The scale of AI is still being underestimated - May Trust Update

Key events

• Easing Middle East tensions pushed oil prices sharply lower, lifting equity markets and calming inflation fears
• Technology stocks delivered strong returns led by semiconductor and hardware companies following robust AI-driven earnings
• Real-world AI usage is growing at a breathtaking pace, with demand continuing to outstrip available supply

Market review

Global equity markets delivered strong gains in May as progress towards a US/Israel-Iran ceasefire deal drove a sharp reversal in oil prices. The MSCI All Country World Net Total Return Index returned +6% and the S&P 500 Index +6.1%. The DJ Euro Stoxx 600 Index returned +3.6% (all returns in sterling terms unless otherwise stated).

Small caps underperformed large caps with the Russell 1000 Growth Index (large and mid-cap stocks) +8.1% and the Russell 2000 Growth (small-cap stocks) +6.7%. Japan was a standout, with the Nikkei surging +10.9%, while South Korea's KOSPI extended its remarkable run, gaining a further +27.2% in May to leave year-to-date (YTD) gains at above 100%.

Equity markets rallied early in the month on reports that the US and Iran were close to a framework that would end hostilities, sending the Brent crude oil price from $114 on 4 May to $100. Optimism subsequently faded as Trump deemed Iran's proposal unacceptable. Bond markets bore the brunt of the resulting anxiety as the 30-year Treasury yield touched a post-2007 high of +5.18% midway through the month.

A slightly hot US core inflation number supported the view that inflation was proving stickier than hoped, even before the full energy price impact had passed through. By the end of the month, a 60-day memorandum of understanding to extend the ceasefire had been agreed, with nuclear negotiations to follow.

Brent crude ended the month at $92, down -19.3%, its biggest monthly decline since March 2020. With energy prices retreating, equity markets made new highs while the 10-year Treasury yield fell for seven straight days for the first time in over a year. The Federal Reserve (Fed) remained in a holding pattern as Kevin Warsh took over as Chair from Jerome Powell.

The technology sector continued to post strong returns, driven by an acceleration in AI-related names as investor confidence in sustained capital expenditure (capex) buildout reasserted itself. The tech-led move higher in broader markets this year has been predominantly driven by earnings as fundamentals continue to improve. Q1 reporting season saw +27% year-on-year (y/y) earnings growth. The S&P is now +11% YTD, while earnings expectations over the next 12 months have been revised c15% higher. Prior to the launch of ChatGPT in November 2022, Goldman Sachs was forecasting 2023 S&P earnings per share of around $224. Today, 2026 earnings per share are expected to be $340, 52% higher than pre-AI expectations. As such, approximately two-thirds of the S&P's rally since November 2022 can be explained by earnings growth revisions, meaning the rally has been built on more than just rising investor optimism.

Technology review

The technology sector outperformed the broader market again in May as investors looked beyond near-term macroeconomic and geopolitical uncertainty and focused on strengthening AI fundamentals.

Small and mid-cap technology stocks outperformed their large-cap peers during the month, with the Russell 2000 Technology Index and the Russell 1000 Technology Index returning +22.4% and +13.8%, respectively. The Magnificent Seven (Mag7) – Apple, Microsoft, Alphabet, Amazon, Meta Platforms, NVIDIA and Tesla – underperformed, returning +7.5%, reflecting the impact on cashflow of their AI infrastructure investments and/or intensifying competitive dynamics.

Samsung Electro-Mechanics (SEMCO) also benefited from its exposure to the accelerating AI infrastructure buildout, reporting strong results at the end of April, driven by high demand for its multi-layer ceramic capacitors (MLCC) and high-end substrates.

The Philadelphia Semiconductor Index (SOX) continued to lead, driven by robust AI-related demand across compute, memory, networking and broader semiconductor content, returning +22.2%. The SOX has gained c81% YTD, outperforming most if not all technology subsectors. Robust earnings growth has underpinned returns, with analysts' average earnings forecasts for the index over the next 12 months up +94%, according to Bloomberg.

First quarter earnings season continued to demonstrate the sector's strong fundamentals. In semiconductors, NVIDIA delivered another exceptional quarter, with revenue rising +20% quarter on quarter (q/q) and +85% y/y to $81.6bn, marking a fourth consecutive quarter of accelerating growth. Guidance implies further acceleration to +95% y/y growth next quarter, with gross margins steady at +75%. Overall, results reinforced confidence in the durability of AI spending and NVIDIA's leadership position, although the stock has lagged many other AI-exposed stocks in part due to broadening AI demand beyond ‘just’ the GPU (graphics processing unit).

Advanced Micro Devices (AMD) also delivered strong results, with revenue up +38% y/y and earnings per share ahead of expectations, driven by +57% y/y growth in data centre revenues from continued server CPU (central processing unit) market share gains. Guidance implied further acceleration to +46% y/y growth, alongside improving margins. AI-related strength saw management double its server CPU market opportunity to above $120bn by 2030. GPU commentary was also encouraging, with the MI450 (AMD’s data centre accelerator GPU) ramp on track and major hyperscaler partnerships progressing well.

Sandisk rallied during May after reporting another outstanding quarter at the end of April, with revenue up +251% y/y, driven by positive NAND – the storage technology used in data centres and consumer devices – pricing and a richer data centre mix. The key focus was newly signed multi-year supply agreements covering over one-third of FY27 bit supply, backed by $42bn of minimum commitments and above $11bn of financial guarantees, reinforcing long-term pricing and volume visibility.

Samsung Electro-Mechanics (SEMCO) also benefited from its exposure to the accelerating AI infrastructure buildout, reporting strong results at the end of April, driven by high demand for its multi-layer ceramic capacitors (MLCC) and high-end substrates. These components are in short supply, leading to increased pricing power and improving order visibility.

Flex, a leading global electronics manufacturing services provider, delivered positive results and announced plans to spin-off of its Cloud and Power Infrastructure business. The transaction should help unlock value by separating the faster-growing, higher-margin AI infrastructure business from the more stable core manufacturing operations and was well-received by the market.

There were pockets of weakness, largely due to profit-taking after an exceptionally strong recent run, particularly for optical stocks due to rumours of NVIDIA product delays that could push out some upside investors had hoped to see in the second half of the year. Celestica, Fujikura and Asia Vital Components all dragged on relative returns despite robust earnings reports. In the case of Celestica, strong quarterly growth was accompanied by a mix towards lower-margin AI programs. These are early in their ramp and cost-heavy due to supply constraints.

Asia Vital Components delivered a positive quarter, with record margins driven by accelerating AI server and liquid cooling demand, but expectations were already high. Management expects continued sequential revenue and earnings growth through 2026, supported by rising adoption of liquid cooling across NVIDIA’s next-generation AI server platform, Vera Rubin, and hyperscaler application-specific integrated circuit (ASIC) platforms. ASICs are chips designed for one specific task, and their increasing use is driving demand for more sophisticated cooling per rack.

Fujikura was unfortunately weak following a medium-term plan that disappointed elevated expectations, with higher investment, optical fibre costs and capacity constraints expected to weigh on near-term profit growth.

In software, Microsoft continued to underperform despite reporting solid results last month. Microsoft Azure and M365 both met expectations and were guided to accelerate next quarter, while Copilot adoption is picking up. Capex guidance was raised well above expectations due to very strong customer AI demand and the need for Microsoft to invest more aggressively in its first-party AI model and product offerings, the latter creating near-term uncertainty that continues to weigh on the stock.

Apple delivered a solid quarter, with revenue growth accelerating to +17% y/y and earnings per share ahead of expectations as margins proved more resilient than feared. iPhone revenue rose +22% y/y despite supply constraints, while Greater China remained particularly strong at +28% y/y. Services growth of +16.3% y/y was broadly in line. Guidance was notably ahead of consensus, implying continued strength in iPhone demand despite ongoing memory cost pressures.

Outlook

AI progress is feeding directly into demand as average enterprise AI token spend – tokens being the basic unit of text that AI models process – has increased 13x since January 2025, according to Ramp. Interestingly, despite prior fears of the opposite, high-end GPU prices have climbed sharply over the past two months, with prices for the B200 – NVIDIA's flagship AI chip – nearly doubling in the spot market. The rollout of higher per-token pricing plans also suggests that demand continues to overwhelm available supply. However, we expect that, as capacity grows in the second half of 2026 and through 2027, true demand will likely become more visible supporting upside to estimates. Goldman Sachs recently forecast that agentic AI – that takes actions autonomously on a user's behalf rather than simply answering questions – could drive token consumption 24x higher than current levels.

While we are 3.5+ years into the AI cycle, the enterprise AI adoption story only started in earnest this year, driven by improved (reliable) model performance and the maturation of agentic tooling.

Google’s annual developer conference, Google I/O 2026, keynote provided a compelling third-party validation of the acceleration in AI adoption and inference demand (inference being the computing required to run AI models in everyday use). Google is now processing 3.2 quadrillion tokens per month, up from 480 trillion a year ago and just 9.7 trillion two years ago. This is a 330x increase in two years and a 7x increase in the past 12 months alone, driven in part by the shift to reasoning models which consume significantly more tokens per interaction. The Gemini app has now surpassed 900 million monthly active users, more than doubling from 400 million a year ago. These figures speak to a broader and accelerating step-change in the velocity of AI consumption globally.

Anthropic's unveiling of its Mythos model – a meaningful step-change across coding, reasoning and cybersecurity benchmarks – points firmly to continued model progress, with scaling laws intact. We are only now seeing the first models trained on NVIDIA's Blackwell architecture and further capability gains are expected as larger clusters of more powerful hardware come online through 2026-27: AMD's MI450, NVIDIA's Rubin Ultra and Alphabet's next-generation TPUs – all custom-designed AI chips.

Importantly, this acceleration is happening at exactly the point that supply remains structurally tight. Capacity constraints also serve to prevent this investment cycle from becoming a bubble, at least in the near term. TSMC's relatively conservative 2026 capex of $52-56bn, combined with the multi-year lead times required to construct hyperscale data centres and connect the necessary power, should mean supply growth remains constrained even as demand explodes.

While we are 3.5+ years into the AI cycle, the enterprise AI adoption story only started in earnest this year, driven by improved (reliable) model performance and the maturation of agentic tooling. NVIDIA has observed that token demand is doubling every two months and, most striking and supportive of this view, Anthropic crossed $47bn annual recurring revenue (ARR) during the month, up from $9bn at the start of the year and $1bn only 16 months ago. Anthropic also raised $65bn in a Series H funding round at a $965bn post-money valuation, higher than the latest OpenAI valuation at its recent round.

Both Anthropic and OpenAI have reportedly recently filed confidentially for IPOs that are expected in the next 3-4 months. These will follow a recent surprisingly oversubscribed $85bn equity offering by Alphabet as well as the imminent (as we write) record-breaking SpaceX IPO. Each of these four deals significantly surpasses the largest previous equity offering, from oil company Saudi Aramco, which raised $26bn in its 2019 stock market debut. If successful, they should help underpin confidence in both the scale and duration of AI capex. While we do not comment on individual IPOs prior to listing, we have been closely following and researching these companies for quite some time.

While the Trust’s pro-AI positioning remains intact, in the near term we have reduced the Fund’s beta – its sensitivity to market movements – via some profit-taking and increased NASDAQ put option¹exposure. This reflects heightened macro/geopolitical risk (Middle East; higher risk-free rates) as well as possible market digestion given the magnitude of expected near-term equity issuance (as described above).

Over the longer term, we continue to believe that both AI capability and demand remain underestimated. The Google I/O token data is perhaps the clearest illustration of this: the trajectory from trillions to quadrillions is not a linear extrapolation but a step-change driven by the compounding effects of better models, wider distribution and agentic workflows multiplying token consumption per task.

1. A put option grants the right to the owner to sell some/all of an underlying security at a specified price, on or before the option's expiration date