The AI divide and the narrow road to outperforming the S&P 500

In a conventional bull market, a rising tide tends to lift most boats, and we see that performance across most market sectors. However, we are currently experiencing one of the narrowest-performing markets in recent history, where only a handful of sectors are outperforming the S&P 500 Index, which has risen by an incredible 18% over the past 12 months (note that this article uses data to 25 August 2026). The market is rewarding AI beneficiaries at one extreme and AI-resistant businesses at the other. This unusual dispersion helps explain why active management has found the current environment so difficult.

The Wall Street Journal, inan article published on 15 August, highlighted the challenge, noting that only 27% of actively managed US large-cap equity funds had outperformed their passive benchmarks over the 12 months to 30 June 2026. The author captured the dilemma neatly: "Investment pros say AI disruption has created a stock picker's market, but they still seem to be struggling to pick the right ones".

So, what exactly did you have to pick to beat the market?

Over the past 12 months, only 38% of S&P 500 constituents have outperformed the market as a whole. This is an unusually low proportion given the strength of the overall market. A quarter of those companies, or 10% of the S&P 500, were direct or indirect beneficiaries of the AI investment cycle. These include semiconductor manufacturers or companies involved in the enormous infrastructure build-out required to support AI data centres.

Another group benefitted from elevated energy prices following the war in Iran, with twenty companies, or 4% of the S&P 500, exposed to energy and related themes. The interesting part was that among the rest of the outperformers, or 25% of the S&P 500, the common theme was that they were the opposite of the AI beneficiaries. These are businesses that the market considers very unlikely to be disrupted by AI but are not necessarily making more money in this environment. They included smokestack businesses like Philip Morris, Coca-Cola, Ross Stores, Monster Beverages, and Union Pacific, as well as defence companies like Lockheed Martin. Their common characteristic was not necessarily exceptional earnings growth. Rather, investors viewed their business models as relatively difficult for AI to disrupt.

Ironically, the market has put the tech giant Apple in this category as well. Unlike the major technology companies funding the AI infrastructure boom, Apple's strategy remains predominantly personal hardware-focused, and it has not committed anything approaching the same level of capital to AI data centres. Investors appear to have rewarded that relative restraint.

The result is an unusual market dynamic: investors have been buying both the companies expected to benefit most from AI and those expected to be least affected by it. Very little in between has been rewarded. It is not just the AI companies' share prices that are doing well; the AI beneficiaries are also expected to generate the biggest chunk of the S&P 500's earnings growth over the next few years:

The chart above shows AI beneficiaries in the S&P 500 in blue vs the rest of the S&P 500 in yellow. For 1Q27, consensus expectations suggest that EPS growth among AI beneficiaries could peak at 60%, far greater than the rest of the market, which is growing at 15%. This helps explain why investors have been willing to pay increasingly high valuations for companies directly exposed to the AI investment cycle. The market is not simply buying the AI narrative; it is increasingly pricing in a meaningful earnings advantage.

The concentration becomes even more obvious when looking at individual share-price performances.

Over the past 12 months, 15 of the 20 best-performing S&P 500 shares have come from the information technology sector, and virtually all these businesses are considered beneficiaries of the AI build-out. The standout performer has been SanDisk, the memory chip manufacturer. Memory prices have surged because of insatiable demand for data centres. The one particularly striking exception is Moderna, which rose 150% on the day it announced a possible cure for melanoma using its mRNA technology, which was famous for its use in COVID-19 vaccines.

A small company whose share price rises significantly still cannot materially impact the index on its own. Because the index is market-capitalisation weighted, the performance of the largest US companies matters more, as they have the biggest influence on the S&P 500.

Among the twenty largest US companies, 11 have outperformed the market over the past twelve months. Almost all of these are tech companies and direct beneficiaries of the AI investment cycle. The other winners were again related to industries that are difficult to disrupt (the AI-resistant category).

Perhaps the most interesting observation is Nvidia. Despite being one of the clearest beneficiaries of AI and one of the largest shares on the market, Nvidia has not outperformed the S&P 500 over the past year. There are numerous explanations for this, but most of it is related to uncertainty around the long-term revenue generation of the AI theme, combined with already having performed so strongly over the past 3 years and perhaps needing to catch its breath. Nvidia is a US$5trn company, after all, and as the old market saying goes: "trees don't grow to the sky".

Alphabet provides another interesting example. After a phenomenal 2025, it has underperformed in 2026. The same is true of several other members of the tech-heavy Magnificent Seven, including Microsoft, Meta, Tesla and Amazon. The common theme across these businesses is that they are spending unprecedented amounts of money funding the AI capex cycle and consequently no longer have positive free cash flow and are taking on debt to finance their investments.

Figure 4 shows the consensus expectation for capex between 2020 and 2030 and the annual expected growth rate. At present, the market expects the five largest hyperscalers to spend US$700bn-US$750bn on AI data centres in 2026, an increase of around 85% from an already massive level of expenditure in 2025. This is where the current market dynamic becomes particularly interesting. Those companies selling the picks and shovels into the AI gold rush can generate immediate revenue from the investment. The hyperscalers, by contrast, must spend billions today in the hope of generating sufficiently attractive returns from AI-related services over many years. When this level of capex is measured against the companies' overall revenue generation (i.e. not just their AI revenue), the scale becomes extraordinary.

Figure 5 below shows that capital expenditure as a percentage of sales for these businesses was often closer to 15% of revenue in the past. Today, that figure is approaching 50%. This is an extraordinary increase that has started to worry the market and is the main reason the hyperscalers have underperformed the AI beneficiaries this year. The market is communicating that it does not believe they can make a meaningful enough return on this investment (at least over the medium term). It places considerable pressure on management teams to demonstrate that today's capital expenditure will translate into tomorrow's earnings, especially while their share prices remain under pressure.

The above dynamic explains the extraordinary performance of large caps such as Micron, Advanced Micro Devices (AMD), and Intel, which have been the biggest beneficiaries of the current environment and the single largest contributors to the market's performance. If you did not own these big companies that went up hundreds of percent then it was extremely difficult to outperform the overall index. For investors, the challenge has not simply been to recognise that AI is transforming the economy but to identify exactly where in the AI value chain economic value is accruing.

Owning the companies spending hundreds of billions of dollars on AI infrastructure has not necessarily produced the best returns. Owning the companies selling the infrastructure to those hyperscalers has often been much more rewarding.

The same dynamic extends beyond semiconductors. Just outside the top 20 performers, Caterpillar has returned about 88%, benefitting from data-centre construction: its reciprocating engines provide backup and increasingly primary power for facilities straining local electrical grids. The AI investment cycle is therefore creating winners well beyond the obvious technology names.

So, what did you have to own to outperform the S&P 500 over the past year? The uncomfortable answer is that investors needed meaningful exposure to a small cluster of AI beneficiaries, particularly the memory and chip makers like SanDisk, Micron and AMD, while largely sidestepping the hyperscalers carrying the enormous capex burden. At the same time, it helped round out the rest of your portfolio with the "undisruptable" smokestack names the market ran to for safety.

That is an extraordinarily narrow needle to thread, which is exactly why active managers have struggled to outperform. The deeper lesson is that this market has effectively split into two camps. On the one side, companies that investors believe will win from AI. On the other, companies they believe AI cannot touch. Very little in between has been rewarded. The important question for investors now is whether this divide can persist. At some point, the hyperscalers will have to demonstrate that the three-quarters of a trillion dollars they are pouring into data centres and AI infrastructure can generate attractive returns. If the answer turns out to be yes, today's underperforming giants may become tomorrow's leaders. If it is no, the narrowness that defined the past twelve months could unwind just as quickly as it formed. Either way, the tide is no longer lifting all boats, and investors would do well to know which side of the AI divide they are standing on.