Powering the AI Era - A briefing by the Carbon Trust, The DCA (Data Centre Alliance) and NZIH

1. Grid capacity and connection queues

Demand for new connections is rising, yet contracted capacity and actual consumption can differ substantially, particularly as campuses are built in phases. Participants disagreed about how much apparent underuse represents inefficient allocation. They agreed that clearer evidence on project maturity, ramp rates, and the basis for connection requests would help distinguish credible demand from speculative capacity reservations and make it easier to plan carbon-optimised infrastructure. Any reform must account for the fact that projects genuinely take years to build.

Developments to watch: Whether connection reform introduces more consistent evidence requirements, staged capacity allocation or incentives to release unused capacity. And how these affect projects with different funding and delivery profiles.

2.Flexibility and shared infrastructure

Data centres may be able to shift some workloads, use batteries, briefly reduce demand, or provide grid services. Options differ markedly by workload and contractual uptime obligation. Participants saw technical possibilities but questioned whether current revenues compensate operators for operational risk. They also discussed community solar and storage, less-firm connections, and sharing capacity during construction, noting that new commercial models may be needed to unlock these opportunities.

Developments to watch: Demonstrations that quantify dependable flexibility, build customer acceptance and develop commercial models that provide appropriate incentives/value exchange for all parties.

3. On-site generation and emissions

Delays to grid connections may encourage greater use of on-site power and microgrids, including on-site gas generation as an interim power source. Some participants at the event raised concerns that what is intended as a temporary investment could become a long-lived asset, extending fossil fuel use beyond the connection delay it was meant to bridge. Participants did not agree on a response, although they put forward efficient use of existing grid capacity and alternative connection products for earlier consideration.

Developments to watch: Whether planning, connection and procurement arrangements begin to specify the emissions performance, operating limits and exit route for temporary generation.

4. AI workloads and the UK’s role

Training and inference have different power, location, and latency needs. Several speakers questioned whether the UK’s electricity costs and infrastructure investment conditions favour very large training campuses, while recognising the potential for inference nearer users. The discussion exposed a more basic evidence gap: neither public planning nor energy forecasts yet provide a sufficiently clear picture of the scale, location and behaviour of future AI workloads.

Developments to watch: Better distinctions between training, inference and other digital demand in national forecasts, investment decisions and spatial planning.

5. Planning and community value

Data centre decisions span digital policy, energy, water, land use and local planning. Participants described fragmented information and uneven capacity within public bodies. They explored a clearer planning pathway, potentially including a dedicated subclassification under B8, with early disclosure of infrastructure needs and community engagement.

Participants also questioned whether the sector sufficiently understands what communities themselves want and need from data centre development.  Discussions explored how local and spatial planning could consider digital infrastructure from the outset, with community needs and potential local benefits weighed alongside energy, water, and land requirements. The discussion of community value went beyond jobs alone, including local energy, amenities, and possible benefit mechanisms.

Developments to watch: A coordinated planning framework that brings grid, water and community impacts into view earlier, with a clearer account of how benefits will be delivered over a facility’s lifetime.

6. Innovation from pilot to deployment

Promising technologies often struggle to reach an operator’s reference design. First deployment carries performance risk, while innovative hardware firms need patient finance and a credible customer base. Participants proposed testing existing prototypes at a relevant scale, pooling demand signals, and working with design and procurement teams early. Standard specifications, including diesel backup requirements, may otherwise preserve familiar solutions even where alternatives could perform well.

Developments to watch: Shared demonstrations and procurement changes that turn validated low-carbon technologies into repeatable designs, supported by measurable performance evidence.

7. Energy, water and transparent reporting

Cooling choices can trade water use against electricity consumption, so a single metric may obscure the overall impact. Participants favoured better comparable evidence on energy sources, water sources, efficiency and grid interaction, while leaving the form of disclosure open. Transparent reporting can also give communities a clearer basis on which to judge local proposals and operating performance.

Developments to watch: Reporting standards that disclose energy and water together, distinguish design claims from measured performance and explain local context.

From evidence to action

Sustainable growth will depend on making better decisions earlier: testing the credibility of demand, matching connection capacity to delivery, understanding when flexibility is commercially viable, and assessing energy, water and community impacts together. It will also require innovation beyond technology, including new commercial models, planning approaches, and ways to share infrastructure and create local value.

Better evidence can help government, networks, industry, and communities make those decisions with greater confidence. The opportunity now is to use that evidence to reduce risk, test new approaches, and help promising solutions move from concept to deployment.

Source: Powering the AI Era event insight summary, 15 September 2026. Discussion points are non-attributed under the Chatham House Rule.

This briefing was prepared jointly by the Carbon Trust, DCA and NZIH.