Emissions from the new data centres set to drive the UK’s AI “revolution” could be hundreds of times higher than government estimates, according to analysis by Carbon Brief.
There are dozens of data centres being developed across the country, potentially driving a surge in electricity demand.
Amid uncertainty about the scale and pace of this expansion, there are mounting concerns that new data centres could pose a threat to the nation’s climate goals.
UK government analysis concluded that the emissions from data centres would be negligible, even if they expand rapidly – a finding one campaigner tells Carbon Brief is “nonsense”.
In contrast, Carbon Brief analysis finds that emissions from powering data centres could be far higher than the government figures suggest, if at least a small amount of the electricity they need is generated by burning gas.
Data centres could run entirely on low-carbon electricity, but some in the sector have argued that the government’s AI ambitions require the UK to use more gas power.
If new data centres source a large amount of their power from gas, it could cause carbon dioxide (CO2) emissions equivalent to at least Denmark’s annual total.
‘AI superpower’
Data centres are energy-intensive computing facilities that are required to train and run complex AI models, among many other things.
The UK is one of the top-ranking nations for data-centre capacity, with roughly 1.8 gigawatts (GW) of facilities consuming more than 2% of national electricity. This could grow rapidly in the coming years as the government aims to make the UK an “AI superpower”.
Companies have already “achieved financial commitment” to invest in 71 new data centres that, if built, would require around 20GW of electricity, according to energy regulator Ofgem.
(For reference, the UK’s average electricity demand in 2025 stood at around 37GW.)
This potential increase in electricity demand has raised concerns from campaigners and some MPs about the impact of data centres on the UK’s climate targets.
Last year, the government’s plan for meeting its 2035 climate target noted that AI growth was “not factored into” emissions projections, although energy secretary Ed Miliband has said new data centres are captured in modelling of “overall electricity demand growth”.
The government is targeting a “clean power system” by 2030, with just a small amount of gas generation remaining. Extra demand from new data centres could require a rollout of clean power that is even faster than the growth already underway.
If clean-power growth does not keep pace, data centres could, therefore, prolong the use of gas power, either by requiring more gas to remain on the grid or by facilities building their own on-site gas generation.
There is significant uncertainty around future emissions from UK data centres, which will depend on the number of centres built, how clean their power is and when they come online.
The government published an analysis of its AI strategy’s climate impact last year, alongside a data-centre “roadmap”.
The analysis, released by the Department for Science, Innovation and Technology (DSIT) suggests emissions from future data centres will be minimal – reaching a maximum of 0.142m tonnes of CO2 (MtCO2) from 11.2GW of AI-related computing power by 2035.
(There is an additional 2.4GW of data-centre demand in this scenario that is not associated with AI, for which emissions are not calculated.)
This figure is based on what DSIT describes as a high-emissions, high-AI growth scenario. Yet it implies that each unit (kilowatt hour, kWh) of electricity supplied to the 11.2GW of AI data centres would be associated with less than 2g of CO2. In other words, their electricity supply would need to be almost completely decarbonised. The government aim is for 50gCO2/kWh by 2030.
In addition, the DSIT figure – for emissions associated with the entire UK data centre fleet in 2035 – is much lower than the emissions estimates reported in planning applications for individual UK data centres made by Google and other companies.
Gas power
The chart below, based on Carbon Brief analysis, shows how data-centre emissions could be far higher than the government’s figures suggest.
Even if gas-fired electricity only accounts for 5% of their supply – indicated by the smallest blue column below – emissions from 11.2GW of data centres would be around 2MtCO2. This is more than 10 times higher than the government’s top estimate for 2035.
If the same data centres rely more heavily on gas, emissions could be hundreds of times higher, exceeding 30MtCO2. This is roughly equivalent to the annual emissions of Denmark. Emissions could rise even higher if capacity increases in line with the extra 20GW of data-centre demand that Ofgem says is in the pipeline, as indicated by the red columns
If data-centre expansion reaches 20GW and those centres rely heavily on gas power, then the figure could be as high as 70MtCO2, the annual emissions of Sweden. This would also be nearly 500 times higher than the government’s upper estimate, which it says is based on a “pessimistic decarbonisation” scenario.
(The numbers are not directly comparable as, unlike the AI-specific 11.2GW figure, it is unclear how much of this 20GW would be for AI, specifically.)
The government’s modelling states that AI emissions in 2035 would be “equivalent to below 0.05% of the UK’s projected total emissions”. It also says “this could be equivalent to the annual emissions of approximately 5,000 to 23,600 UK households”.
On the contrary, Carbon Brief’s analysis suggests data centres could, in fact, be equivalent to as much as 20% of the UK’s projected total emissions in 2035.
As for the number of households, Carbon Brief estimates that future data centres could result in emissions equivalent to as many as 11.4m homes, roughly a third of all UK households.
Dr Tim Squirrel, head of strategy at Foxglove – part of an NGO group calling for more government scrutiny of data-centre emissions – tells Carbon Brief the DSIT figures are “nonsense and threaten to derail our carbon budgets”. He says:
“The figures that DSIT projects here wildly downplay data-centre emissions, even by the standards of the most optimistic energy transition scenario. There is no way that the amount of compute they anticipate can be built and produce the miniscule emissions they’re calculating.”
In its analysis, the government attributes the low emissions figures to “more efficient models and hardware” and “the UK’s ambitious targets for electricity grid decarbonisation”.
When asked by Carbon Brief, DSIT declined to provide any more information about its analysis.
Clean growth
While the UK is prioritising data centres for AI, there is mounting industry pressure to allow gas-power expansion for this “critical” infrastructure, as is happening in, for example, the US and Ireland.
Developers in the UK have reportedly already “turned to gas” via private electricity supplies, due to struggles securing a connection to the public network.
Yet, new data centres could be completely emissions-free if they are powered entirely with on-site clean energy or using electricity from a decarbonised grid.
As it stands, most data centres are connected to the electricity grid. Some enter power purchase agreements (PPAs) in which they financially support renewable-energy operators, allowing them to describe their electricity as clean.
Katie Davies, head of energy and infrastructure policy at techUK, a trade association representing the technology sector, highlights this expansion of PPAs as important for driving the growth of wind and solar power:
“In doing so, data centres actively contribute to additionality by unlocking extra carbon-free capacity that might not otherwise come online.”
A report last year by Aurora Energy Research found that data centres could provide a “route-to-market” worth up to £35bn for 19GW of UK renewables. However, it added:
“If renewables capacity and networks don’t keep pace, additional data centre demand will likely be met by carbon-intensive sources of generation.”
The UK’s “AI opportunities action plan” includes the establishment of “AI growth zones“, which the government says will be in areas with “available clean energy”. It is also overhauling the grid connection queue, which Davies says is important:
“Reducing this queue through strategic alignment and the removal of speculative applications will be vital to ensuring [data-centre] operators do not have to turn to higher-carbon energy sources as a last resort.”
Responding to Carbon Brief’s analysis, a government spokesperson said:
“We want the UK to be at the forefront of AI, but we are clear this must be done sustainably. That is why our AI growth zones are supporting development in areas with access to clean power, while the AI Energy Council is exploring how AI can be powered by responsible, clean-energy sources.”
Methodology
There is considerable uncertainty around data-centre power demand and emissions, with much of the relevant information not in the public domain. Carbon Brief has performed some rough calculations based on available data.
The government figure comes from an annex to DSIT’s UK compute roadmap. DSIT analyses the emissions impact of expanding the UK’s data-centre capacity to between 7.4GW in a “low compute-demand scenario” and 13.6GW in a “high compute-demand scenario” by 2035. (The majority of the demand in each scenario is from AI.)
DSIT also uses an “AI environmental impacts model” to estimate the greenhouse gas emissions from AI compute, only covering the 11.2GW AI component of data-centre capacity. It concludes that AI emissions in 2035 could range from 0.025MtCO2 to 0.142MtCO2. This includes both “direct” and “indirect” emissions, indicating that it covers more than just emissions from the electricity used to power the data centres.
A widely reported consultation by the energy regulator, Ofgem, found that there are proposals for around 140 new data centres in the UK, which would require 50GW of electricity if they were all built.
In reality, it is highly unlikely that all of these data centres will be completed, with a “significant number” expected to fail when trying to secure funding or planning permission.
The 20GW figure used in this analysis is based on the 71 “mature” projects that have “achieved financial commitment with final investment decision”, according to Ofgem.
Carbon Brief used the top government figure of 0.142MtCO2, even though it represents a “pessimistic grid decarbonisation” and “high compute demand” scenario.
To calculate the emissions from powering data centres in the future, Carbon Brief assumes a data-centre “load factor” of 90%, which is in line with other analyses. The analysis uses different shares of gas in the centres’ power supplies to indicate a range of future possibilities, assuming emissions from gas power are 0.4MtCO2 per terawatt hour.
The post Analysis: CO2 from UK data centres could be ‘hundreds of times’ higher than thought appeared first on Carbon Brief.
Analysis: CO2 from UK data centres could be ‘hundreds of times’ higher than thought
Claver Gatete is Executive Secretary of the UN Economic Commission for Africa. Jason Veysey is Energy Modeling Program Director and Senior Scientist at the Stockholm Environment Institute. Lisa Sachs is Director of the Columbia Center on Sustainable Investment at Columbia University.
The case for global energy transition has rarely been clearer. The closure of the Strait of Hormuz earlier this year exposed the cost of unplanned, fossil-dependent systems, while the falling cost of renewables, the rising penetration of electric vehicles, and the growing value of demand flexibility have made the direction of travel obvious. The benefits of a clean, secure, integrated system are no longer in dispute. What remains unclear is how to build it.
Countries around the world have called for faster renewable energy deployment and alternative energy arrangements. A secure, affordable, resilient, decarbonised system requires specific investments in specific places in a specific sequence, optimised across sectors and borders. But very few governments have the analytical foundation to translate those imperatives into investment.
The two instruments that are supposed to determine investment priorities for decarbonisation – Nationally Determined Contributions (NDCs) and country platforms – cannot answer the most basic question facing any country undertaking an energy transition: what should the energy system look like?
-
New coal mine openings slow as East Asian demand plateaus
China and Australia drove global decline in new capacity additions in 2025, but planned production kept growing
To close this gap, every country needs a bankable, economy-wide optimisation model for its energy system. A model is not a plan, but it can help answer the critical question of what the future energy system should look like. It shows how optimal scenarios vary as assumptions and policies are adjusted, calculates investment requirements and sequencing, and quantifies how system costs are affected by assumptions, policies, and exogenous variables like trade policy and financing terms.
Tool for efficient investment
Optimisation is a simplified way of simulating an energy system, but it can be an extremely powerful tool for moving energy planning from reactive (how do we manage the disparate actions in the energy system?) to intentional (what energy system underpins our national objectives?). A model can show how optimal scenarios vary as assumptions and policies are adjusted, and how investment requirements are quantified and sequenced.
Optimisation models can treat the energy system and the sectors it serves as an integrated whole, optimising across sectors and projects in ways that can be mutually reinforcing. If considered independently, growth in industrial demand, transport electrification, and digital infrastructure can add stress to the energy system. But an optimised plan can arrange these and other changes in an efficient, synergistic way.
Two to tango: How governments can unlock private investment for national climate goals
New load can be added where low-cost power is available; industrial customers can ensure the viability of investments in energy supply; electric vehicle charging policy can smooth load curves and reduce costs for all consumers.
Additionally, optimisation modeling can also change the financeability of investments. Taken alone, each project faces uncertainty about the rest of the system, which raises the cost of capital and causes projects to stall or unwind after contracts are signed. A coherent, optimised plan makes visible the coordination that private capital would otherwise have to bet on: identified offtake, sequenced and committed transmission, contracted power supply, and so on.
What COP31 and COP32 should do
The upcoming COPs in Turkey and Ethiopia can shift the center of gravity of international climate cooperation from fragmented commitments to planning. Three moves are urgently needed.
First, optimised, economy-wide, long-term energy system planning must be the foundation on which any meaningful NDC, country platform, or finance commitment rests. NDCs are typically drafted by environment or single-line ministries, with limited cross-sectoral input from ministries of energy, finance, and planning. They contain targets, derived from sectoral strategies or national commitments, not from an analytically grounded picture of what the energy system should look like and what investments would make it work. Country platforms are generally a portfolio of investments assembled from existing project pipelines, rather than derived from a system-level analysis of what an optimised, decarbonised energy system would require.
Second, recognise regions as a key planning unit. Modern integrated energy systems are inherently regional. Renewable endowments are unevenly distributed; balancing variable supply across borders lowers aggregate cost, reduces redundant backup capacity, and unlocks economies of scale no individual nation can achieve. Many energy investments in Southeast Asia, East Africa, Southern Africa and Central Asia may only be financeable in a regional context. Assessing domestic infrastructure without regional optimisation perpetuates the perception that decarbonisation is more expensive than it is.
COP31 leaders unveil global targets, with spotlight on electrification
Third, finance the planning capacity. A coordinated commitment by multilateral development banks, bilateral donors, and philanthropic partners to help every region and its constituent countries develop and maintain their own modelling capability, with open-source tools and regional analytical hubs, would close the most consequential gap in the current architecture. The cost is small relative to current spending on country platforms, failed project preparation, and misallocated infrastructure investment.
This includes supporting regional institutions such as the ASEAN Centre for Energy, the African Energy Commission, regional power pools, and the Latin American and Caribbean Energy Organization to determine what optimised regional systems require. Country-by-country pledging, repeated at every COP, will not deliver what meaningfully integrated systems can.
The 2026 energy crisis made the cost of unplanned, fossil-dependent systems newly visible. That window of clarity will close. The international community should seize the moment to build the planning foundation that has been missing for thirty years, rather than commissioning another round of NDCs or pledges, striving for outcomes neither was designed to deliver.
The post Every country needs a model to help optimise its energy transition appeared first on Climate Home News.
Every country needs a model to help optimise its energy transition
The world is currently experiencing what is expected to become the strongest El Niño on record – dubbed a “super El Niño” by many.
El Niño is the warm phase of a recurring climate pattern in the tropical Pacific that releases heat from the ocean into the atmosphere.
This temporarily raises global temperatures and reshapes rainfall and extreme weather around the world – impacting the lives of billions of people.
The current El Niño event began in June and is expected to last into 2027.
El Niño is part of a wider climate pattern called the El Niño-Southern Oscillation (ENSO) cycle.
The ENSO cycle also has a cool phase, known as La Niña, as well as a “neutral” phase. El Niño and La Niña events typically last between nine and 12 months, but can go on longer.
Below, Carbon Brief explains how the ENSO cycle works, its impacts on extreme weather and global temperatures and why this El Niño event is projected to be the most intense since records began.
The post Explainer: How the ‘super El Niño’ will reshape the world’s weather appeared first on Carbon Brief.
https://interactive.carbonbrief.org/el-nino-explainer/index.html
The second-largest reservoir in the US reached a record-low water height on Saturday – just days after the country’s largest reservoir broke its own record.
Both Lake Mead and Lake Powell are located on the Colorado River.
They provide water for populations across seven US states in the south-western US, with around 40 million people getting some or all of their municipal water from the Colorado River.
The river also provides water for around 5.5m acres (22,258 square kilometres) of farmland across Colorado, Arizona, California and the other states in the river basin.
Experts tell Carbon Brief that climate change, population growth and over-consumption are all contributing to the current record-low levels of the reservoirs.
Record lows
At full capacity, Lakes Mead and Powell can hold a combined 68 cubic kilometres of water – enough to supply all household consumption in the contiguous US for nearly 1.5 years. However, the water level in both reservoirs has been declining for decades.
The chart below shows the water level of Lake Mead, in metres above mean sea level. The reservoir, which began to fill in 1935 following the construction of the Hoover Dam, has a “full pool” maximum capacity of 347.60 metres. The water level in Lake Mead reached a record low of 317.11 metres on 7 August.
The following chart shows the water level of Lake Powell, in metres above mean sea level. Lake Powell’s full-pool level is 1,127.76 metres.
While the reservoir reached its maximum capacity several times in the 1980s, it has not done so since. On 15 August, the water level in Lake Powell was recorded at a new record-low of 1,072.87 metres.
Both reservoirs have continued to decline in the days since breaking their respective records. The downward trend will largely continue in both lakes until next spring, when the snowpack in the mountains of the Upper Colorado River Basin begins to melt, says Dr Jack Schmidt, a senior research scientist at Utah State University’s Center for Colorado River Studies. He tells Carbon Brief:
“The big dilemma of the moment is that we’re only in the middle of August, and we have no assurance of what the coming winter will be. The only thing we can be sure of is that we will be depleting overall total basin reservoir storage from now until, roughly, early April.”
Compounding factors
The record lows across the two reservoirs are the result of several compounding factors, experts tell Carbon Brief.
Since the turn of the 20th century, the amount of water flowing along the Upper Colorado River has declined by about 20%. Research suggests that half of this decline can be attributed to human-induced climate change.
Most of the river’s streamflow comes from the snowpack of the Upper Colorado River Basin, which stretches across five western US states but is primarily located in Colorado and Utah.
This region has been gripped by a historic “megadrought” for more than a quarter of a century. Nearly half of the megadrought’s intensity over 2000-18 is attributable to climate change, according to a 2020 study.
At the same time, the increasing population in the US south-west has put added pressure on the Colorado River’s water supply. The number of people obtaining some or all of their water from the Colorado system has grown by 15 million (around 60%) since 1992.
Schmidt tells Carbon Brief:
“There’s an ultimate cause of the present water crisis, and there’s a proximate cause. The ultimate cause is a warming climate, a warming planet and a pretty clear correlation between warming conditions and decreased runoff in the Colorado River Basin.
“The proximate cause is that in this messy democratic republic of ours, big policy decisions that match the variability of the climate occur painfully slowly – with intense political negotiations – and only incrementally.”
On 31 July, the US Bureau of Reclamation, which manages water resources in the western US, released an environmental impact statement on its proposed post-2026 strategy for managing Lakes Powell and Mead. The strategy itself has not been released yet.
Schmidt notes that the statement does appear to give the Bureau flexibility to “respond to crisis” by reducing the delivery of water to several states. However, he adds:
“They acknowledge it won’t work if we just stay critically dry, and of course every climate model for the 21st century, especially with a continually warming planet, says that that’s exactly what’s going to happen.”
The post Analysis: The two largest reservoirs in the US have hit record-low levels appeared first on Carbon Brief.
Analysis: The two largest reservoirs in the US have hit record-low levels
-
Climate Change1 year ago
Guest post: Why China is still building new coal – and when it might stop
-
Greenhouse Gases1 year ago
Guest post: Why China is still building new coal – and when it might stop
-
Greenhouse Gases2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago
Bill Discounting Climate Change in Florida’s Energy Policy Awaits DeSantis’ Approval
-
Renewable Energy10 months agoSending Progressive Philanthropist George Soros to Prison?
-
Greenhouse Gases1 year ago
嘉宾来稿：探究火山喷发如何影响气候预测
-
Carbon Footprint2 years agoUS SEC’s Climate Disclosure Rules Spur Renewed Interest in Carbon Credits