Sam Richards, who runs the pro-building campaign group Britain Remade, opened a City AM column this week with Lewis Carroll’s Red Queen: in her country, it takes all the running you can do to keep in the same place. His argument is that Britain has spent decades assuming it lives in a slower world, and that AI has quietly moved the race to a speed the country is not built for. Strip away the literary framing and there is a concrete claim underneath, one that lands directly on UK businesses: the AI economy will be built somewhere, and where it gets built decides who captures the growth and who pays rent for it.
The problem is not ambition, it is throughput
Britain does not lack AI ambition. It has frontier labs, a dense financial sector, world-class universities, and two international AI safety summits to its name. What it lacks, in Richards’ telling, is the ability to build the physical things AI runs on at the pace the technology now sets.
His chain of dependency is worth taking literally because it is the part most strategy documents skip. AI requires compute. Compute requires data centres. Data centres require electricity. Electricity requires power stations, transmission lines, and grid capacity. Almost every link in that chain runs through the UK planning system, and the planning system is where things go to wait. Richards notes it has been over thirty years since Britain built a new nuclear power station, and that the one currently under construction faces further delay because regulators judged the £700m already spent on fish protection insufficient.
Strategic Reality: The binding constraint on UK AI is not model quality or talent. It is megawatts and the years it takes to connect them. A country can have the best researchers in the world and still end up renting its compute from someone else’s grid.
That is the uncomfortable core of the piece. Growth has been treated as something that happens independently of physical infrastructure, as if an economy could thrive without building anything. AI breaks that assumption because it is unusually hungry for the one thing Britain has become slowest at providing: power, at scale, in a specific location, soon.
Makers and takers, and why the line matters now
Richards draws a distinction that deserves to sit at the centre of any UK board’s thinking about AI. Today, Britain can buy its way around most technology gaps. It does not manufacture many semiconductors, but it can import them. It does not dominate cloud computing, but it can rent it. That has been a perfectly workable position for decades.
His claim is that AI changes the economics of being a buyer. The technology divides the world into makers and takers. Makers own the models, the chips, and above all the compute, and they capture the largest share of the economic gains. Takers consume AI services built elsewhere and pay ever-rising rents to foreign firms for the privilege. In a world where the most powerful systems are controlled by a handful of American companies and the Chinese state, being a taker is not a neutral position. It is a standing transfer of value out of the country.
| Dimension | Maker economy | Taker economy | UK’s current trajectory |
|---|---|---|---|
| Compute | Owns and hosts capacity domestically | Rents capacity from foreign providers | Renting, with limited domestic build |
| Value capture | Retains the margin on AI services | Pays rising rents that leave the economy | Exposed to rent extraction |
| Resilience | Controls availability during shortages | Subject to others’ allocation decisions | Dependent on external allocation |
| Leverage | Sets or shapes terms of access | Accepts terms set elsewhere | Rule-shaper abroad, price-taker at home |
The strategic point for business is that this is not only a national question. A UK firm whose entire AI capability rests on capacity rented from providers building everywhere except Britain inherits the country’s exposure whether or not anyone chose it. When compute is scarce and expensive, allocation decisions made in another jurisdiction become your operating constraint.
Competitive Reality: If domestic compute stays thin, the cost of running AI in Britain tracks decisions made in Virginia, Dublin, and the Gulf. Firms that assumed cloud pricing only ever falls are planning for a world that scarcity is quietly closing.
What “speed up” concretely requires
“Speed up” is easy to say and hard to cash. Richards is a campaigner for building, so his prescription is unapologetically physical: build the energy, the grid, and the infrastructure now, and stop treating caution as a form of safety. That is the right altitude for a national campaign. For a business reading it, the useful move is to translate the national bottleneck into the questions it actually changes on your side of the fence.
The first is where the constraint bites. Grid connection queues, planning timelines, and power availability are not abstractions if you are choosing where to site a workload, sign a multi-year cloud commitment, or plan capacity for an AI-heavy product. The lead times on UK power and data-centre capacity are long enough that they belong in three-year plans, not procurement footnotes.
The second is who bears the risk of slowness. If the national build stays slow, the practical hedge is portability. A firm that can move a workload between providers and regions in weeks is insulated from any single grid’s constraints in a way that a firm locked to one datacentre region is not.
Implementation Note: Treat compute location and provider concentration as a board-level risk, not an IT preference. The question is not only “what does inference cost today” but “what happens to that cost and availability if UK capacity stays scarce for the next five years”.
The third is the timeline mismatch. AI capability is compounding on a scale of months. British infrastructure moves on a scale of parliaments. Richards’ Red Queen is really a warning about that gap: the faster the technology runs, the more expensive every year of planning delay becomes, because the thing you eventually build has to catch up to a further-away target.
The part the column understates
Richards is making an argument, and a good argument leaves things out. Two are worth naming, because a business acting on the piece needs them.
The first is that being a taker is not uniformly ruinous. Most UK firms will never own compute, and for the vast majority the sensible strategy is to be an excellent, discerning buyer rather than a frustrated would-be maker. The risk Richards identifies is real at the national level and for compute-intensive sectors, but the individual firm’s job is usually to extract maximum value from rented capability whilst managing the concentration risk, not to fight a battle over sovereign chips it was never going to win.
The second is that speed and care are not always opposites. The column frames caution as the enemy, and in planning terms it often is. But the same country that struggles to build a power station also earned genuine convening authority by moving early and thoughtfully on AI safety. The honest position is that Britain needs to be fast on infrastructure and deliberate on governance at the same time, and that these are different muscles rather than a single dial to turn up.
Critical Context: The UK’s structural advantage in AI is not compute, which it will not win, but trust and rule-shaping, which it has already started to build. A strategy that trades all its caution for speed risks spending the one asset the country actually leads on.
Recommendations for UK organisations
The column is a call to national action. Here is how it converts into decisions a business can make this quarter, sorted by how far along the AI adoption curve you are.
For organisations early in AI adoption:
- Put compute dependency on the risk register before you scale. Record which providers and regions your AI workloads rely on, and what your cost and availability look like if that capacity tightens.
- Design for portability from the first build. Abstracting the model and infrastructure layer is far cheaper now than after you are locked into one provider’s region and pricing.
For organisations already running AI in production:
- Map your concentration. A stack that depends entirely on one hyperscaler’s UK region is a single point of failure against exactly the scarcity Richards describes.
- Bring infrastructure lead times into your planning horizon. If your roadmap assumes cheap, abundant, instantly available compute in 2028, pressure-test that assumption against how slowly the underlying capacity is actually being built.
For organisations in compute-intensive sectors:
- Engage with siting and energy now, not when you need the capacity. Grid connection and planning timelines mean the useful conversations start years ahead of the workload.
- Watch the policy signals on planning and energy reform. Where the national bottleneck is genuinely loosening is where the economics of building in Britain change, and that is worth tracking rather than assuming.
Take Action: Ask one question at your next planning session — if UK compute stays scarce and expensive for five years, which of our AI plans still work? The answers tell you where your real exposure sits.
The challenges that will not show up on the roadmap
Four risks in this picture are easy to miss until they cost something.
The first is invisible dependency. Teams adopt rented AI capacity because it is easy and available, and rarely record where it physically sits or what happens if that changes. The exposure is real long before anyone measures it.
The second is the assumption of abundance. Much AI planning quietly assumes compute will keep getting cheaper and more plentiful. If domestic build stays slow and global demand keeps climbing, that assumption is a bet, not a baseline.
The third is the delay compounding. Every year a piece of infrastructure waits in a queue, the AI workload it was meant to serve has moved further ahead. Slow build is not a fixed cost; it is a widening gap.
Hidden Cost: The cheapest AI capability today can carry the highest long-run cost if it depends on capacity that Britain is not building, priced by a market the UK does not control. Today’s inference bill is the smallest line in that sum.
The fourth is misreading the fix. “Just move faster” can become an argument for cutting the wrong corners. The corner worth cutting is planning delay on infrastructure. The corner not worth cutting is the governance credibility that gives British AI standing in the first place.
The takeaway for UK business
Richards’ warning is aimed at government, and rightly so, because the biggest levers here are national ones: energy, grid, and planning. But the column contains a lesson that does not require waiting for Whitehall. The distinction between makers and takers is already shaping the cost and availability of AI in Britain, and firms that treat compute as an infinite, location-free utility are the ones most exposed when that turns out to be false.
Three things matter most. Know where your AI capability physically depends, and do not let that dependency concentrate by accident. Build for portability so that scarcity in one place is a migration rather than a crisis. And read the national infrastructure story as a business signal, because the speed at which Britain builds power and compute will show up in your cost base whether or not you were watching.
The Red Queen framing is dramatic, and the panic Richards wants to inject is partly rhetorical. The steadier truth underneath it is that Britain’s position is still open. The country can be a shaper of this era rather than a renter of it, but only if the building starts now, and only if businesses stop inheriting their AI exposure by default and start choosing it deliberately. For more on how AI policy and infrastructure land on UK organisations, explore our insights and latest news, or get in touch with a tip or a question.
Analysis based on the opinion column by Sam Richards for City AM, “Britain must speed up to survive the AI era” (1 July 2026). Strategic interpretation and UK business context by Resultsense.