The AI build is inside British inflation now, and the Bank has started listing the ways in.

The Bank of England's Agents publish a summary of what British firms tell them. This month's carries a box on artificial intelligence, and most of the attention it gets, including in this edition, is on the part about jobs. The other half is about prices, and it is the first time a British institution has set out the AI build as an inflation input rather than as an investment story.

Start with the constraint list, which is unusually specific.

The Agents report that strong demand for cloud computing and AI services is driving significant investment in data centres, and that expansion is held back by limited electricity capacity, lengthy grid connection processes, planning and land-use constraints, and shortages of key inputs including semiconductors, transformers and skilled labour.

Every item on that list is a physical thing in a particular place. A transformer is a large object of steel and copper with an order book measured in years. A grid connection is a queue. Planning is a committee. None of it is software, and none of it scales the way software scales.

The consequence, in the Bank's words, is cost pressure across energy infrastructure, construction, IT hardware and associated operating expenses, and longer delivery times.

Then the part that reaches the till. The Agents report more mentions of higher imported inflation for investment goods, especially technology such as servers, on higher memory chip prices. They also report rising prices for consumer electronics including mobile phones, laptops and games consoles, reflecting higher component costs.

That is a chain worth following to the end. Companies elsewhere are buying enormous quantities of memory to train and run models. Memory gets dearer. British firms pay more for servers, and British households pay more for laptops. The bill for a data centre that may not be in this country arrives in a shop that is.

The Bank records the other direction too, and it should be said plainly, because this is not one-sided. The same box says productivity improvements from AI are helping some firms cut costs and labour requirements, lowering unit costs and increasing pressure to reduce prices for routine tasks that can be partly or fully automated. It then says those savings are offset in part by rising spending on software, cloud services and AI licences, and that concentration among large technology providers is pushing costs up in parts of the value chain.

So the technology now sits on both sides of the inflation ledger, and the Bank is candid that this makes its job harder rather than easier.

The structural point is the one about electricity and the queue. The argument about datacentre power is not new here. What has changed is that it has stopped being only about whether the sheds get built, and started being about what the build costs everyone else while it happens. A grid connection queue is not a technology problem. It is a planning and infrastructure problem, which is the kind a state can choose to fix or choose to leave.

One thing the box does not do, and it is worth naming rather than walking past. It does not quantify any of this. There is no figure for how much of British inflation is attributable to AI infrastructure, in either direction. The Bank is describing a mechanism, not measuring one. Anybody who gives you the number this week has made it up.