The UK’s AI plans have run into a very practical question: how much energy will all this computing need?
According to reporting by the Guardian, updated government evidence now suggests that AI datacentres in the UK could be responsible for far more carbon emissions than earlier official figures implied. The figures sit alongside the government’s UK Compute Roadmap, which was first published in 2025 and updated again on 23 April 2026.
For most people, “compute” sounds abstract. It means the servers, chips, networking and datacentres that make modern AI work. When you ask a chatbot to summarise a document, generate an image, write code, analyse a photo or keep a voice assistant running, the answer does not appear by magic. It is processed somewhere, often in a large datacentre that uses electricity for both the computing itself and the cooling needed to keep equipment running.
What changed?
The Guardian says the revised estimate for AI datacentre emissions is now between 34m and 123m tonnes of CO₂ over the 2025 to 2035 period. The higher end is a striking number, and the government page confirms that the Compute Evidence Annex was replaced on 23 April 2026 after earlier environmental modelling was removed in March for updating.
That does not mean every AI search or chatbot question is suddenly disastrous. It does mean the national picture matters. If AI use spreads through workplaces, schools, customer service, home devices and creative tools, the combined demand for servers can become a serious infrastructure issue. The important shift is from thinking of AI as only a software feature to seeing it as a service with physical costs: land, water, power, cooling, cables and backup systems.
Why ordinary users should care
Most UK users cannot choose which datacentre handles a chatbot request. You also should not have to become an energy analyst before using a useful tool. But this story is a reminder that convenience still has trade-offs.
For households, the clearest lesson is to be more intentional. AI can be genuinely helpful when it saves time, explains something clearly, helps with admin or makes work less frustrating. It is less compelling when apps add AI for novelty, when a simple search would do, or when a device keeps listening and processing because that makes the product feel futuristic.
That distinction matters as AI moves into more everyday tools. We have already looked at how Gemini for Home can keep a conversation going, and why users should check what is being listened for and when. Energy use is another reason to ask whether a feature is solving a real problem or just adding more background computing to your day.
What to check before turning on another AI feature
Start with usefulness. If an AI feature helps you complete a task better or faster, it may be worth using. If it is mostly decorative, leave it off. This is especially true for always-on assistants, automated summaries you never read, image generation for throwaway posts, and tools that reprocess the same information again and again.
Then check settings. Many apps now let you control whether AI features are enabled, whether chats are saved, whether data can be used for training, or whether a tool has access to files, photos, microphone input or email. The privacy setting is not the same as the energy setting, but both are part of the same practical question: do you want this service running, and what are you getting in return?
Finally, look for clearer claims from the companies selling AI tools. If a firm says its product is greener, cheaper or more efficient, ask what that means in practice. Is it using renewable power? Is the model smaller? Is processing happening on your device rather than in the cloud? Is the company publishing useful environmental reporting, or only vague reassurance?
This is also a policy question
Individual habits help, but the bigger decisions sit with government, regulators, energy planners and the companies building datacentres. The UK wants AI infrastructure because it may support research, public services and economic growth. At the same time, the country has climate commitments and an electricity grid that already needs major upgrades.
That tension should be discussed openly. Datacentres can bring investment and jobs, but they can also put pressure on local power supplies, water use and planning decisions. If AI is going to become part of public services and everyday work, the public deserves clear numbers about the cost as well as the benefits.
The practical takeaway is not “never use AI”. It is to treat AI like any other powerful service: useful when it earns its place, wasteful when it is switched on by default without a clear reason. As more products add AI buttons, UK users should feel comfortable asking a simple question before clicking: what problem is this actually solving?
