Britain's AI plans overlook the computing power everything depends on
According to TechRadar, the UK's new AI partnership with the US highlights security, but the strategy has yet to address where computing power will come from.
According to TechRadar, Prime Minister Andy Burnham announced a new AI partnership between the United Kingdom and the United States at the UN General Assembly last week. The stated goal is to protect critical national infrastructure, with the outlet describing the scope as ranging from sea lanes to airspace.
TechRadar presents the move as evidence that Britain is beginning to treat artificial intelligence as a matter of national security as well as economic growth. The outlet notes that AI is moving beyond chatbots and productivity tools into defense and other systems where reliability matters most.
The article argues that the strategy leaves out a basic question: where the computing power needed to run these systems will come from. TechRadar points to the country's universities and startup ecosystem as real strengths, while suggesting that public attention has centered on models rather than the hardware and capacity beneath them.
The reporting does not detail how the partnership will be structured, funded or governed. It also does not set out what specific commitments, if any, accompany the announcement on domestic computing capacity.
Background
Modern AI systems rely on large amounts of computing power. Training advanced models and serving them to large numbers of users both depend on specialized processors, most often graphics processing units or similar accelerators. These chips sit in data centers that need reliable electricity, cooling and high-speed networking, so each layer of the system brings its own constraints.
Supply is also concentrated. Advanced chips are designed by a small number of companies and manufactured by an even smaller group of specialized fabrication plants. Export controls imposed by major governments can further limit which countries and companies can buy the most capable hardware, which makes access a diplomatic and commercial question as well as a technical one.
For many countries, the practical result is dependence. Organizations without their own facilities often rent capacity from large cloud providers, many headquartered in the United States. That arrangement can be efficient, but it raises questions about pricing, priority during shortages, and who can restrict access in a crisis. Governments that see AI as strategic have responded in different ways, including funding public research clusters, offering incentives for data center construction and negotiating supply arrangements with allies.
Energy is an increasingly central limit. Data centers draw large, steady loads, and connecting them to the grid can take years in some places. Planning rules, water use and local opposition can slow projects further. These physical realities mean that AI policy is partly a matter of power, land and infrastructure, areas that software-focused strategies often treat lightly.
Why it matters
The partnership signals a shift in how governments think about artificial intelligence. Treating AI as part of national security means that questions once confined to defense procurement, such as supply chains, redundancy and trusted vendors, now apply to software that runs everyday services. For the public, that shift could change how quickly new tools reach hospitals, transport networks or energy systems, and under what conditions.
The compute question sits beneath every such ambition. A strategy built around models and research talent can look complete on paper while still depending on hardware and capacity controlled elsewhere. If access to computing power is constrained, the practical effects could include slower deployment, higher costs for domestic companies, or reliance on overseas providers whose priorities may not match British interests during a crisis.
There is also a question of coordination. Cooperation with a major ally can strengthen security and improve access to technology, but it can also create dependencies that are hard to unwind. How the partnership handles shared standards, data handling and export rules will determine whether it builds British capacity or extends existing reliance. For businesses planning long-term investments, clarity on those points may matter as much as the headline itself.
What to watch
Watch for details of the partnership: its scope, governance, funding and any named projects. Look for whether the government sets out a plan for domestic computing capacity, including where data centers would be built, how they would be powered, and how advanced chips would be secured. Pay attention to how the agreement treats critical infrastructure in practice, including testing, certification and responsibility when systems fail. Finally, track whether public and industry discussion focuses on compute and energy or stays centered on models and regulation. That emphasis will show whether the compute gap is being taken seriously.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
Announcements about national security and AI partnerships are easy to make and hard to measure. The core issue raised here is simple: a strategy is only as strong as the hardware and power it can rely on. So far the public record offers a headline about protecting infrastructure, but not a clear account of how the UK will secure the computing capacity that such protection would require. Which facilities will be used? Who owns them? What happens if access is restricted during a disagreement between partners? None of these questions is unreasonable, and none has an obvious answer in the coverage so far. Skeptics should not assume the worst, but they should ask for specifics before applauding. A partnership that sounds comprehensive may still leave the most basic dependency untouched. The real test will be whether future documents name concrete commitments on capacity, energy and independent oversight, or whether the language stays at the level of intent.
The Builder the engineer's view
From an engineering standpoint, the compute gap is not abstract. Running reliable AI in defense or infrastructure settings means planning for latency, failover, security patching and hardware replacement, all of which depend on where and how processors are housed. A model that performs well in a lab can behave differently when it must run continuously under load, be audited, and keep working when a link to a remote region fails. Builders tend to ask unglamorous questions first. How much capacity is needed? Where will the power come from? Which chips, supplied by whom? How are updates tested before they reach critical systems? These are not footnotes. They determine whether a national plan can be delivered in months or in years. If the UK wants resilience, it should invest in the less visible parts of the stack as seriously as it invests in research, and publish enough detail for engineers to judge whether the plan holds together.
The Rulebook privacy, safety and regulation
Putting AI into critical infrastructure raises questions that go beyond performance. Who is accountable when an automated system makes a harmful decision? What data is used, where is it stored, and who can access it? International partnerships can make sense for security, but they also need clear rules on information sharing, oversight and redress. The rulebook for this area is still being written, and compute is part of that story because it determines where sensitive data is processed and under whose legal authority. A facility located abroad, or run by a foreign provider, may fall under different obligations than a domestic one. Regulators, lawmakers and the public will want to know which safeguards travel with the technology, not just what it can do. Good intentions about protecting infrastructure should be matched by enforceable standards, independent review and transparent reporting when things go wrong.