Artificial intelligence has moved from experiment to infrastructure. The question for investors and executives in 2026 is no longer whether companies will adopt AI, but who will pay for the physical layer that makes it run — chips, data centres, power and the networks connecting them.
The shift from software to physical assets
For a decade, technology investment meant software: low capital expenditure, fast deployment, high margins. AI changes that equation. Training and serving large models demand specialised accelerators, enormous amounts of electricity and cooling capacity that traditional data centres were never designed to provide. Capital is flowing into concrete, copper and power contracts as much as into code.
Where the money is going
- Compute: demand for GPUs and custom silicon remains the bottleneck. Suppliers are sold out for quarters in advance, and buyers are signing multi-year commitments to secure capacity.
- Data centres: new campuses are being built outside traditional hubs, driven by land availability and access to cheap, reliable energy.
- Energy: power has become the constraint. Utilities, nuclear restart projects and long-term renewable contracts are now part of the AI supply chain.
- Networking and memory: faster interconnects and high-bandwidth memory are as critical as the processors themselves.
How the build-out is financed
The scale of spending has pushed financing beyond operating cash flow. Hyperscalers still fund much of it internally, but a growing share comes from corporate debt, infrastructure funds and private credit. That introduces a new dynamic: AI capacity is increasingly financed like infrastructure — with long amortisation schedules and returns that depend on sustained utilisation.
For lenders and investors, the key metric is no longer growth alone, but utilisation and pricing power over the life of the asset. A data centre built for a demand curve that never arrives is an expensive stranded asset.
Risks worth watching
- Overcapacity: if model efficiency improves faster than demand grows, utilisation could fall short of projections.
- Energy prices: electricity costs can erode margins in energy-intensive facilities.
- Concentration: a handful of buyers account for a large share of demand, creating dependency risk.
- Regulation: energy use, data residency and export controls can reshape the economics overnight.
What it means for businesses
Companies outside the technology sector do not need to build infrastructure, but they do need a sourcing strategy. That means understanding where their AI workloads run, negotiating contracts that account for price volatility, and avoiding lock-in to a single provider. The organisations that treat AI capacity as a strategic input — rather than an IT line item — will be better positioned when supply tightens.
The bottom line
AI infrastructure is becoming a capital-intensive industry in its own right, with its own financing structures, risk profile and cycle. The winners will not simply be those who spend the most, but those who match capacity to real demand and secure energy and supply chains early.
