Infrastructure shocks change siting choices

OpenAI has paused its Stargate data-center plan in Britain after saying the United Kingdom’s power costs and rules do not support the project. (reuters.com) Reuters reported on April 9 that OpenAI halted its main United Kingdom data-center effort over an “unfavourable regulatory environment and high energy costs.” Bloomberg reported the same day that the company still plans to revisit the project if conditions improve. (reuters.com) (bloomberg.com) The project had been tied to Britain’s push to attract artificial intelligence infrastructure, and Politico Europe reported it was announced during Donald Trump’s September 2025 state visit to the United Kingdom. The pause leaves the government of Prime Minister Keir Starmer without one of the marquee investments linked to that campaign. (politico.eu) A data center is a warehouse full of chips, networking gear, and cooling equipment, and the power bill can decide where it gets built. The Next Web reported that industrial electricity prices in Britain are about four times higher than in the United States, while unresolved copyright rules for artificial intelligence also remain a hurdle. (thenextweb.com) At the same time, Amazon Web Services says demand is running so hot that some customers want to reserve huge chunks of supply before anyone else can. In his April 9 shareholder letter, Amazon chief executive Andy Jassy said two large Amazon Web Services customers asked to buy all of the company’s Graviton processor capacity for 2026, and Amazon refused. (aboutamazon.com) Jassy also said Amazon’s Trainium 3 chip started shipping at the start of 2026 and is already “nearly fully subscribed,” according to coverage of the letter by CIO and Data Center Dynamics. Amazon is planning about $200 billion in capital spending in 2026, with much of it aimed at artificial intelligence infrastructure. (cio.com) (datacenterdynamics.com) (cnbc.com) Google and Intel moved in the same week to secure another part of that stack: the chips that run cloud services and route traffic inside giant server farms. On April 9, the companies announced a multiyear deal to keep Intel Xeon processors in Google Cloud and to expand joint work on custom infrastructure processing units. (newsroom.intel.com) Intel said those infrastructure processing units are custom chips built to improve efficiency and utilization at scale, which means they handle the plumbing around artificial intelligence systems rather than the model itself. The announcement framed modern artificial intelligence as a “heterogeneous” system, with central processing units, accelerators, and networking chips all sharing the load. (newsroom.intel.com) (intc.com) The split is now clear across the market: one country lost a flagship siting bid over energy and policy, one cloud provider is rationing scarce capacity, and two chip suppliers are locking in longer partnerships. In artificial intelligence infrastructure, location, electricity, and component supply are starting to matter as much as software. (reuters.com) (aboutamazon.com) (newsroom.intel.com)