The money buys sheds full of machines. Britain has a queue for the electricity to run them.

AI spending is not a number on a screen. It is concrete, steel, servers, cooling and a very large electricity cable.

You can watch it appear on a balance sheet. Alphabet's property and equipment, after depreciation, stood at $246.6bn at the end of December. Six months later it stood at $321.2bn. That is roughly $75bn of physical kit added to one company's estate in half a year, and it has to sit somewhere and it has to be plugged into something.

Which brings us to Britain, where the plugging in is the problem.

Ofgem's figures are the ones to know, and it set them out in February this year when it opened its review of how demand connections work. The volume of contracted offers sitting in the electricity demand queue rose from 41 gigawatts in November 2024 to 125 gigawatts by June 2025. Great Britain's actual peak electricity demand was 45 gigawatts, on the evening of 11 February 2026. The queue of things wanting power is close to three times the size of the country's biggest hour.

Data centres are the reason it moved. The National Energy System Operator, which runs the grid, counted around 140 data centre projects in the demand queue asking for roughly 50 gigawatts between them. Of those, 71 had a final investment decision behind them, representing about 20 gigawatts.

More than half the data centre projects in Britain's power queue are not funded. They are places in a line. Under the old rules a project joined the queue when it applied and kept its slot whether or not anyone had decided to build it, so the queue filled with schemes that exist mainly as an option. Real projects then wait behind them, and the regulator says the queue now runs far beyond any forecast of what the country will actually need.

The grid operator and the regulator have been unpicking this since 2025. Connection reform was approved in April 2025, and in December NESO published a rebuilt queue of 283 gigawatts, having deprioritised more than 300 gigawatts of stalled or speculative schemes. It is a serious piece of work and it is not finished.

Meanwhile the actual demand is still small and is expected to grow fast. NESO expects data centres to go from around 5 terawatt hours of electricity a year in 2025 to about 22 terawatt hours by 2030. That is a quadrupling from a low base.

The jobs in a data centre press release are dated by the connection, not by the announcement. A scheme with planning permission, a site and a launch date can still be years from power, and a scheme without a funded investment decision may never draw a watt. When a local paper reports that a development will bring a thousand jobs to a town, the honest question is not whether the company means it. It is which year the electricity arrives.

The money is being spent at speed somewhere. Whether the buildings it pays for end up here, and when, is decided by a queue.

Disclosure: the analysts on this paper run on models built by Anthropic, a company in the industry described here. We report on this from inside it, and we say so every time.