A power-first approach to data centre strategy

David Knox, Global Director of Energy & Sustainability at Colt Data Centre Services, explains why grid constraints are forcing developers to rethink where, when and how new capacity is built.

The UK government revealed that the queue for connections to the UK power grid has grown by 460% in just six months. This rise is linked to a significant increase in speculative applications. Demand from AI is rising faster than the grid can deliver.

This is fundamentally changing the approach to data centre design and location selection. Access to power has become a deciding factor, shaping when and where sites can be built, how projects are phased and which models offer long-term value.

Success depends on preparing for uncertainty and reflecting this in design, prioritising flexibility while adopting a power-first approach to location and delivery.

The new reality of grid access

AI workloads are materially reshaping what typical power requests look like. The compute required for AI data centres demands several times the power density of traditional data centres, and the number of operators seeking access to this level of capacity is growing.

At the power generation level, most countries have sufficient capacity to support the growth of AI data centres. However, demand often exceeds what local power networks can accommodate. Combined with growing connection queues, these local constraints are becoming a defining challenge, creating delays and adding complexity to the development and operation of AI data centres.

Grid access is no longer a straightforward connect-and-expand process. It’s an ongoing negotiation that needs to be managed early because it now affects design assumptions and delivery risks from the outset.

Flexibility first

Organisations must adapt how they approach grid access for AI data centres and prioritise flexibility amid uncertain timelines and outcomes. One way to achieve this is through practical design and delivery choices, particularly modular power systems and hybrid supply models.

Modular power systems support phased construction and energisation, allowing providers to respond more effectively if grid delays occur. Hybrid supply models can also introduce additional power sources, reducing reliance on the grid and improving resilience.

What was once viewed primarily as a failsafe or temporary measure may now have a role as a capacity-bridging tool. Hybrid supply models can help sites operate while grid upgrades catch up, with on-site power providing additional capacity to support growing AI compute requirements.

There are, of course, trade-offs to consider when implementing hybrid supply models, including cost, operational complexity and sustainability objectives. However, in a period of constrained grid access and growing power demand, developers should consider these approaches as potential options for supporting AI-led growth.

Intentional location selection

For AI data centres, where competitiveness depends on the certainty of power delivery and the ability to scale quickly, developers are being forced to rethink location strategy from the ground up.

Historically, site selection has followed a familiar pattern, focusing on proximity to users, strong connectivity and suitable land. Power was largely assumed to be available through a standard grid connection process. That assumption no longer holds.

In many established data centre hubs, grid capacity is constrained or already fully allocated, and sites that appear ideal on paper can conceal significant delivery risks. These may include long connection delays, complex network constraints or substantial and unexpected reinforcement costs.

As a result, the strategic shift is clear. Developers must now balance traditional considerations such as latency, network ecosystem and market access against the availability of reliable power and the ability to scale. In the AI era, the strongest location decisions are those built around speed to power, certainty of delivery and long-term scalability.

Preparing for the unknown

AI compute demand is growing beyond the capacity that local grids were designed to support. Connection queues are growing rapidly, while local network constraints continue to be a barrier to entry.

Power has become a primary determinant of the successful delivery of AI data centres. Developers must therefore lead with a power-first approach to help ensure projects are delivered on time, with stable funding and in line with customer demand.

Grid strategy must be at the forefront of the site selection process and concept design, with developers engaging proactively with delivery partners and network operators rather than leaving it until later.

To be resilient, power strategy needs to be rooted in flexibility. This can be achieved through modular power systems, hybrid supply models or designs that can adapt to different grid outcomes.

Location selection must be integrated with a power-first model, recognising that speed to power, certainty of delivery and scalability are critical.

The developers that succeed in the era of AI will be those that understand the grid rather than trying to avoid it. Building programmes and sites that can secure power, adapt to change and scale will be central to that success.

Categories

Related Articles

More Opinions

It takes just one minute to register for the leading twice weekly B2B newsletter for the data centre industry, and it's free.