‘Enablers’ are the AI sweet spot for investors

The writer is global head of thematic research and sustainability research at Morgan Stanley

In the AI boom, the providers of infrastructure for the technology may capture some of the most durable returns for investors.

As AI capabilities improve at a lightning pace, constraints such as electricity supply are emerging as an important issue. Companies that can alleviate these will be in strong demand. This is a prospect that is already being recognised by investors, judging by the performance of a basket of stocks that span sectors across the AI infrastructure value chain, from utilities to information technology to hyperscalers.

The AI Enabler basket has outperformed the MSCI World Index by 50 per cent in the last three years and AI enablers in the US have captured 65 per cent of capital raised in seed to late-stage venture rounds in the first half of 2026.

This reflects confidence in the increasing demand for AI as the technology develops. Over the course of 2025, the percentage of questions that large language models could answer in a demanding test of intelligence and capability known as Humanity’s Last Exam has risen from 10 per cent to 50 per cent. And the pace of development will accelerate as LLM developers pursue a goal of what is called recursive self-improvement — the point at which models can develop autonomously.

In turn, AI adoption is accelerating. Usage of tokens — the units of data processed by models — rose 10-fold between January 2025 and April 2026, according to the fintech Ramp.

For many businesses, the value created by AI is already rapidly outstripping the expense. Yet, we are still in the early stages of AI diffusion, where workflows are redesigned, job tasks reallocated and capital spending redeployed. The productivity gains we are seeing today are likely a fraction of what can ultimately be achieved.

Our maths implies a massive shortage of computing power relative to probable future demand. Lack of access to the power grid may be the most serious bottleneck to ease this. We project that US data centre developers will face about a 40-gigawatt grid access shortfall through 2028, about half the power they need. For comparison, large US cities typically consume a few gigawatts of power daily.

The structural nature of the mismatch between the pace of AI infrastructure demand and the multi-decade timelines of grid investment is often under-appreciated. The lead times for power transformer deliveries in the US now average more than two years versus a pre-Covid norm of 12 to 16 weeks, and in some regions, utilities are quoting interconnection timelines of a decade or more.

There is also political pushback against data centre construction in some areas because of increases in power bills for other customers and concerns about the impact on air quality and water availability. If power developments are halted or delayed, this could exacerbate the energy shortfall for AI development as demand accelerates.

AI’s shift from the training of LLMs to “inference” usage — where models analyse data to make predictions — and agents carrying out tasks is also creating electricity demand patterns that existing grids were not designed to handle. Workloads are moving from batch processing to real-time execution, shifting the systemic challenge from energy availability to power demand volatility and response speed.

Recommended

We see solutions for these challenges emerging from AI enablers such as providers and developers of fuel cells, which convert an energy source into electricity. These can often be installed faster than new grid connections and typically provide reliable on-site power with limited water use and fewer harmful emissions. Such fuel cell “power islands” can be built off the grid, eliminating impacts on other power customers.

With power demand becoming more volatile, energy storage systems are another solution. Batteries are emerging as a system-level necessity, an “inventory of electricity” that does not replace generation but reduces power usage at times of peak demand, defers costly capital infrastructure and allows data centres to participate in grid stabilisation.

The enablers alone are unlikely to satisfy the rising demand for energy. But the scarcity of grid capacity is increasing the value of companies that can supply power efficiently while reducing the political cost of data centre expansion. They are the picks and shovels of this investment boom.