether to do it" to "what to use and how much to spend." In addition to the put option solution Goldman Sachs provided, according to Bank of America Global Research's "Global Equity Volatility Insights," the bank recommends staying long technology but expressing it through asymmetric structures — buying out-of-the-money QQQ October call options, while hedging macro rate risk in the bond market using a combination of TLT puts and call spreads; Bank of America also noted that AI-related heavyweights in the Japanese market already account for 27% of the Nikkei 225, versus only 5.5% for TOPIX, and the difference in sensitivity of the two benchmarks to the same theme is itself structural.

The necessity of layered trading is also supported by data in emerging markets. According to RBC Global Asset Management's autumn outlook (as of August 31), the top five constituents of the MSCI Emerging Markets Index have reached about 35% in weight, "index concentration unprecedented" — and the top five happen to be AI hardware heavyweights such as TSMC, Samsung Electronics, and SK Hynix; the firm estimates that another 100–150 companies in emerging markets are related to AI through links such as semiconductor design, ASICs, power infrastructure, cooling technology, and testing equipment. JPMorgan Asset Management argues from the other side (index exposure data as of August 11): European benchmark indexes have a lower weight in AI "enablers" than the S&P 500 and are more "adopters," which makes Europe relatively resilient when AI stumbles while still sharing in infrastructure dividends when AI delivers.

Returning to Haines' point — what these AI value chain indexes give investors is modular, rules-based building blocks that let them raise or lower exposure layer by layer, rather than treating AI as an all-or-nothing bet that cannot be selectively deconstructed. When divergence within a single chain is as wide as "semiconductors +55%, software -18%," what index providers are really selling is not a new index, but a new granularity of risk management. Whether that granularity can outperform the "all-in bet" will depend on who validates it in the next earnings season.