How investors can assess opportunities in AI boom

UBS Global Wealth Management has identified three layers to the AI market opportunity and they’re not all centred around chip-making giants such as Nvidia.

The potential of artificial intelligence has started a capex boom in data centres.

The launch of ChatGPT on November 30, 2022 was an inflection point for the broad adoption of artificial intelligence.

The term AI originated in 1955 at Dartmouth College so the concept is not new. However, ChatGPT was the first usable and accessible AI tool.

Generative AI uses vast amounts of data to identify patterns to generate content. Its ultimate benefit will be to improve productivity, particularly for the one billion knowledge workers worldwide.

The potential of AI has kicked off a capex boom in data centres. The range of activities or use cases that AI can address is growing. But to justify the capex expense, use cases have to materialise at scale. This creates opportunities for investors who understand the AI value chain.

UBS Global Wealth Management sees three layers to the AI market opportunity.

The first is the Enabling Layer. This consists of AI data centres, and the servers, chips and power that goes into them. A company may own their own data centre, or use cloud service providers like Amazon AWS, Microsoft Azure or Google GCP.

The key elements to a data centre are banks of servers and the general processing units, or GPUs, that go into the servers. We estimate GPUs account for 74 per cent of the cost of a server, and servers represent 57 per cent of the cost of a data centre.

Nvidia comprehensively dominates GPU market share. However, competitors are working on substitutes. Apart from Nvidia, the other big five US tech companies are developing custom AI chips.

Custom chips are application specific. Sometimes called accelerator chips, they have a specific functionality. Companieswho fabricate semiconductors and provide the equipment for chip manufacturing, such as laser lithography, are benefiting from data centre capex. Despite attractive industry structure, these businesses are subject to the volatile chip-manufacturing capex cycle, although AI is providing a strong tail wind at the present.

There is a growing realisation that enormous amounts of power will be required to run the planned AI centre build-out. We estimate the cost of power, and distribution and cooling systems, account for 31 per cent of data centre costs.

Companies that provide cabling, copper for the cabling, power supplies, power efficiency solutions, cooling and other equipment are experiencing sales strength. These “old-economy stocks”, often European, are seeing a renaissance. Likewise, clean energy companies that can provide carbon-neutral power are enjoying strong demand.

The second layer is the Intelligence Layer. This includes generative AI algorithms and large language models (LLM). This layer functions as the “brain”of AI. It uses the massive amounts of data and compute capacity in data centres to identify patterns and linkages. The models store the information in an organised and systematic format that applications can interrogate and deliver in a usable output.

An example of an LLM is ChatGPT, which was developed by OpenAI. Microsoft owns the right to 49 per cent of OpenAI’s profits. Google (Gemini) and Meta (Llama) are also developing LLMs. Pricing models are usage based or subscription and are differentiated by whether the LLM is used directly or through an application, the size of the model and the limited number of LLMs so far. Industry structure is favourable, with certain models dominant in different regions of the world. That could change and new competitors arrive. However, what will likely differentiate the models in the future is access to data, which gives the biggest six US tech firms a significant advantage.

The third layer is the Application Layer. Copilots are at the centre of applications given their ability to boost office productivity. The term “copilot” refers to tools embedded in workflow software. Examples are Microsoft’s aptly named CoPilot and Adobe’s Firefly.

The addition of copilots allows software companies to raise prices. Workflow management software is another segment we favour. This part of the value chain is at an early stage but growing at pace.

Examples of use cases include increasing the speed of computer coding, identifying novel drug compounds for specific illnesses, improving call centre outcomes for customer questions and the diagnosis of error alerts in industrial processes.

Advertising should also benefit through image and text creation, the use of chatbots to proactively meet customer needs, and through targeted, personalised content.

Finally, a potentially large market is embedding AI in internet-enabled devices such as smart phones, PCs and home appliances. AI-enabled devices are referred to as AI edge computing devices. Specialised chips are required that don’t need the compute power of data centre GPUs. They have specific functionality and can process data locally. This is not only an opportunity for chip designers, but also the dominant US, Korean and Chinese smart phone producers to increase prices and encourage updating to the latest phone.

Calculating the size of the AI opportunity with certainty is impossible. It will depend on the productivity gains that can be achieved. The International Labour Organisation estimates if AI can improve productivity of each knowledge worker by 15 per cent, its value is around $US4.4 trillion ($6.7 trillion) annually, larger than the mobile phone and PC market combined.

But a word of caution, technology valuations have been pushed up. It is a volatile sector and highly competitive. There will be successes and failures. However, the above framework is one way in helping long-term investors identify the opportunities in this potentially world-changing technology.

Andrew McAuley is managing director of UBS Global Wealth Management.