The next decade of investing: AI at the centre, everything else in tow
17th October 2025
The shape of the next cycle
If you zoom out from the hype, what is happening now looks familiar. A breakthrough technology moves from the lab to the real economy and drags an entire supply chain behind it. In the late 2020s and into the 2030s, artificial intelligence is that engine. What makes this cycle different is the breadth of what it pulls along: data centres, chips, optics, cooling, electricity, grids and the minerals underneath it all.
Independent analysts put eye-watering numbers on this. McKinsey’s latest estimate suggests the world will need roughly $6.7 trillion of data-centre build by 2030 to keep up with compute. The International Energy Agency projects data-centre electricity consumption will more than double by the end of the decade, with AI the main driver. In plain English, this is a multi-year build-out that touches steel in the ground, not just software in the cloud.
If you zoom out from the hype, what is happening now looks familiar. A breakthrough technology moves from the lab to the real economy and drags an entire supply chain behind it. In the late 2020s and into the 2030s, artificial intelligence is that engine. What makes this cycle different is the breadth of what it pulls along: data centres, chips, optics, cooling, electricity, grids and the minerals underneath it all.
Independent analysts put eye-watering numbers on this. McKinsey’s latest estimate suggests the world will need roughly $6.7 trillion of data-centre build by 2030 to keep up with compute. The International Energy Agency projects data-centre electricity consumption will more than double by the end of the decade, with AI the main driver. In plain English, this is a multi-year build-out that touches steel in the ground, not just software in the cloud.
Beyond the obvious names
Owning “AI” is not just a question of picking the cleverest model. It is semiconductors and memory to feed the model, optics to move data around at 800G and beyond, and the kit to keep ever-hotter racks cool and powered. It is also the power stations, transformers, cables and control systems that keep the whole thing humming. Hyperscale operators already run more than a thousand facilities globally and are still adding capacity. That spend has to land somewhere and, for investors, the more durable moats often sit in the picks-and-shovels.
Where the moats live
Some parts of the stack have unusually strong competitive positions. ASML remains the only supplier of EUV lithography tools, which the industry needs for the most advanced chips. High-bandwidth memory has become a structural bottleneck as model sizes grow, and leading players are investing heavily to keep up. In networking, the shift to 800G and then 1.6T optics underpins a multi-year upgrade inside the data hall. None of this is glamorous. All of it is necessary.
Power is the constraint
Electricity is the new oxygen. Data-centre power demand is rising quickly, and in some regions operators are exploring on-site generation because the grid cannot connect them fast enough. Over the decade, that implies sustained demand for generation, grid reinforcement, high-voltage equipment, and storage. It is also a reminder to keep portfolios global. Policy and permitting timelines are not the same in Texas, Dublin and Tokyo, and capital will flow to where it can get built.
Owning “AI” is not just a question of picking the cleverest model. It is semiconductors and memory to feed the model, optics to move data around at 800G and beyond, and the kit to keep ever-hotter racks cool and powered. It is also the power stations, transformers, cables and control systems that keep the whole thing humming. Hyperscale operators already run more than a thousand facilities globally and are still adding capacity. That spend has to land somewhere and, for investors, the more durable moats often sit in the picks-and-shovels.
Where the moats live
Some parts of the stack have unusually strong competitive positions. ASML remains the only supplier of EUV lithography tools, which the industry needs for the most advanced chips. High-bandwidth memory has become a structural bottleneck as model sizes grow, and leading players are investing heavily to keep up. In networking, the shift to 800G and then 1.6T optics underpins a multi-year upgrade inside the data hall. None of this is glamorous. All of it is necessary.
Power is the constraint
Electricity is the new oxygen. Data-centre power demand is rising quickly, and in some regions operators are exploring on-site generation because the grid cannot connect them fast enough. Over the decade, that implies sustained demand for generation, grid reinforcement, high-voltage equipment, and storage. It is also a reminder to keep portfolios global. Policy and permitting timelines are not the same in Texas, Dublin and Tokyo, and capital will flow to where it can get built.
The materials that make it possible
Copper really is the metal of electrification. Forecasts vary, but many credible scenarios show demand outstripping supply into the 2030s without accelerated investment and recycling. Rare earths matter too. They go into high-performance magnets that turn in wind turbines, EV motors and industrial kit, with a large share of the processing concentrated in China. Concentration is not the same as catastrophe, but it does mean occasional price spikes and policy noise are part of the journey. We price that risk in rather than pretending it does not exist.
How we own it for clients
At Brigantia we prefer a diversified, evidence-based approach. Our core remains broad and cost-effective. Around that, we tilt into the underlying infrastructure that AI needs to function: memory and lithography where moats are real, optics and interconnect where volumes are compounding, liquid cooling where physics leaves little alternative, and grid equipment where backlogs support pricing. We balance platform exposure with valuation discipline and we rebalance regularly. If 2026 brings a pause in headline capex, we prefer to be the investor harvesting volatility rather than the tourist chasing it.
The bottom line
AI is the headline, but the investment story of the next decade is the ecosystem that makes it possible. The companies that wire, cool, power and supply it are not optional extras. They are the compounding spine of the theme. Our job is to own that spine sensibly, keep costs honest, and stay patient while everyone else argues about which chatbot sounds cleverest.
Copper really is the metal of electrification. Forecasts vary, but many credible scenarios show demand outstripping supply into the 2030s without accelerated investment and recycling. Rare earths matter too. They go into high-performance magnets that turn in wind turbines, EV motors and industrial kit, with a large share of the processing concentrated in China. Concentration is not the same as catastrophe, but it does mean occasional price spikes and policy noise are part of the journey. We price that risk in rather than pretending it does not exist.
How we own it for clients
At Brigantia we prefer a diversified, evidence-based approach. Our core remains broad and cost-effective. Around that, we tilt into the underlying infrastructure that AI needs to function: memory and lithography where moats are real, optics and interconnect where volumes are compounding, liquid cooling where physics leaves little alternative, and grid equipment where backlogs support pricing. We balance platform exposure with valuation discipline and we rebalance regularly. If 2026 brings a pause in headline capex, we prefer to be the investor harvesting volatility rather than the tourist chasing it.
The bottom line
AI is the headline, but the investment story of the next decade is the ecosystem that makes it possible. The companies that wire, cool, power and supply it are not optional extras. They are the compounding spine of the theme. Our job is to own that spine sensibly, keep costs honest, and stay patient while everyone else argues about which chatbot sounds cleverest.