AI is a Global Arms Race, Is the UK Playing to Win?
The race for AI dominance isn’t coming. It’s already here.
The US is pouring $500 billion into AI infrastructure. Saudi Arabia is investing $250 billion in AI infrastructure. The Middle East is projected to capture $320 billion in AI-driven economic benefits by 2030.
The UK? £14 billion. It’s betting on AI Growth Zones, 500MW of power per site, fast-tracked planning rules, and AI-ready data centres, to stay in the game. 
Science Secretary Peter Kyle calls it a transformative plan:
“These new AI Growth Zones will deliver untold opportunities, sparking new jobs, fresh investment and ensuring every corner of the country has a real stake in our AI-powered future.”
But AI is no longer about building data centres faster than the next country. Compute power is just one part of a highly complex, interdependent infrastructure challenge, and Britain risks making the wrong bet. While policymakers focus on electricity and land use, the real AI superpowers are building something more valuable: a system where workloads move at the speed of intelligence, not the pace of bureaucracy.
The Real AI Bottleneck: Not Power, But Data Mobility
The government’s approach assumes AI workloads will be static, housed in massive data centres, consuming local grid capacity. But AI is not like traditional computing. Its biggest challenge is not where to train and store models, but where to run them for commercial and strategic benefit, and today’s networks are not designed for that level of mobility.
AI inference workloads are bursty, latency-sensitive, and need to be processed close to real-time demand. Hyperscale cloud providers, AWS, Azure, Google, are already proving inadequate for distributed AI. Their pricing models penalize moving data, their network architectures introduce unpredictable delays, and their business incentives are built around locking customers into centralised compute locations rather than enabling flexible, dynamic AI processing.
If the UK wants to lead in AI, it needs to start thinking beyond the data centre. Platforms like Stelia offer a preview of what’s possible, a model where AI workloads can move to wherever the infrastructure conditions are most favourable. It’s not enough to have power; the ability to shift workloads dynamically based on energy pricing, bandwidth availability, and demand spikes will determine who wins this race.
No Fibre, No AI: The UK’s Forgotten Infrastructure Gap
Electricity is a prerequisite for AI, but without high-speed fibre connectivity, it is useless. Many of the UK’s best locations for AI Growth Zones are chosen based on their access to 500MW+ of power, yet these same sites, especially those near intermittent wind, solar, and battery storage, lack the fibre density required to support AI inference at scale.
Wind farms, for instance, may generate massive amounts of surplus electricity during off-peak hours, but AI workloads cannot be routed there if there is no data infrastructure to support them. The government has recognized that energy capacity needs to scale, but it has not accounted for the fact that AI’s true bottleneck is moving petabytes of data, not just securing megawatts of power.
The Middle East is already solving this problem. Saudi Arabia’s Vision 2030 infrastructure projects are integrating AI-ready fibre networks directly into their energy and smart city developments. The UK, by contrast, is planning its AI expansion without upgrading its digital backbone. Without immediate investment in fibre infrastructure, AI Growth Zones will be dead on arrival.
AI Should Be an Energy Asset, Not a Liability
Right now, AI is being framed as a power problem. However, data centres are seen as enormous energy drains, competing with homes and businesses for grid capacity. But this is the wrong way to think about AI’s energy impact.
In Texas, AI compute is already being used to stabilize the grid rather than overload it. Data centres there are integrating into demand-response programs, adjusting AI workloads dynamically based on real-time grid conditions. When renewables are generating surplus power, compute ramps up. When demand on the grid spikes, AI workloads are temporarily shifted elsewhere.
The UK has an opportunity to take this idea further. AI should not be a static infrastructure burden, it should be designed as a grid asset that flexes with the availability of energy. This requires more than just power supply, it demands an AI ecosystem capable of routing workloads to the cheapest and most sustainable energy sources in real time.
That vision, however, is entirely absent from the UK’s AI Growth Zone strategy. The current plan treats AI infrastructure as a fixed demand load, rather than as something that can be intelligently optimized to balance national energy resources.
The UK is Falling Behind While Others Surge Ahead
The government is focused on accelerating data centre development, but the countries truly leading in AI are building the full-stack infrastructure to support inference at scale.
- The US has Project Stargate, a $500 billion AI infrastructure initiative designed to build next-generation computing networks.
- Saudi Arabia is investing $250 billion into AI-driven cities, energy, and data hubs, ensuring that AI workloads can be dynamically allocated where conditions are best.
- The Middle East as a whole is projected to capture $320 billion in AI economic benefits by 2030.
- The UK’s £14 billion investment, by comparison, looks insufficient to position itself as a true AI superpower.
Right now, the UK is investing in static infrastructure while its competitors are investing in AI mobility and optimisation. The AI Growth Zones plan might succeed in attracting data centre investment, but if those data centres are locked into rigid, power-hungry locations, they won’t be able to compete with the global AI infrastructure emerging elsewhere.
The UK Needs an AI Infrastructure Strategy Fit for the Future
If the UK wants to be a leader in AI, it needs more than data centres. It needs an AI infrastructure strategy that is designed for inference, not just training. That means:
- Workload mobility. AI compute must be able to move in real time based on power availability, network conditions, and cost.
- AI-native networks. Fibre expansion must be prioritised in locations where AI workloads will run, not just in existing urban centers.
- Energy-aware AI. Compute should flex with grid conditions, absorbing excess renewable energy and shifting dynamically when demand spikes.
The UK has a choice. It can continue investing in infrastructure that will be outdated before it is fully operational. Or it can hack the system, using AI-driven workload mobility to sidestep grid constraints, optimise power use, and position itself as a true AI superpower.
This isn’t a theoretical problem. It’s already happening.
The world isn’t waiting. Why is the UK?
