3 min read

UK AI growth may hinge on fixing data center power

Rising electricity costs and grid bottlenecks could push UK data center projects overseas just as AI demand surges.

Image: TechRadar

The UK’s push to become the fastest AI-adopting country in the G7 could run into a more basic problem: power. According to a report by Oxford Economics for the Nuclear Industry Association, rising grid constraints and uncompetitive industrial electricity prices risk sending data center investment overseas just as Britain needs more capacity.

That pressure is building fast. Data center energy demand could increase fivefold by 2035, while operators are also under pressure to secure cleaner electricity in line with climate targets. Even if AI data centers are prioritized for new grid connections, that only addresses demand-side access, not the wider issue of expensive electricity and limited network capacity.

Data center power demand is outpacing supply

The mismatch is stark: data centers can be built in 12–14 months, but large renewable energy projects may take 12–14 years to come online. At the same time, Britain faces rising non-commodity electricity costs including Transmission Network Use of System (TNUoS) and Nuclear Regulated Asset Base (RAB) charges.

The article notes that Ofgem has warned data center electricity use could significantly exceed Britain’s current peak power consumption, raising the risk of even higher costs.

One proposed answer is a broader shift toward flexible electricity contracts, which let operators buy power in chunks rather than lock all consumption into a fixed rate. That matters for AI data centers, whose energy use can be more volatile than conventional cloud facilities. Flexible deals can help operators hedge only part of their usage, then buy the rest on the spot market when prices are favorable.

Recommended reading

Infinity lands $15M to make AI chips easier to use

Examples cited in the piece include models such as:

  • tolerance banding, which guarantees a set price within a defined range of consumption
  • contracts that price electricity above or below expected demand more dynamically
  • cash-out arrangements that let buyers lock in energy when prices are low and purchase additional power days or months ahead

CPPAs and credit barriers

The article also points to Corporate Power Purchase Agreements (CPPAs) as a way for data centers to secure long-term, traceable green electricity at more stable prices. In some cases, private-wire PPAs with onsite generation could even let facilities sell excess power back to the grid while avoiding some grid-related charges. The piece says those non-commodity costs, including TNUoS, make up 60% of electricity bills and are expected to rise further.

CPPAs may also help support the UK target to cut operational emissions from buildings by 76%. But they come with hurdles. These agreements are often complex, can require strong credit backing, and may lock customers into 10–15 year commitments. The article argues that this is especially difficult for startups and smaller data center operators, whose revenues are typically spread across many customers rather than anchored by a few large long-term contracts.

Possible workarounds include security deposits, bank guarantees, parent-company guarantees, and intercompany guarantees from larger customers such as Amazon. The piece also highlights PPA import sleeving contracts as a way to integrate pre-purchased power into a facility’s existing grid supply.

The broader argument is simple: if Britain wants sovereign AI capacity and sustained data center growth, it needs easier access to secure, affordable, low-carbon energy — and faster, less restrictive contract models to get there.

Marcus Vance

Enterprise Editor

Marcus follows the money. He covers enterprise software, cloud architecture, and the tectonic shifts in Big Tech strategy. He translates dense earnings calls and complex M&A activity into actionable insights about where the industry is actually heading. If a tech giant makes a silent pivot, Marcus is usually the first to notice.

via TechRadar

// Keep reading