The rapid development of generative artificial intelligence (AI) tools has been a central concern of businesses and markets over the past three years. Demonstrated by the dominance of the ‘Magnificent Seven’, valuations reflect the widespread expectation that this will be an ongoing phenomenon.
Much of the discussion about AI technologies has focused on the hypothetical. Speculation is at fever pitch regarding the potential for AI to transform tasks as mundane as grocery shopping and as advanced as surgery. Yet, few tools have become mainstays of daily life.
Up until now, activity in the AI sector has focused on training models and machine learning. This has aimed to produce systems sophisticated enough to support complex requests and applications.
The significant transition that is only just beginning is for the technology to rotate from training to usage.
It is here that the outlook for the sector becomes murkier. The ‘winners’ in terms of both models themselves and their makers have not been determined.
However, capital expenditure has been – and continues to be – significantly ramped up in a bid to capture future markets. Companies such as Meta and Microsoft have staggering sums of money to spend, supported by their prodigious earnings growth from their existing businesses and cash-rich balance sheets.
In the first quarter of 2024, Microsoft’s capex rose 79% to $14bn, while Meta plans to increase its capex in 2024 overall by at least $5bn.
The question that most readily comes to our minds is: where is all that money going?
It is here that a more certain set of winners lies. AI may itself have a dramatic impact on our lives, society and physical world. The resources needed to support its evolution, though, are already experiencing – and creating – such an effect.
The expansion of AI requires significant infrastructure support – and that needs to be firmly in place before any transition can happen.
With this in mind, tangential operations such as data centres have become increasingly central to the AI revolution. Estimates suggest that AI-supporting data centres use two and a half times more energy than legacy data centres. With extra processing comes extra heat produced. As such, cooling technology has become a central requirement in the AI transition, accounting for up to 40% of that energy use.
Another example is memory. Devices will need to have additional memory capabilities to support the most challenging AI applications. As these functions come online, that hardware will need to be in place already.
As Nvidia’s meteoric share price rise demonstrates, semiconductors are a crucial component of this transition. This opportunity is investable via semiconductor producers themselves, through to the producers of the equipment used to manufacture the chips. The AI investing pool runs deep.
Indeed, down to the cabling required to carry so much additional data both within the data centres and around the world, a global supply chain is establishing itself.
Another question to ask ourselves though is: who will be the eventual winners among the AI service providers?
Ultimately, this is hard to forecast. One thing that seems certain is that our technology platform is likely to expand dramatically. Cisco has predicted that 500 billion devices will be connected by 2030, a rise from 13 billion in 2013.
This ultimately circles back to having the resources in place to support any applications that do come online. The demand for energy has already surged with the use of large language models. This exacerbated the increase in electricity use that was already coming from electric vehicles, heat pumps and other areas of the green transition.
The tension that this represents was most clearly reflected when Amazon purchased a data centre with its own nuclear power source earlier in the year.
With Western societies unused to energy needs increasing, this presents a significant challenge. However, among established energy producers, it also presents an opportunity.
Ben Lofthouse is the fund manager of Henderson International Income Trust. The views expressed above should not be taken as investment advice.