Grid Bottlenecks Force AI Datacenters Onto Gas Power, Jeopardizing Cloud Carbon Accounting

A new investigation by tech justice non-profit Foxglove has revealed that two major planned datacentres in England—Wapseys Wood in Buckinghamshire and Quest Park in Bedfordshire—will generate over 4.5 million tonnes of carbon emissions annually once fully operational. Together requiring 1.3 GW of power capacity, the operators of these facilities plan to construct dedicated on-site gas-fired generation units to circumvent years-long queues for national electricity grid connections. The projected annual emissions from these two AI and cloud hosting sites alone will surpass the entire UK operational footprint of major fossil fuel producers. This development marks a critical shift for enterprise cloud practitioners and sustainability leads. For years, cloud migrations have been justified partly by hyperscaler efficiency gains and clean power purchasing agreements (PPAs). However, the explosive computational demand of generative AI model training and inferencing has created severe local grid congestion. When datacenters deploy bridge fossil fuel solutions like on-site gas turbines to achieve faster time-to-market, cloud consumers inadvertently inherit substantial Scope 3 emissions increases that undermine corporate ESG commitments and climate disclosures. This incident reflects a broader global conflict between rapid AI infrastructure expansion and decarbonization mandates. Across major cloud hubs in North America and Western Europe, power utility interconnect backlogs now span five to eight years, while datacenter capacity demand continues to surge. Consequently, the assumption that cloud workloads are automatically running on a decarbonizing grid is no longer guaranteed. While cloud providers continue to buy virtual renewable energy certificates (RECs), physical operations in power-constrained regions increasingly rely on localized carbon-heavy generation during peak periods. For DevOps, platform engineers, and enterprise architects, this reality necessitates an immediate pivot toward operational carbon awareness. Teams can no longer treat cloud carbon intensity as static across an entire provider's fleet. Instead, engineering organizations must implement carbon-aware scheduling tools, routing non-latency-sensitive batch AI jobs and data pipelines to regions with verifiable 24/7 carbon-free energy (CFE). Furthermore, procurement teams must demand rigorous, granular location-based emissions telemetry from colocation providers and hyperscalers rather than relying on annual market-based carbon offsets.