When AI and a brewery compete for the same electricity, who wins? Dr Phil Wu, the founder and director of Absolar Solutions, a commercial solar and battery storage company, shines the spotlight on the energy allocation question that the AI boom is quietly forcing onto the agenda.
Somewhere in the United Kingdom, a planning application is being approved for a new data centre. It will consume more electricity than a small town. It will run continuously, every hour of every day, regardless of what else is happening on the grid. And it will almost certainly have been designed, permitted, and brought to operation faster than any new power station built to serve it.
This is not a criticism of data centres, or of the AI infrastructure they support. It is an observation about pace. And pace, in energy systems, is where problems tend to begin.
A different kind of demand
Every major technological shift in history has changed the shape of energy demand. Electrification, mass manufacturing, and the rise of computing each placed new pressures on infrastructure that was not originally designed to accommodate them. In each case, the system adapted, eventually.
The growth of AI infrastructure is different in one important respect: its speed relative to the infrastructure required to support it. A hyperscale data centre can be designed, approved, and operational within two to three years. A new nuclear power station takes fifteen or more. A large offshore wind farm, including grid connection, rarely takes fewer than seven. Even solar, the fastest-deploying form of generation available, is constrained by a UK grid connection queue that now stretches years into the future for many applicants.
The result is a structural mismatch. Demand, driven by software decisions made in boardrooms and data labs, is scaling faster than the physical infrastructure required to meet it. That gap has to be absorbed somewhere. The question is where.
The allocation problem nobody is talking about
Consider a scenario that is already playing out across regional grids. A food and drink manufacturer — a brewery, say, or a dairy — and a new data centre are both drawing from the same local network. Capacity is under pressure. Both are legitimate businesses. Both employ people and contribute to the economy. But their energy profiles could hardly be more different.
The brewery operates on tight margins, employs people across a range of skill levels, uses locally sourced ingredients, and produces goods that sit in supply chains serving retailers, pubs, and restaurants across the country. Its energy demand is substantial — refrigeration, production, packaging — but bounded, and linked to physical output. The data centre operates on high margins, employs relatively few people per megawatt consumed, and provides services that are economically valuable but, for most users, not immediately essential to daily life. Its energy demand is large, constant, and growing.
If grid access becomes genuinely constrained and a price-based rationing mechanism takes effect, capital wins almost by definition. The data centre, backed by a balance sheet that can absorb higher energy costs or negotiate long-term contracts that smaller users cannot, secures its supply. The brewery faces the residual: higher costs, less certainty, and eventually, questions about whether it can remain economically viable on a grid it no longer has reliable access to.
This is not a hypothetical edge case. Industrial location decisions in the UK and across Europe are already being shaped by grid access. Businesses that require reliable, large-scale power are asking questions about availability before they ask about almost anything else. Energy is quietly becoming a constraint on economic geography in ways that have not been fully acknowledged in public policy.
The counterarguments, taken seriously
There are reasonable responses to this concern, and they deserve to be engaged with honestly.
The most compelling is that AI infrastructure, over time, increases productive capacity across the whole economy. If AI genuinely delivers on its potential to optimise industrial processes, reduce waste, and improve supply chain efficiency, the net effect on energy demand per unit of economic output could be positive. The data centre consuming power today might be enabling the smart factory that consumes considerably less of it tomorrow.
A second argument is that demand response technology could allow large digital loads to flex in ways that traditional data centres could not. If AI workloads that are not time-critical can be shifted to periods of surplus generation, data centres could in theory become assets to grid stability rather than threats to it.
Both of these are plausible medium-term outcomes. The difficulty is that the constraint being described is a near-term reality, and medium-term solutions do not help a manufacturer facing an energy bill that has moved beyond what its margins can absorb.
The question that needs asking
Energy has always been allocated by markets, tempered by regulation. That system has worked adequately in conditions where demand grew incrementally and infrastructure kept reasonable pace with it. Those conditions are changing.
What is not yet happening, at least not openly, is a societal conversation about priorities. When grid capacity is genuinely finite and demand from high-capital, low-employment technology users competes directly with demand from lower-margin, higher-employment traditional industries, how should that be resolved? By price alone? By policy intervention? By physical allocation?
These are not comfortable questions. They involve trade-offs between economic growth, industrial policy, employment, and food security that governments tend to prefer not to make explicit. But the absence of an explicit framework does not mean the allocation is not happening. It means it is happening by default, through market mechanisms that were not designed with this scenario in mind.
What is not yet happening, at least not openly, is a societal conversation about priorities. When grid capacity is genuinely finite and demand from high-capital, low-employment technology users competes directly with demand from lower-margin, higher-employment traditional industries, how should that be resolved? By price alone? By policy intervention? By physical allocation?
Dr Phil Wu, founder and director of Absolar Solutions
What industry can do while policy catches up
Solar will not power a hyperscale data centre. The load is too large, too constant, and the intermittency of generation is fundamentally incompatible with the uptime requirements of critical digital infrastructure. That much is straightforward.
But the more relevant question is not whether solar can serve the data centre. It is whether solar can help the brewery, the dairy, and the logistics hub reduce their dependence on the grid that the data centre is increasingly competing for.
For commercial and industrial users, a well-specified rooftop solar installation with battery storage can realistically offset 30 to 60 percent of grid consumption. That is not energy independence, but it is a meaningful reduction in exposure to the price volatility and capacity constraints that are becoming defining features of the UK grid. A food manufacturer generating a significant share of its own power is, in practical terms, removing itself from competition with larger, better-capitalised users for that portion of its demand.
Scaled across the commercial and industrial base — and the adoption gap here remains enormous, with fewer than 4 percent of UK buildings currently carrying any solar installation — distributed generation of this kind represents a material reduction in aggregate grid pressure. It does not solve the structural mismatch between AI-driven demand growth and generation capacity. But it creates headroom. And headroom, in a constrained system, matters.
The brewery and the data centre are both legitimate. Both contribute to the economy in ways that are real, if different in character. The question of who gets priority when the grid cannot adequately serve both is one that deserves a considered answer from policymakers. In the meantime, the most practical thing traditional industries can do is reduce how much of that question applies to them.
—
