Active Thinking: AI's Android Moment

Artificial intelligence’s (AI) warp-speed evolution could reshape entire industries, creating unprecedented opportunities for investors.

10 March 2025

• As AI technologies become truly accessible and cost-effective, we are witnessing a transformative shift akin to the Industrial Revolution, one that will spur a new ecosystem of applications and services for end users.
• Josh Sambrook-Smith, Investment Manager, and Wendy Chen, Senior Investment Analyst on the Global Equities team, discuss how investors can capitalise on emerging opportunities in the shapeshifting AI landscape.

Since integrated circuits and silicon microchips were independently developed by Jack Kilby of Texas Instruments and Robert Noyce of Fairchild Semiconductor in the late 1950s, the industry’s mission has been to achieve greater power and performance. Gordon Moore, co-founder of Intel, famously observed as early as 1965 that the number of transistors on a microchip doubles every two years, while the cost of computing power roughly halves over the same period.

Progress since then has largely borne out Moore’s Law, as capabilities have developed from a few dozen transistors per chip in the sixties to tens of billions today, and processing power quantum leaps from then-thousands to trillions of operations per second today. However, while the focus over the last 10 years or so has been on shrinking chips to squeeze in ever-more circuits using ASML’s lithography technology to increase processing power and improve energy efficiency, this route to progress is increasingly reaching its physical limits. Shrinking, as a route to performance improvement, is getting harder, and much more expensive.

Splitting, not hitting, the wall

As you would expect from the ultimate innovation-driven sector, the semiconductor industry has been investing heavily in new technologies, microchip architecture and techniques to overcome these constraints. And in our view, these advances could in turn create enormous commercial opportunities for designer and equipment vendors across the sector, and for those investors nimble enough to adapt to the pace of change.

Custom chips

During the early 2000s, producers like Intel and AMD rode the wave of demand for increasingly powerful general-purpose microchips, driven by growth in the PC market and the rapid rise of the internet. More recently, Nvidia has come to the fore with GPUs (graphic processing units) that have morphed from gaming applications to professional visualisation, data centres and AI/machine learning applications. However, notwithstanding growing demand for general purpose, so-called merchant chips, we are witnessing an even bigger surge in demand for customised chips, tailored for customer application-specific needs, bringing benefits in terms of cost and performance.

Custom Chips (ASICs)

Opportunity #1

With demand growth for custom chips outstripping that for their general-purpose equivalents, we believe companies such as Broadcom, the largest custom chip designer in the world, already supplying customers like Google and ByteDance – and soon to add Apple and OpenAI to their client list – are well-positioned to capitalise on this market. Broadcom management have projected a major revenue opportunity in AI, seeing a serviceable addressable market (SAM) of up to USD 90 billion from just three customers – Google, Meta and ByteDance – by 2027.¹

Transistor innovations

Every decade or so has witnessed a seismic shift in the architecture of transistors – the devices that control the flow of electrons – on microchips. The latest phase – the switch to Nanosheet (designs which improve the electrical performance and efficiency of a transistor) architectures – is now gaining real momentum, with the changing shape of transistor structures creating massive opportunities for semi-cap equipment vendors and companies behind the advanced software enabling the new chip designs. While we believe these chip architecture opportunities would arise independently of the AI revolution, improved hardware performance is an enabler of more powerful AI, so we see a potential double upside of these advances as tens, if not hundreds, of billions of dollars are invested in new-generation manufacturing equipment and the software/design applications driving them.

Pack IT in – opportunities in advanced packaging

As chips become smaller and more complex, the requirement grows for advanced packaging techniques, stacking more and more chips of sub-three nanometres in size into a single package. In our view, this presents a huge market opportunity for equipment vendors such as BE Semiconductor Industries specialising in these technologies.

As well as the hardware – the physical tools behind the AI revolution – we are also constantly analysing what end-customers are using AI for, and whether/how it is boosting their bottom line. Most importantly of all, we are considering where the most attractive investment opportunities are.

Established (and emerging) AI use cases

AI text generation took a major leap forward in 2022 with the launch of ChatGPT. Since then, rivals such as Google DeepMind’s Gemini have joined the fray with knowledge and creative writing services. Microsoft’s CoPilot and GitHub offer software code generators and marketing support services, while Midjourney and Adobe are among those offering image and video generation services to customers such as artists and marketers. These services are already being commercialised, both on a pay-per-image or subscription basis, and the rapidly evolving capabilities, especially in video generation, are creating disruption and investment opportunities across the media industry.

Established Use Cases

Already being commercialised

Personalised AI assistants are already a ‘thing’, with Agentic AI’s conversational personal assistant acting on verbal instructions like “Make a dinner reservation for tomorrow evening at my favourite restaurant.”

Pharma grows: Why AI could be a game-changer in health and agri applications

Companies like Google’s AlphaFold have developed AI technology capable of generating proteins with the potential to combat specific diseases in crops. Further along the emerging use case line lies the potential to create proteins to fight diseases including cancer and neurodegenerative disorders such as Alzheimer’s, reducing suffering while also creating extraordinary investment opportunities in the healthcare sector.

Beyond pure hardware plays – the disruptive potential of DeepSeek, and others

Anyone guilty of lazy assumptions about the AI evolution being founded purely on ever-more processing power had a rude awakening at the end of January.

DeepSeek's AI model, apparently developed at a fraction of the cost of other leading models by a young team of tech prodigies and recent graduates from China’s most prestigious universities, raised questions about the necessity of high AI capital expenditure (capex), particularly as the model appeared to deliver similar results to established players. DeepSeek claimed to have spent only USD 5.5 million², a pittance in relative terms, to train their model on H800 chips that are much less powerful than the newer, muscular H100 GPUs favoured by the likes of OpenAI, a necessity ironically partly resulting from US technology export sanctions from the US CHIPS and Science Act of 2022. And to rub salt in the wounds, the Chinese company claimed it uses just a few thousands of these H800 chips to run its model across its 20 million daily active users (DAUs), a fraction of running costs of its more illustrious US peers.

DeepSeek: deep evolution, seek disruption

Similar performance, fraction of costs. How? Maximise existing tech to combat sanctions

The emergence of DeepSeek caused a valuation shakeout of some of AI’s established poster boys, such as Nvidia, reflecting the disruptive potential and the market's sensitivity to unexpected AI developments. Although share price setbacks in more familiar AI-related names were soon largely reversed on strong trading updates, especially from semiconductor names like ASML, as well as concerns about DeepSeek’s data security practices, the whole episode reminded investors how rapidly perceptions can change in the nascent AI sector.

Strategic implications in the rapidly changing world of AI

The AI investment landscape is evolving at breakneck pace, far faster than even tech investors during the dot.com boom in the early 2000s could have imagined. While established hardware names like Nvidia will remain core to the theme, for many investors AI exposure has been crowded around power-based infrastructure names. For example, only a week before the spotlight fell on DeepSeek, the US government announced the “Stargate” investment plan which will see USD 500 billion spent on AI capex over a four-year period. Moreover, just a week after DeepSeek gained attention from a global audience, the four major AI hyperscalers, aka Google, Meta, Amazon and Microsoft, collectively guided their 2025 capex to USD 320 billion, a remarkable 44% year-on-year increase that beat Wall Street’s expectation by 20%.³

Such developments evidenced our view that the capex theme still has some way to run, noting the observations of 19th century British economist William Jevons, who found that as technological improvements raise the efficiency of resource usage overall consumption can still rise, as, counterintuitively, the falling costs of a resource can lead to increased overall demand for the technology as uptake rises. While the resource in question for Jevons was coal during the Industrial Revolution of 1760 to 1840, the principle is equally applicable to computing power in AI’s present-day Industry 4.0 revolution.

End of AI computing capex story? Not so fast

Jevons Paradox suggests that cost efficiency helps to drive more demand

The AI applications party is just getting started

In our view, the lower cost of training AI models, as demonstrated by DeepSeek, should lead to increased usage and spending on AI technologies, both on the hardware and the software side. As Satya Nadella, CEO of Microsoft, recently noted, for AI to fulfil its true potential, the functionality needs to be more commoditised, echoing Jevons by comparing AI to the steam engine during the Industrial Revolution in the sense that making AI more accessible will boost uptake and subsequently productivity and economic growth.

In our view, the increasing adoption of open-source models, akin to the Android operating system that fostered the development of the app eco system on non-Apple mobile devices, will accelerate the development of democratised AI-applications, spreading out investment opportunities to the wider tech world as an entire AI ecosystem rapidly develops, serving evolving user demands on a scale that many investors are only now beginning to imagine.