Data centres are no longer just a technology story. They are becoming a power, planning, and rates story, which is a polite way of saying that physics has entered the spreadsheet.

Demand is structural, capacity is not

UK data centre demand still looks structurally tight rather than cyclical. Three forces sit behind it: hyperscale cloud expansion, AI related workloads, and enterprise colocation that has not gone away just because every board deck now contains the letters AI.

The key point is that demand is not the scarce item. Power, planning permission, and actual execution are. That changes the investment question from who can sign leases to who can deliver megawatts without pretending the grid is an infinite vending machine.

London remains the anchor market because it has fibre density, mature cloud ecosystems, and availability zones. That mix gives it a network advantage. It also pushes the market toward land and power constraints, which is where elegant growth stories become messy engineering problems.

Capital is still interested, but it is becoming more selective. That is a healthy change. A data centre deal that ignores grid access is not a technology investment. It is a real estate bet wearing a hoodie.

Power is becoming the credit check

The eastern US heat dome gave a live demonstration of the constraint. More than 150,000 households lost power as temperatures approached 40C, and electricity prices jumped as utility grids came under stress.

Data centres are attractive tenants because they want huge amounts of electricity and pay on long contracts. That is also why they make local politics nervous. A constant server load is useful for cash flow, but it is not invisible to the grid.

The better operators will underwrite substations, generation, grid connection queues, and local consent with the same seriousness they apply to lease terms. The weaker ones will present demand charts and hope someone else handles electrons. Hope is not infrastructure.

This is why power is becoming a credit check. Cheap capital helped the sector grow. Reliable power decides who can actually open the doors.

Planning risk now has a ticker feel

Blackstone’s QTS ended a Virginia data centre project after protests. That matters because Northern Virginia is the old capital of data centres, not a fringe market where problems can be dismissed as local noise.

Planning opposition is no longer a soft nuisance. It can kill projects, slow delivery, and change the economics of an entire campus. When the market needs new capacity fast, a zoning fight can become a revenue event.

Investors usually model lease rates, cap rates, debt costs, and vacancy. They now need a sharper model for political friction. A community that sees higher power demand, more transmission equipment, and limited direct employment may not care that AI inference needs another hall of servers.

The lesson is blunt. Capacity growth will depend less on the press release and more on the boring map: substations, water, land, roads, permits, and local patience.

Soft jobs data keeps markets cheerful

The US economy added 57,000 jobs in June, below an estimate of 110,000, after a 3 month streak of stronger data. Stocks and bonds reacted well because weak labor data can reduce pressure on the Federal Reserve to keep policy tight.

This is not a morality tale. It is duration math. Lower expected rates support assets where a lot of the value sits in future cash flows, and that includes large parts of the technology and infrastructure trade.

Data centre equities and AI linked capital spending both benefit when rates fall, because cash flows are distant while capital spending arrives immediately. Cheaper debt can make a marginal project look less marginal. That does not make the power grid bigger.

The awkward part is that markets want soft data, but not too soft. Weak enough to calm the Fed is pleasant. Weak enough to damage enterprise demand is less pleasant. This is the usual financial dance, performed in expensive shoes.

Energy risk has moved, not vanished

Gasoline prices have fallen for seven consecutive weeks by an average of 9.6 cents as crude eased after the Strait of Hormuz reopened. That helps consumers at the margin and takes some heat out of inflation expectations.

Canada’s plan for a new oil pipeline to supply Asia with 1 million barrels per day points in the other direction. Energy security is still being redesigned around trade risk, export routes, and reduced dependence on single corridors.

For data centres, the lesson is similar. Energy is not background scenery. It is a core input. No model survives if fuel, power, or grid access is assumed to be a permanent constant.

The market can price falling gasoline and easier rates in one column. It still has to price power scarcity, local opposition, and grid investment in the next one.

What to watch

First, watch UK grid connection queues and planning decisions, not just demand forecasts. London can remain strategically important and still become harder to build in.

Second, watch whether Virginia style resistance spreads to other data centre clusters. One cancelled project is a story. Several cancelled projects become a capacity constraint.

Third, watch the Federal Reserve reaction to labor data. Lower rates can support the trade, but they do not solve power, land, or permit scarcity. Finance can discount the future. It cannot plug in a server rack by itself.