Artificial intelligence is often described as living in the cloud. In reality, it operates from vast data centres filled with high powered servers that consume extraordinary amounts of electricity. The world is now accelerating investment in these facilities at a pace rarely seen in modern infrastructure cycles.
Technology giants including Oracle Corporation and Amazon are committing hundreds of billions of dollars to expand data centre capacity. Globally, AI related infrastructure investment is projected to reach into the trillions of dollars over the next decade. These projects are essential to support the computational intensity of advanced AI systems.
The scale of cost
The numbers behind this shift are significant.
Hyperscale data centres can cost between $10 billion and $20 billion to deliver once land, construction, grid connections, high voltage systems, cooling infrastructure and specialist fit out are included.
A 100 MW facility can consume as much electricity as a small town. At typical US industrial electricity rates, that level of demand can translate to more than $6 million per month in power costs alone.
Capital carries weight as well. Financing a $15 billion project at around 6 percent interest would equate to approximately $75 million per month in capital carrying costs before operating expenses are even considered.
Taken together, a large AI data centre can face combined power and financing exposure comfortably exceeding $80 million per month once fully built and funded.
Those figures explain why investors are watching the sector closely.
The investment debate
The infrastructure is real, the demand is growing, and many facilities are profitable. The question being asked in financial markets is whether revenue growth from AI services will consistently justify the scale and speed of capital deployment now underway.
When commitments reach tens or hundreds of billions of dollars, timing matters. Companies are constructing large scale digital infrastructure in anticipation of sustained growth in usage and pricing. The long-term direction may prove correct, yet the upfront exposure remains substantial.
AI may feel intangible. The balance sheet impact is anything but.
Why this matters for property and risk
Data centres differ markedly from conventional commercial buildings. Their electrical systems are bespoke, highly specified and extremely expensive to replace. As demand increases, complexity rises and reinstatement costs follow.
Greater specialisation leaves less room for error in insurance valuations. When declared values do not reflect current construction and infrastructure costs, exposure develops quietly in the background.
In a market committing billions of dollars to highly technical assets, accuracy is essential.
