Positioning with conviction through the AI selloff - July Trust Update

Key events

• Middle East tensions drove oil prices sharply higher, adding to a volatile month for the tech sector
• After strong returns in the first half of the year, semiconductor and other AI infrastructure companies sold off sharply
• Big tech firms raised AI spending plans, reinforcing our long-term confidence in the AI theme

Market review

Global stock markets were volatile in July, though overall returns were fairly muted: the MSCI All Country World Net Total Return Index declined -1.3% and the S&P 500 fell -1.7%. European shares fared better, with the DJ Euro Stoxx 600 Index up +0.6%. Smaller companies lagged larger ones, as the Russell 1000 Index of large-cap US shares returned -2.0% against -4.6% fall for the small-cap Russell 2000 Index (all returns in sterling terms). 

A sharp selloff in AI and semiconductor shares dominated the headlines, alongside renewed conflict in the Middle East. The technology-heavy NASDAQ 100 fell -8.1% and the Philadelphia Semiconductor Index (SOX) dropped -21.9%, its worst month since 2008, while South Korea's memory-focused KOSPI fell -17.6% despite a late rebound.

Commodities were led higher by crude oil, the best-performing major asset class in the month. Prices rose after President Trump declared that June's agreement with Iran was "over" and both sides resumed strikes. Prices later eased back as tensions showed tentative signs of stabilising, leaving Brent up +23.6% over the month and close to +50% higher so far this year, though still some way below its spring peak. These inflationary pressures pushed global bond yields higher even as major central banks held interest rates steady.

The Federal Reserve (Fed) left its key interest rate unchanged at 3.50-3.75% for a fifth consecutive meeting, though three policymakers voted in favour of a rate rise. A softer-than-expected reading of US inflation for June, flat on a month-on-month basis against expectations of +0.2%, helped keep short-term borrowing costs in check. The Bank of Japan also left policy unchanged while signalling a likely rate rise in autumn. The yen ended the month as the best-performing major currency, rebounding after apparent US-coordinated intervention on 30 July following its weakest level against the dollar since 1986. 

Technology review

The technology sector underperformed the broader market in July, with the Dow Jones Global Technology Net Total Return Index (W1TECN) returning -5.8% against-1.3% for the broader MSCI All Country World Net Total Return Index.

The Philadelphia Semiconductor Index (SOX) bore the brunt of the selloff, down -21.9%, while the NASDAQ Internet Index (QNET) and iShares Expanded Tech-Software ETF (IGV) returned +5.9% and +2.7% respectively. Large technology companies outperformed small and mid-sized ones, with the Russell 1000 Technology Index down -4.6% against a -13.7% fall for the Russell 2000 Technology Index (all returns in sterling terms).

The supply chain behind AI infrastructure saw heightened volatility despite very strong underlying demand that is still running ahead of supply, and AI-related companies continuing to post strong results, together driving around a third of Q2 earnings growth for the S&P 500. Indeed, S&P 500 earnings per share (EPS) growth for Q2 is tracking at +26% year-on-year (y/y), excluding gains from the largest technology companies' equity investments, with nearly two-thirds of S&P 500 companies beating profit expectations by more than one standard deviation¹, one of the highest rates on record, according to Goldman Sachs.

We believe the July selloff was driven mostly by technical factors rather than a change in the underlying fundamentals.

In semiconductors, TSMC delivered a strong quarter, with revenue up +34% y/y and guidance ahead of expectations. Demand for high-performance computing chips used in AI remained the main growth driver, supported by robust AI demand, while smartphone demand stayed weak against a backdrop of high memory prices. Although gross margin guidance disappointed some investors, reflecting the ramp-up of its new 2nm chip technology and costs from expanding overseas, demand for the most advanced AI chip capacity continues to outstrip supply. Management raised full-year revenue growth guidance to more than 40%, lifted 2026 capital expenditure (capex) guidance to $60-64bn from $56bn, and announced a further $100bn investment in its Arizona manufacturing cluster, underscoring confidence in sustained AI demand.

Intel also delivered a strong quarter, with revenue up +25% y/y, its strongest growth in 15 years, driven by server chips and its AI infrastructure-related chip design business. Its manufacturing operations continue to improve, with strong progress on its next-generation 18A and 14A chip technologies. Management raised 2026 capex guidance to more than $20bn from $15bn and indicated that 2027 spending would be significantly higher.

Corning, an optical fibre and specialty glass manufacturer, reported results modestly ahead of expectations, supported by strength in optical communications and solar. However, softer Q3 revenue guidance raised some concerns about the near-term pace of AI-driven growth. Management reiterated confidence in the longer-term outlook, pointing to strong demand for optical connectivity used in scaling up data centres, and expanding opportunities across AI infrastructure, which it believes can support average annual sales growth of around 19% through 2030.

Memory chip makers, including Sandisk and Kioxia Holdings*, also weakened in July despite supportive conditions for DRAM, NAND2 flash memory and high bandwidth memory (HBM). The selloff reflected investors taking profits in AI-related shares as they questioned the durability of AI capex growth. Sentiment was further pressured by rising concerns about Chinese memory competition, particularly the expansion ambitions of Chinese producer CXMT* and broader domestic semiconductor investment, although even CXMT remains constrained by capacity and is understood to be seeking significantly higher prices from Western customers.

KLA reported solid Q2 results, with revenue and margins ahead of expectations, supported by strong demand for chip-inspection equipment used in advanced manufacturing. However, Q3 revenue guidance was slightly below elevated investor expectations, weighing on its shares and those of other semiconductor equipment companies, including Applied Materials* and Tokyo Electron*. Management remained upbeat, raising its forecast for the wafer fabrication equipment market and expressing confidence in continued growth in advanced chip packaging and process control through 2027.

In the internet sector, Alphabet (we are underweight (u/w) compared to the benchmark), delivered a solid quarter, with Search revenue growing +17% y/y and Cloud revenue growth accelerating to +82% y/y, ahead of expectations. Management raised 2026 capex guidance to $195-205bn from $180-190bn to support growing AI infrastructure demand, although this weighed on the outlook for profit margins and free cash flow. Investors were also disappointed by cautious comments on future AI models, with no timeline given for its next-generation Gemini model.

Meta Platforms (Meta; u/w) delivered a mixed quarter, with revenue growing +28% y/y and advertising trends remaining healthy, supported by higher impressions and pricing. Investors focused on higher-than-expected operating expenses and a rise in the lower end of 2026 capex guidance to $130-145bn from $125-145bn. News that Meta was considering offering AI compute capacity as a service was initially misread as a sign that its own AI plans were stalling, or that it had overbuilt capacity and might cut future capex. Those fears were later eased by chief executive Mark Zuckerberg, who said he does not know anyone in the industry who feels they have too much compute capacity and that Meta is currently using all the computing power it has.

In our view, while this has added to near-term turbulence, it likely remains supportive of the broader AI investment case. The so-called 'AI race' is clearly alive and well...

Amazon delivered one of its strongest quarters in some time, with revenue up +20% y/y, supported by robust growth across Amazon Web Services (AWS), retail and advertising. AWS revenue increased +37%, while margins came in well ahead of forecasts. Management raised 2026 capex guidance to $220bn from $200bn to support continued investment in AI infrastructure, reinforcing confidence in the company's long-term growth outlook. Amazon chief executive Andy Jassy delivered a strong message underscoring the company's confidence in future returns on its AI spending, which helped address broader investor concerns about returns on hyperscaler capex, important given companies are planning sustained capex growth into 2027 and beyond.

In software, Microsoft (u/w) delivered a strong quarter, with Azure cloud revenue growth accelerating to +43% y/y, ahead of expectations, driven by robust AI demand and improving compute efficiency. Microsoft 365 Commercial growth was guided to accelerate, while adoption of its Copilot AI assistant exceeded forecasts. Management indicated that underlying 2026 AI infrastructure investment remained broadly unchanged, although capex is expected to increase further in 2027 to support continued strong demand.

Apple (u/w) outperformed through July as investors rotated into technology companies with lower capex needs, favouring its cash-generative business model over the heavy infrastructure spending pursued by other large technology companies. Shares gave back some of these gains after a mixed update, which also marked Tim Cook's final appearance as chief executive before handing leadership to John Ternus. Revenue was broadly in line with expectations, as strong Mac and iPhone sales were offset by weaker Services growth and softer demand in Greater China. Management guided to slower revenue growth in Q3, citing increasing supply constraints at the most advanced chip factories.

Outlook

We believe the July selloff was driven mostly by technical factors rather than a change in the underlying fundamentals. Investors taking profits ahead of Q2 earnings, a more cautious mood heading into the quieter summer months, hedge funds reducing their exposure and losses for retail investors in highly borrowed products all led to a sharp unwinding of momentum, with the year's winning shares sold off and heavily shorted³shares, including software, rebounding sharply.

This unwinding of momentum also intensified some existing, more fundamental concerns around competition from open-source AI models, the capacity to fund AI infrastructure and the return on investment (ROI) from AI. These concerns, which cannot really be proved or disproved in the near term, found a more receptive audience as share prices were falling, although none of them are new or materially different from earlier in the year.

In our view, competition from rival AI models remains the greatest near-term challenge. Investors faced echoes of an earlier period of concern as new AI models appeared to narrow the gap with the most advanced systems, often at a much lower cost. These included closed-source models such as Meta's Muse Spark 1.1 and SpaceX's Grok 4.5, as well as Chinese open-source models from Zhipu, Kimi, Alibaba and DeepSeek. This raised questions about how durable the competitive advantages are and the ability to fund infrastructure investment, at leading AI companies such as Anthropic. High expectations and crowded positioning left AI-related shares very sensitive to any sign of weaker returns on investment, while investors continued to question how the value created by AI will ultimately be shared out across the industry.

Our long-term conviction in the AI investment theme remains unchanged, and we have used the weakness to add to our preferred positions.

In our view, while this has added to near-term turbulence, it likely remains supportive of the broader AI investment case. The so-called 'AI race' is clearly alive and well, contrary to fears that some competitors could drop out, and the scaling laws that have driven AI progress so far, where increases in the size of models, the data used to train them and the compute power involved tend to produce predictable gains in capability, look firmly intact. If anything, competition for AI leadership appears to be intensifying.

Investors also became increasingly concerned about the ability and willingness of credit markets to fund future spending by the hyperscalers. Goldman Sachs estimates $489bn of AI-related debt has been issued so far this year, 40% directly by the hyperscalers themselves, against $322bn for the whole of 2025. This year has already seen 10 new bond issues of investment-grade debt greater than $20bn, against just 11 such deals in the prior seven years combined. It is therefore unsurprising that, as debates about returns and competition intensified, long-dated borrowing costs for hyperscale companies rose sharply during the month, before easing back on strong earnings results.

Concerns about circular financing, where companies invest in and lend to one another to fund AI expansion, also resurfaced towards the end of July after Bloomberg reported that NVIDIA was pursuing a fresh round of AI deals worth more than $750bn. It is an unavoidable fact that building infrastructure on this scale will require a huge amount of funding and that the capital-intensive nature of AI model training, along with the need to build capacity ahead of profitable inference demand for using AI models, makes the need for capital particularly pronounced while the technology, its ecosystem and business models are still relatively new.

While the scale of investment required is substantial, 2026 AI investment is estimated at $1trn globally, and just under $600bn in the US, and growing, these levels are consistent with those seen in earlier waves of transformative new technology, according to Goldman Sachs. Based on analyst forecasts for company spending, we estimate AI capex will rise from 1.8% of US economic output (0.9% globally) in 2026, to 2.5% (1.3% globally) in 2027 and to 2.8% (1.4% globally) in 2028.

Sentiment improved towards month-end after OpenAI reported a sharp acceleration in July, adding more annual recurring revenue than in the previous three months combined. Competition between AI models also appears supportive of AI infrastructure demand in the near term: Silicon Data, a provider of pricing data for Graphics Processing Units (GPUs), specialised chips widely used to power AI, indicated that "GPU rental price indices we track show no signs of softening demand for compute whatsoever. On-demand rental prices continue to either rise or hold firm across all major GPU types".

Q2 results from the hyperscalers clearly benefited from firm pricing, with a record addition of new recurring revenue across all the major cloud providers, and better-than-expected profitability providing tangible evidence of strong returns on investment. Combined cloud revenue growth accelerated to +48% y/y from +39% in Q1 and order backlogs remain up more than 100% y/y, reaching more than $2.4trn. Amazon's Andy Jassy gave more detail on the company's "clear line of sight to strong financial returns." Amazon expects to be supply-constrained in 2026 and 2027 while demand for 2028 is "striking" enough to support what could "very possibly be $1trn annual revenue business for us in time." Given strong returns and improving AI capabilities, it is not surprising that overall estimates for hyperscaler capex in 2027 increased by a further $125bn following Q2 results.

While not immediately obvious at the time, it later became clear that the collapse of former OpenAI researcher Leopold Aschenbrenner's high-profile Situational Awareness hedge fund, which had borrowed heavily to increase the size of its investments and was worth around $45bn at its peak, played a significant role in the unwinding of momentum, which culminated in the biggest three-day reduction in risk and borrowing since November 2022, according to data from prime brokers.

Volatility could continue through August, given this is traditionally a quieter period during which investors tend to take some risk off the table and lock in gains. Despite this, our long-term conviction in the AI investment theme remains unchanged, and we have used the weakness to add to our preferred positions, taking the Trust back to a more fully invested position.

While debates will persist, we expect sentiment to recover as AI progress continues rapidly, supply constraints ease and positive earnings revisions come through. We are mindful of the impact of powerful new AI models: OpenAI's Astra model has reportedly solved 10 major open problems in mathematics, quantum physics and computer science. Upcoming high-profile AI stock market listings, including Anthropic and, at some stage, OpenAI, could also help restore confidence in the AI investment case by removing uncertainty about their ability to fund growing capex requirements.

AI progress and adoption are accelerating, and we believe the recent choppiness is best understood as an unavoidable period of digestion. The AI industry has already scaled to a $175bn annual revenue run rate in little more than three years but the next stage will require supply to expand further to meet accelerating demand. New supply coming online often brings complexity and volatility, even as the underlying AI fundamentals improve. We would encourage investors to be prepared for further volatility, but not to lose sight of the extraordinary transformation of the economy taking place around us, and the opportunities we believe this offers to equity investors.