Investors should not be focusing on which AI lab can develop the best large-language model but on which companies are supplying the underlying infrastructure for the whole AI revolution, according to Grégoire Kounowski, head of advisory at wealth management platform Norman K.
For nearly two years, markets have linked AI almost exclusively to semiconductors, cloud computing and the platforms built on top of them, but Kounowski reckons that association is now breaking down.
"The central question is no longer simply which companies will develop the best AI models, but which companies will have the necessary infrastructure to run them at scale," he said.
"As AI becomes more widespread, the demand for computing power continues to rise at a spectacular rate. Data centres are now one of the main drivers of growth in global electricity consumption."
He cited International Energy Agency projections that electricity demand from AI data centres could more than double by 2030 as evidence that investment opportunities are being created outside of the tech names at the heart of AI development.
A recent paper from Edmond de Rothschild's global investment research team also forecast that global data centre electricity consumption will double by 2030 to roughly 945 terawatt-hours, with AI-specific demand tripling over the same period. Some projections put total consumption above 1,200 terawatt-hours by 2035.
Kounowski said: "Whilst investors have so far focused on chip manufacturers and software companies, their attention is now turning to physical infrastructure: electricity generation and distribution, transport networks, industrial equipment, cooling systems, the acquisition of strategic land and the construction of the data centres themselves.
"Major technology groups are fully aware of this. Having previously devoted the bulk of their investment to computing capacity, they are now seeking to secure their energy supply for decades to come."
That shift can be seen in several sectors outside of tech, such as power generation and distribution. Data centre operators are signing long-term renewable supply contracts, with technology companies accounting for roughly 40% of all renewable electricity purchase agreements signed in 2025, according to Edmond de Rothschild.
They are also reviving nuclear capacity, shown through agreements between Constellation Energy and Vistra and hyperscale cloud providers, and backing small modular reactors (SMR): the pipeline of conditional agreements between data centre operators and SMR projects grew from around 25 gigawatts to 45 gigawatts in a single year. Where grid connections are too slow, some operators are turning to on-site gas generation, with GE Vernova adapting turbines to power sites directly.
Industrial and electrical equipment are also attractive, according to Edmond de Rothschild. An executive at Eaton said the industrial company's order backlog was equivalent to 11 years' worth of what it built in 2025, with data centre-related orders up roughly 200% in the fourth quarter of that year alone. Caterpillar expects to quadruple sales of generators used to supply data centre power by 2030.
Edmond de Rothschild estimates the investment required to electrify AI data centres could reach approximately $1,400bn by 2030, with Schneider Electric, ABB, Siemens, Legrand and Hitachi Energy named among the suppliers of switching equipment, inverters and transformers benefiting from extended delivery times and resulting pricing power.
A third opportunity is cooling. Server densification and the shift to liquid cooling have made thermal management critical, an area where Vertiv holds a leading position alongside Johnson Controls, Carrier, Trane and Modine, according to the bank's research.
"In construction and installation, players such as Quanta Services – whose order backlog reached a record high by the end of 2025 – and connectivity providers such as Corning in fibre optics round out the ecosystem," Edmond de Rothschild added.
"A notable strength of this sector lies in its relative resilience: the same equipment manufacturers benefit from the broader modernisation of electrical grids, which provides them with a certain degree of visibility even if specific demand from data centres were to moderate."
Norman K's Kounowski said this should feed into portfolio construction. He argued that the resulting equity market implications could be significant, given how concentrated stock market performance has been.
"Since 2024, stock market performance has been largely concentrated amongst a small number of technology companies," he said.
"However, the next phase of the cycle could extend to much more traditional sectors. Certain manufacturers, electrical equipment suppliers, engineering firms, energy producers and infrastructure operators could benefit indirectly from the massive spending on AI. This development is all the more interesting given that these companies often have more reasonable valuations than those seen in certain technology segments."
He drew a comparison with previous infrastructure build-outs, pointing out that the beneficiaries of past phases of electrification or the rollout of the internet were ultimately companies much wider than those developing the technology itself.
On positioning, Kounowski favours balance rather than rotation out of technology altogether. Exposure to technology leaders is justified thanks to their dominant position and their capacity for investment and innovation.
However, diversifying some exposure into infrastructure, energy, certain industrial sectors and strategic commodities gives investors the chance to capitalise on a long-term structural trend while reducing their reliance on the increasingly expensive technology sector.
"The market is gradually shifting from speculation on the promises of AI to an analysis of the concrete needs it generates," he said. "In this new phase, the question may no longer be simply who will develop the best AI, but who will provide the electricity, networks and infrastructure enabling the entire ecosystem to function."