Japan Added a Data Centre Table to Its Energy Report, and Set PUE at 1.4

Japan’s Agency for Natural Resources and Energy publishes a manual each year telling large energy users how to fill in their annual report. The 2026 edition is dated 6 April 2026. Its own list of what changed since the previous edition runs to four lines. The first line adds a reporting form for data centres (ANRE, FY2026 reporting manual, indexed at the agency’s support tools page).

Japan now measures the energy behind its AI build-out inside a report that already existed. The new instrument is one table in that report.

Who files, and by when

The law is the Act on the Rationalisation of Energy Use and the Shift to Non-fossil Energy, Act No. 49 of 22 June 1979, amended in April 2023 to bring non-fossil energy inside its scope. It reaches any business using 1,500 kilolitres of crude-oil equivalent a year or more. Those businesses are designated as 特定事業者 and file two things: a periodic report of what they actually used, and a medium-to-long-term plan for what they intend to change. The manual gives the deadline as the end of July, every year (ANRE pamphlet on the revised Act).

The new form is 特定-第8表, “Status of energy use in the data centre industry”. Its scope note limits it to data centre operators whose own annual energy use reaches that same 1,500 kilolitre threshold.

The table divides an operator’s business into three kinds, and the definitions are worth reading closely because they decide who counts a given kilowatt-hour.

Housing type covers a company that provides the space and the supporting equipment while somebody else owns the IT hardware, which is colocation in ordinary English. In the hosting and cloud owner type, one company owns the building and the IT equipment and sells the processing that runs on it. The tenant type takes its space from somebody else and owns only the equipment it puts inside.

An operator running more than one of these reports each separately. A tenant-type operator’s IT load lands on the tenant’s own report. The supporting load around it lands on the landlord’s.

That allocation changed fairly recently. Until the FY2023 periodic report, a data centre’s energy use was counted by the owner whatever the business form. From that report onward the owner counts everything except the equipment a tenant brought in, while the tenant reports the part it exclusively occupies. Where the tenant’s area is not separately metered, the manual asks the owner to estimate it by a reasonable method and pass the figure on, and if nothing arrives the tenant estimates it instead.

Buying cloud capacity does not put the energy onto your own report. The manual says that users of the IT equipment of a hosting and cloud owner-type operator do not include that energy use at all, because the operator counts it.

One warning sits in the manual’s own text, and it is easy to miss. Everything above describes how energy use is allocated for the report in general. The manual notes separately that 特定-第8表 uses different categories for that allocation. The split deciding your headline energy figure is not the split that fills in the data centre table.

What the table asks for

The form opens with contracted power and maximum receiving capacity, both in kilowatts.

Then it asks for electricity in kilowatt-hours, split between the information-processing equipment and the ancillary equipment that keeps it cool and powered. A third line carries the total.

Next to each of those three electricity lines sits a smaller box: 「うち、非化石電気の割合 ％」. The share of that electricity which came from non-fossil sources. An operator reports it for the computing load and for the supporting load as separate numbers, not once for the building.

That separation is the demanding part of the form. A company that buys one clean-power contract for a site and reports a single percentage at the meter does not have what the form wants.

After the electricity comes the PUE value, given as an operator average and again for each business type. Then energy consumption intensity with its year-on-year change, a five-year average change, and the utilisation rate as a percentage.

The benchmark is 1.4

Japan runs a benchmark system inside this law. Industry categories each carry an indicator and a 目指すべき水準, the level to aim for. Blast-furnace steelmaking is set at 0.531 kilolitres per tonne of crude steel. A hotel is measured against the average for a hotel of the same size, service level and occupancy. A shopping centre is measured per square metre of floor area.

Category 16 is the data centre industry. Its indicator is the PUE of the data centre the business operates. The level is 1.4 or below.

PUE is total facility electricity divided by the electricity reaching the IT equipment. At 1.4, every watt of computing carries roughly four-tenths of a watt of overhead, most of it cooling. Lower is better, so coming in above 1.4 is the miss.

The treatment of a miss is set out in the manual’s instruction for 特定-第7表 1-2. An operator compares its figure for the year against the level to aim for, then writes in the name of the site concerned, the reason the level was not reached and the circumstances the business is dealing with.

The manual attaches no penalty to that entry. It is filed with the ministry as part of the periodic report, and the state publishes that report if the company joins the disclosure scheme, which has been in full operation since FY2024.

The roof, from FY2027

The second line of the change list adds a rooftop solar form, to 特定-第10表 and 指定-第8表. It is the only item in the list marked 【2027年度報告から】, so it starts with the FY2027 report rather than this one.

It asks for the roof area that already carries panels, and for the roof area that could carry them, leaving out the parts occupied by other structures, those in daily use for a specific purpose, and anywhere another statute forbids installation. Then it asks for one number: of the roof that meets the conditions, the percentage that already has solar on it.

Producing that percentage means surveying the company’s own roof area against each of those conditions.

What this adds to the rest of Japan’s data centre policy

Japan’s compute capacity is growing in particular places for particular reasons. We have written about the Hokkaido cluster and the HVDC link intended to put servers near wind, and about the interconnection queue that governs how quickly anything new can connect at all. The GX-ETS, which entered its mandatory phase in April 2026, puts a price on the emissions of large emitters.

The reporting form collects what those do not. Each year it records the computing load and the ancillary load with a non-fossil share on each, alongside a utilisation rate showing how full the building ran. Gathered from every operator over the threshold, that becomes a national picture of compute efficiency, which nobody currently holds.

The benchmark also gains a reported figure to sit beside it. Before this table, no PUE number was collected under the Act.

What an operator should actually check

The first question is the boundary: which of the three business types the operation falls into, and whether it falls into more than one, since that decides which loads are yours and which belong to a counterparty. The next is whether your non-fossil share exists as two numbers or one, because the form wants the computing load and the supporting load separately. The PUE worth knowing is the measured one, taken per site across the year rather than read off the design specification. And the roof can be surveyed a year ahead of the form that will ask about it.

Tech for Impact Summit 2027 is in Tokyo on 18–19 May 2027, co-hosted with the United Nations University. The summit is invitation-only. If you work on data centre energy, compute efficiency or corporate disclosure, join the waitlist.