Regardless of the return on investment that the hyperscalers will eventually achieve, the capex boom is having an extremely positive impact on the recipients of this spending today.
“Is AI a bubble?” has been one of the most important, and most difficult, questions facing investors since the pandemic. In our 2026 Investment Outlook (published last November), we explained why we believed it was impossible to make a definitive call on whether the booming levels of capex related to artificial intelligence (AI) would deliver a sufficient return on investment. This was because it was simply too early to tell the extent to which AI would enhance corporate profitability and how much enterprises would be willing to pay for it. We therefore argued that tech exposure was still an important part of equity allocations but required careful risk management.
What we have learned over the past six months has generally been positive for the tech sector. Survey-based indicators highlight increasing corporate adoption, with the Ramp AI Index now suggesting that more than half of companies surveyed are paying for an AI subscription, up from 40% a year ago. Anthropic’s latest update to the Claude family of large language models (LLMs) helped to add to the excitement, with the new release more clearly targeted at augmenting professional tasks.
We also take reassurance from increased return dispersion in tech-related areas of the market. Dispersion has risen across subsectors (software vs. semiconductors), across stocks within subsectors (US hyperscalers) and across regions. The implication is that investors are scrutinising individual company fundamentals rather than placing options on the overall market, which would be more common behaviour in the late-stage euphoria of a bubble forming. Increased dispersion also implies a stronger environment for active management to generate alpha.
It is clear, however, that questions remain: the effect that AI will have on the global economy and the profits that can be reaped from it, as well as who will take the largest share of the profit pie.
To avoid the brunt of these uncertainties, investors are demonstrating a preference for parts of the AI supply chain – such as high bandwidth memory – where bottlenecks are allowing companies to achieve strong pricing power and generate robust profits. Conversely, industries that are perceived to be most at risk of disruption have been punished hard: software and media companies are down 3% and 32% respectively since the start of 2025 (see Exhibit 7).
Below we outline three strategies to capitalise on the tech opportunities available while seeking to guard against areas of potential risk.
Prepare for hyperscaler volatility
US hyperscalers are going all-in to win the AI race. Sell-side analyst estimates for 2026 capex are now a remarkable USD 697 billion for just five US companies, up by USD 173 billion since the beginning of the year. At the same time, enormous financing requirements are pressuring balance sheets for some names. As a result, pairwise correlations – the extent to which stocks are moving in the same direction – among the US hyperscalers have reached new lows (see Exhibit 8). The spread of performance between the best (Alphabet) and worst (Meta) performing US hyperscaler over the past year to 8 June is a remarkable 125 percentage points.
Given the uncertainty about who the winners will be, investors are now conditioning the reward for additional capex spending on evidence of greater final demand, which is a trend we expect to continue. In the latest earnings season, higher capex plans only drove stronger performance when matched with higher revenue estimates. Balance sheet strength will also remain under scrutiny. With cash piles run down, AI capex has gone from 33% of the hyperscalers’ cash flow from operations in 2023 to an estimated 93% in 2026.
Ultimately, for the hyperscalers to win back investors’ favour, the equation is clear: demonstrate that demand is sufficient to deliver a positive return on investment, and do so quickly enough to avoid placing too great a strain on existing cashflow.
The AI supply chain should see less noise
Regardless of the return on investment that the hyperscalers will eventually achieve, the capex boom is having an extremely positive impact on the recipients of this spending today. Companies that manufacture semiconductors, memory and electronic components are all seeing incredible earnings upgrades, as are companies linked to the build-out of data centres.
Consensus estimates see US semiconductor companies’ earnings growing by 98% in 2026, more than five times faster than those of the US hyperscalers (see Exhibit 9). While Nvidia remains the market leader in designing cutting-edge chips – with operating margins of 75% – the profitability of companies linked to semiconductor production has been soaring around the world. Since January, the 2026 earnings-per-share (EPS) estimates for Asian, US and European semiconductor names have been revised up by 89%, 22% and 17% respectively.
In the long run, it is unsustainable for one subsector to capture such a large share of the total profit pie. Some degree of reconvergence is therefore likely, either through a “catch up” scenario, where AI demand validates the hyperscalers’ capex, who will require a bigger share of the overall gains. Or via a “catch down” scenario, where a pullback in AI capex would likely hit semiconductors harder than other subsectors, given their recent rise and the magnitude of earnings expectations already baked in.
While the extent of the recent semiconductor outperformance cannot continue indefinitely, valuations provide a reason to remain constructive. US semiconductor valuations have actually fallen over the past year, and at 22x 12-month forward earnings, their valuation is only one point higher than the valuation of the S&P 500. Given the expected earnings growth trajectory, this looks like a relatively small margin.
One might argue that unrealistic earnings growth assumptions are artificially lowering valuations. Measuring the trailing price-to-earnings growth (PEG) ratio is, therefore, useful in that it compares the P/E valuation with realised earnings growth rather than expected earnings growth. Surprisingly, the “backward looking PEG” for the US semiconductor sector is not only lower than that of other subsectors involved in the AI build-out, but has also fallen more than others over the past year (see Exhibit 10).
The opportunity is now global
The weight of tech exposure in regional benchmarks has once again been a key determinant of performance year to date (see Exhibit 11). While US and Chinese benchmarks have the largest exposure to hyperscalers, huge amounts of AI capex have flowed to suppliers in other countries, especially Korea and Taiwan. Earnings growth in emerging markets has been so strong that valuations have fallen in 2026 despite returns of 18%, with the tech sector having contributed more than 100% of this return (see Exhibit 12).
With AI increasingly seen as a critical part of countries’ national security, how the AI build-out evolves geographically bears watching. The US and China both appear determined to win the AI race, with the US leading in terms of frontier innovation “blueprints”, although China’s open-source/open-weight models offer much lower costs. (It remains to be seen whether other regions, including Europe, will be open to utilising AI capacity provided by others or whether they will become more determined to create their own domestic capabilities.)
China’s strategy appears focused on achieving the broadest usage rather than the most powerful models, with AI agents and models aiming to become a commodity that every business can utilise. While the depth of adoption remains nascent, Chinese businesses seem eager to opt in: over 50% of small and medium-sized enterprises are already applying AI in operations and 27% plan to adopt it within the next two years.
China is also advanced in deploying robotics at scale, and seems more able and more willing to field real-world applications (for example, drones and uncrewed mobility), aided by a less restrictive domestic regulatory posture. Almost 40% of China’s AI-related investment goes to robotics.
Finally, China has a “physical edge” thanks to its ample and rising power capacity, as well as its supply of critical minerals. This advantage allows the country to build data centres at a much faster pace than anywhere else in the world.
We see stronger investment opportunities today in Asian companies involved in the AI build-out (semiconductors, robotics, industrial equipment etc., both in China and broader emerging Asia) rather than in the Chinese hyperscalers. Because of Chinese government subsidies, the uncertainty over the ability of Chinese hyperscalers to monetise their current investments is higher relative to their US peers.
Capturing the pie, minimising volatility
AI-related developments since we published our 2026 outlook have generally been positive. Rising dispersion across the AI ecosystem suggests that investors are scrutinising company fundamentals. This process of discovery will continue to generate volatility, as we learn both how AI will influence the economic pie, as well as who will garner the largest slice. We suspect that the recipients of capex spend will enjoy a smoother ride.
Until there is much greater evidence to answer the return-on-investment question, a selective approach remains warranted. The next key test lies in several high-profile initial public offerings ahead, which will gauge investors’ appetite for yet more AI exposure. We will tackle this subject in greater detail over the coming weeks, but ultimately don’t expect these events to change the narrative. Signs of corporate demand for AI, whether good or bad, are likely to be the more important signal over the coming quarters.