Вход на сайт

Просмотр новости

Найдите то, что Вас интересует

IDCA warns of ‘delay tax’ as AI datacentre power surges 50%

Дата публикации: 14-07-2026 19:01:00

Global datacentre electricity consumption reaches 1.9% of all generation, with ‘zombie workloads’ wasting 3GW of US capacity alone and a 100MW project delay costing up to $1bn

Основное содержимое страницы с новостью.

Global datacentre electricity consumption reaches 1.9% of all generation, with ‘zombie workloads’ wasting 3GW of US capacity alone and a 100MW project delay costing up to $1bn

Antony Adshead

By

Published: 15 Jul 2026 0:01

Artificial intelligence (AI)-focused datacentres grew electricity consumption by 50% during 2025, with demand now at 67.7GW – 1.9% of all electricity generation – and the International Energy Agency (IEA) projecting it will more than double to 945TWh by 2030.

Datacentres are also increasingly subject to a “delay tax”, where slow grid upgrades can cost more than $10,000 per MW per day. Meanwhile, more than 3GW of datacentre capacity – as much as 13% of US datacentre power – supports “zombie workloads”, unused cloud containers and services that were never properly decommissioned. 

That’s all according to the latest Global energy report from the International Data Center Authority (IDCA), which also pointed to 11 US state moratoria or stop-build legislation, operators moving behind the grid with their own onsite generation.

Also, on emissions, IDCA has introduced a CO2 efficiency index measuring tonnes of CO2 per $1m of GDP, and finds the UK ranks as the 11th most efficient globally at 86.4 tonnes, compared with 156 for the US and 647.9 for China, against a world average of 316.5.

The delay tax arises from a fundamental mismatch in infrastructure timelines. A modern datacentre can be deployed in 18 to 24 months, but upgrading grid connections typically requires three to five years. IDCA says average lost revenue costs exceed $10,000 per megawatt, per day, and that a delayed 100MW deployment can approach the billion-dollar mark when the cost of capital and demand shifting elsewhere are factored in.

The zombie workload problem points to a different kind of waste. Before building new, capital-intensive power generation, IDCA urges operators and governments to audit their “logical layers” to eliminate these massive inefficiencies. The 3GW of wasted capacity in the US alone is presumed to be increasing worldwide as cloud computing adoption rises. Reclaiming it, the report suggests, could represent a cheap and fast way to free up grid capacity for AI growth.

The community veto phenomenon in the US is no longer episodic friction, IDCA finds, but is “highly organised, armed with technical consultants, pre-written ordinance language, and heavily funded ratepayer coalitions”, in which resident groups fight to avoid subsidising transmission and generation upgrades required by datacentres. Maine appears to be the first state moving a statewide datacentre moratorium through its legislature, and similar measures have been introduced in at least 10 other states.

Behind the meter power

The shift to sovereign power – onsite generation behind the meter – is now “federally backed grid policy”, says the report. It cites June’s FERC orders that explicitly directed US Regional Transmission Organizations (RTOs) to build tariffs that accommodate co-location and behind-the-meter generation. 

Operators including Google, Microsoft, Equinix and Amazon have already signed deals for nuclear power from SMRs and restarted reactors and existing plants. LNG-fired gas turbines are also being widely used as a bridging fuel, though IDCA warns of a manufacturing bottleneck.

Demand for large-frame gas turbines from GE Vernova, Siemens Energy and Mitsubishi Power is so intense that an order backlog now exists for some units until 2030, with turnkey project costs reportedly doubling from roughly $1bn per GW to as much as $2.5bn per GW.

Regulatory landscape shifts

The regulatory landscape continues to shift rapidly on both sides of the Atlantic.

In the US, the June 18 FERC orders, issued under Section 206 of the Federal Power Act, require major RTOs to report on their ability to supply within 30 days and justify tariff proposals within 60 days. FERC deliberately left retail and facility siting to individual states, but singled out the Southwest Power Pool’s High Impact Large Load process as a potential model for fast-tracking operator applications.

In Ireland, the government officially ended its multi-year datacentre moratorium in December 2025 by introducing a Large Energy Users Connection Policy. New facilities must now operate under a “bring your own power” model, installing onsite or proximate generation capable of matching their maximum requested demand. Operators must also meet at least 80% of annual electricity demand through additional renewable energy projects generated within the Republic of Ireland within six years of connection.

In Virginia – home to “Data Center Alley” in Loudoun County – a rate class takes effect from January 2027 that requires datacentres to cover 85% of transmission and distribution costs and 60% of generation demand, to directly addressing the cost-allocation tensions that have fuelled community opposition.

Read more on Datacentre capacity planning

Схожие новости

#Наименование новостиТональностьИнформативностьДата публикации
1Data centre delays expose AI cloud power limits0508-07-2026
2Shifting data center power to off-peak hours could cut grid costs in the age of AI0526-06-2026
3AI Data Centers Use Far More Water Than Most Tech Giants Report0703-07-2026
4Datacentres need to consider alternatives to grid connection013.9913-07-2026
5UK slows AI data center roll-out over energy crunch012.5530-07-2026
6Battery start-ups see ‘crazy’ demand to smooth power surges in data centres2702-07-2026
7OPINION. « Data centers, la facture cachée des promesses de l’IA », par Dejan Glavas, directeur de l’Institut AI for Sustainability07.1214-08-2026
8Data centers may emit 57% more CO₂ than expected in 2025, study finds-2730-06-2026
9 Comment on AI Electricity use: a lot or a little by Tm -2606-07-2026
10Bloomberg: дата-центры для ИИ к 2035 году могут потреблять 4,4% электроэнергии мира0019-09-2025

Классификация: . Схожих патентов: 0. Схожих новостей: 10. Тональность: 0. Информативность: 12.71. Источник: www.computerweekly.com.