The AI cycle broadens beyond early adopters - December Trust Update

Key events

• Technology markets performed well in 2025, driven by a strong economy, interest rate cuts and growing investment in AI
• The Trust outperformed the index in 2025 due to its pro-AI positioning and avoidance of subsectors potentially at risk from AI
• We remain ‘AI maximalists’ and our outlook for the technology sector remains positive, supported by strong earnings and ongoing AI innovation
  

Market review

Global equity markets declined in December, with the MSCI All Country World Net Total Return Index giving up -0.6%. US equities followed suit, with the S&P 500 Index down -1.5%, however European markets were strong with the DJ Euro Stoxx 600 Index up 2.3% (all returns are stated in sterling terms).

During the year, the MSCI All Country World Net Total Return Index gained 13.9%, the S&P 500 9.7% while the DJ Euro Stoxx 600 significantly outperformed, up 27.3%. The trade-weighted US dollar fell 9.4%, its worst year since 2017, weakening against every other G10 currency as the Federal Reserve (Fed) pivoted from a ‘higher for longer’ narrative towards delivering multiple interest rate cuts in the second half. Meanwhile, gold returned 65%, its best year since 1979, in part due to dollar weakness itself, but also fears over government spending and debt levels.

2025 was defined by solid global growth, central bank rate cuts and growing optimism over the potential tailwinds from AI and upside to AI capital expenditure (capex). The market's upward trajectory was punctuated by volatility around DeepSeek in late January and the early April Liberation Day tariff uncertainty, followed by a strong relief rally as reciprocal tariffs were paused.

Market conditions stabilised mid-year as Middle East tensions eased, inflation moderated and labour markets cooled, prompting the Fed to resume rate cuts. The late summer and autumn were marked by fluctuating rate expectations, a prolonged US government shutdown, pockets of credit stress and renewed tariff threats. Momentum stalled a little in the fourth quarter although equities held up, supported by resilient earnings, easing inflation, supportive monetary policy and continued strength in AI-related capex. Encouragingly, most of the S&P 500 returns were driven by earnings per share¹ growth, while the forward price to earnings² multiple although high versus history ended the year at 22 times, only marginally above where it began at 21.5 times.

Economic data released in December suggested a continued, orderly cooling in the US labour market, further easing inflationary pressure. The headline Consumer Price Index rose 2.7% year on year (y/y) in December, below expectations of 3.1% and the lowest level since July.

A more benign inflation backdrop reinforced the case for easier monetary policy. The Fed delivered a further 25 basis points³ (bps) rate cut in December, lowering the federal funds target range to 3.50% to 3.75% and taking borrowing costs to their lowest level since 2022. The decision was not unanimous, however, highlighting ongoing debate within the Committee.

Looking ahead, political and policy uncertainty remain elevated. President Trump indicated he will announce his nominee for the next Fed Chair in January, raising potential questions over the future independence of monetary policy. In parallel, the Supreme Court is expected to rule on the constitutionality of Trump's tariff regime, although the administration has indicated contingency plans to reimpose tariffs should their current approach be struck down.

Yale economists have derived scaling laws for economic productivity which suggest that each year of AI model progress reduces task time by 8%, which implies continued model scaling could boost US productivity by 20% over the next decade.

Our base case remains that despite macro challenges and political gyrations, the US economy will not be sufficiently weak to derail the AI story. The broad economic expansion should continue as “the consumer is resilient, spending is strong and delinquency rates are actually coming in below expectations” (JP Morgan Q3 2025 earnings call). The US economy should also be supported by a positive fiscal impulse from the One Big Beautiful Bill Act and the potential for further policy support as the US midterm elections approach in November. A reduced drag from tariffs, further deregulatory initiatives and tax cuts or rebate cheques should also help. However, the K-shaped economy reflects strong wealth effects and consumer spending for upper income groups while less affluent groups, and the companies that serve them, may fare less well.

Disinflationary growth tailwinds from AI are also building. Productivity growth has ticked up post-Covid, running at c2% per year up from c1.5% and reaching 4.9% in the third quarter of 2025. If the proliferation of AI across the economy begins to take hold, potential real GDP should be lifted above prior assumptions and could allow growth above 2.5% without triggering inflation. Yale economists have derived scaling laws for economic productivity which suggest that each year of AI model progress reduces task time by 8%, which implies continued model scaling could boost US productivity by 20% over the next decade.

Technology review

During December, the technology sector modestly underperformed the broader market, the Dow Jones Global Technology Net Total Return Index declining -1.0%. The Philadelphia Semiconductor Index declined -0.7% while the NASDAQ Internet Index and iShares Software fell -1.9% and -0.1% respectively (all figures in sterling terms).

During the month, NVIDIA announced a $20bn non-exclusive deal to license Grok's Language Processing Unit (LPU) technology. The AI chip startup was founded by Jonathan Ross, part of the original Google Tensor Processing Unit design team, and as part of the deal Ross and other Grok staff will join NVIDIA. Adding LPU technology to NVIDIA's roadmap enables it to address more of the inferencing total addressable market, in areas where Graphics Processing Units (GPU) are less efficient.

December also saw the final earnings reports of the year. In the semiconductor sector, Broadcom beat expectations and raised guidance as AI semiconductor revenue grew 76% y/y and the company guided to 100% year-on-year growth for the next quarter. Broadcom's fourth processing unit customers more than doubled their order for AI racks to $21bn from $10bn last quarter. However, the stock was weak as selling full server racks will be a headwind to gross margins while the backlog number was below some bullish investor expectations.

Semiconductor memory maker Micron Technology (Micron) reported a strong quarter, guiding revenue $4.3bn above the street's $14.4bn estimate, a higher dollar amount than NVIDIA in its breakout May 2023 quarter. 2026 price and volume agreements for its high bandwidth memory (HBM) has now been contracted and the company pulled forward its $100bn HBM total addressable market estimate to 2028 from 2030. Supply is extremely limited across both HBM and commodity DRAM/NAND⁴ due to a lack of industry clean room space. This should support a record pricing environment through 2026 and into 2027, reflected in Micron's gross margin guidance of +68% versus consensus expectations of +56%.

Similar to Micron, AI networking leader Credo Technology Group Holding (Credo) delivered a robust quarter, beating revenue expectations by +14% and gross margin by +2.7%. However, its financial third quarter guidance was a staggering $100m, or +37% above consensus, driven by existing and new customers ramping Active Electrical Cables. The stock sold off, despite strong results, on profit-taking and concerns about next generation server content opportunities.

Ciena noted an expansion of the revenue opportunity with its first hyperscaler customer as well as wins with two additional customers for scale-across applications, meaning connecting data centres over significant distances.

Ciena's 4Q25 results were reassuring as order acceleration in H2 gives the company improved visibility on revenue next year. As such, the company raised revenue guidance for its next fiscal year to +24%, up from +17% last quarter, and $5bn of the $6bn revenue guide is already in the backlog so it appears somewhat conservative. In addition, Ciena noted an expansion of the revenue opportunity with its first hyperscaler customer as well as wins with two additional customers for scale-across applications, meaning connecting data centres over significant distances.

The software sector underperformed in 2025 with the IGV (iShares Expanded Tech-Software Sector ETF) down -1.8% in sterling terms. Fears of AI disruption, the commoditisation of code and threat to seat-based business models have raised terminal value questions. Software valuations are now below 2013-18 averages. Architectural and business model disruptions often spark mergers and acquisitions (M&A) for incumbents, and this month ServiceNow* announced the $7.8bn acquisition of Armis, by far the largest deal in the company's history and a push into an adjacent cybersecurity market. The company reaffirmed its guidance, but the story has become more complex and investors are concerned about a potential slowdown in the core business.

MongoDB delivered another good quarter as its Atlas cloud offering continued to accelerate to +30% y/y. This brought upside across margins, earnings per share and free cashflow⁵. Guidance was also very strong, +7% above street expectations. New customer additions remained strong and a sharper focus on direct customer engagement is driving enterprise growth.

Peer Snowflake also beat expectations, albeit by a smaller amount than the previous quarter. Product revenue was +29% y/y, a +1.5 percentage point deceleration from the previous quarter. Q4 product revenue was guided to +27% y/y. Despite a disappointing gross margin guide for Q4, other key metrics remained healthy and AI revenue is now at a more than $100m run rate.

Pure Storage posted solid results with revenue ahead of expectations accelerating to +16% y/y. Guidance was also raised and gross margins were particularly strong at +76%. The stock sold off after its results were announced given the company's large Meta Platforms (Meta) order has now been delivered, cautioning it will need to increase spending to address new opportunities.

Outlook

The market remained volatile into year-end as investors weighed rising AI capex and financing concerns, compounded by perceptions that OpenAI's models have recently lagged peers such as Anthropic's Claude Opus 4.5 and Google's Gemini 3, alongside the emergence of highly competitive models from China. In our view, such fluctuations are to be expected during the early stages of what we believe will be a durable and intensely competitive race for frontier model leadership, not least because AI adoption and growth continue unabated. Three years ago, fewer than one million people used AI which has grown to well over one billion active users, a feat that took decades during the internet era. ChatGPT alone hit more than 800 million weekly active users in mid-2025.

2026 will be a pivotal year for AI and OpenAI. Model performance will continue to improve as new Large Language Models (LLMs) trained on larger, more powerful compute clusters are released. Everyone is waiting keenly for the next major OpenAI and xAI models, expected in 1Q26, which will be the first trained on NVIDIA's Blackwell architecture. Alphabet's Gemini 3 model success gives confidence that pre-training scaling laws remain intact and that larger coherent compute clusters will bring more performant models. Benchmarks will be beaten, new approaches progressed and we remain hopeful that AI capabilities will continue to positively surprise. This should embolden leading AI labs to invest even more aggressively in the next generation of models and continue to support AI capex.

The speed of AI revenue growth so far has been extraordinary: OpenAI reached $10bn annual recurring revenue (ARR) less than three years after the launch of ChatGPT and Anthropic is on course to do the same in four years. This compares to eight years at Google and 10 years at Meta. OpenAI reportedly increased its revenue projections and now expects $100bn in 2028, up from $86bn previously. Anthropic is projecting $40bn revenue in 2027 and $70bn in 2028. While the market is treating these forecasts with healthy scepticism, the pace of current AI demand growth is unparalleled.

Perhaps most exciting is the potential for agentic AI to take off in 2026 as AI agents complete tasks previously the domain of humans, from booking a hotel to chasing an unpaid invoice. Agents should also benefit from a more mature tooling ecosystem and as they start to execute more complex work, demand for inference tokens should grow meaningfully. Work will increasingly be unconstrained by the number of humans available or hours in the day as agents work autonomously on longer and ever more sophisticated tasks, with potential for significant productivity benefits.

Strong demand for AI infrastructure is likely to persist as better models find more valuable and sophisticated use cases.

We expect 2026 will see a flourishing of AI use cases across every sector although the distribution will be extremely uneven between companies that can deliver and compound value by adopting AI and less successful peers that are left behind. Goldman Sachs estimates a 10-30% earnings per share boost from full AI adoption to most S&P 500 sectors.

Strong demand for AI infrastructure is likely to persist as better models find more valuable and sophisticated use cases. Infrastructure to train and serve AI models will remain supply constrained and with accelerating demand we expect leading labs and hyperscalers to remain compute-constrained through the year. There are already signs of this in hyperscaler backlogs, with $1.2trn of backlog combined across AWS, Azure and Google Cloud, up from c$660bn at the end of 2024, and commentary that AI demand is accelerating at a faster pace than capacity can be added.

We agree with Goldman Sachs' assessment that “the AI spending boom does not look particularly large when appropriately benchmarked against investment cycles associated with other general purpose technologies”. Bank of America expects big-five hyperscaler capex to rise to $612bn in 2027 while combined operating cashflows to help cover this spending could reach $1.1trn by 2029. For context, Gartner forecasts total worldwide IT spending to grow 9.8% in 2026 to pass $6trn. Near term, data centre financing needs for 2026 can be funded from hyperscaler cashflows and investment grade bond markets: Alphabet, Microsoft, Amazon and Meta have $700bn of additional capacity based on 1x net debt to 2026 EBITDA and $1.45trn based on 2 times. Net debt issuance in the tech sector was $145bn in 2025 and only represents 6% of notional stock outstanding versus 12% for telcos in 2000.

Valuations are elevated but appear reasonable and certainly not bubble-like with the S&P IT sector trading at circa 26 times forward earnings, 1.2 times the S&P 500, and NASDAQ at c25 times, in line with its seven-year average. Instead, most of the performance in recent years has been driven by strong fundamental growth rather than higher market valuations. Investors also appear to be highly sceptical about the return on AI investments and the durability of the AI trade, which is hard to reconcile with an irrational market bubble. Meanwhile OpenAI has reportedly held preliminary discussions to raise as much as $100bn at a valuation of $750bn.

Unlike previous booms, Initial Public Offering markets have been subdued since 2021's enormous issuance, but we expect this will change. It is quite possible one of the leading AI labs will go public towards the end of next year, due to the amount of capital required to train better models and, perhaps more importantly, run revenue-generating inference on them. More likely, however, there are several other companies at more than $1bn ARR which are credible candidates for a public listing and may also encourage the next generation of AI-native companies to come to market sooner, especially since many are scaling to more than $200m ARR faster than ever seen before.

Our constructive outlook on the tech sector in 2026 is primarily driven by our conviction in the AI story itself and is underpinned by strong earnings growth and several supportive impulses into next year: monetary policy, fiscal policy, fund flows and AI investment which should help the market climb the ‘wall of worry’.

Bull markets do not typically end when the Fed is cutting rates, earnings are growing double digits and the world is breaking ground on the infrastructure for a transformative new technology. With AI advancing at extraordinary speed, we expect 2026 to be the year when the capabilities of these models become unmistakable and the impact of AI increasingly difficult for investors to ignore. However, volatility should be expected as a normal feature of new technology cycles.

*not held

1. A company’s profitability expressed on a per share basis and calculated by dividing the company’s annual earnings after tax by the weighted average number of shares in issue.

2. A way to estimate the future earnings potential of a particular company or investment trust. It is calculated by taking the current share price and dividing it by the earnings per share (EPS).

3. A basis point is a common unit of measure for interest rates and other percentages in finance; one basis point equals 0.01%

4. DRAM is fast, short-term working memory for a device, while NAND is slower, long-term storage that keeps data even when the power is off.

5. The amount of cash a company has after it has covered its capital expenditures