AI data-centre build-out hits equipment snag
What happened
America’s push to build more AI data centres is slowing after shortages of critical electrical equipment — much of it sourced from China — raised supply risks and cost uncertainty. The bottleneck, coupled with tariff and trade worries, is forcing firms to treat capacity and hardware availability as genuine business constraints rather than distant engineering problems. That changes system-design conversations about regional failover, capacity planning and cost-aware rollouts. (firstpost.com) (cnbc.com)
Why it matters
Analysts at Sightline Climate warn that 30–50% of the data‑center capacity announced for 2026 may be delayed or canceled, leaving a large portion of the roughly 16 gigawatts of announced capacity at risk and only about 5 gigawatts actually under construction. (sightlineclimate.com) One high‑scale example: an eight‑building campus in Abilene, Texas that will be used by OpenAI is expected to draw about 1.2 gigawatts of power — roughly the amount needed to run nearly 1 million U.S. homes — at the same time that cloud companies plan to spend more than $650 billion on AI infrastructure this year, putting unprecedented strain on supply chains. (bloomberg.com) The immediate bottleneck is in electrical hardware: transformers (machines that raise or lower voltage so transmission lines can feed a site safely), switchgear (control and protection panels that connect and isolate electrical circuits), and batteries (devices that store energy for short‑term backup); lead time — the delay between placing an order and receiving the equipment — for large transformers has stretched from roughly two years before 2020 to as long as five years in some cases, while many AI campuses expect to deploy inside a 12– to 18‑month window. (eepower.com, sightlineclimate.com) Because U.S. factories cannot meet that surge, builders have turned to global suppliers; Bloomberg reports imports of high‑power transformers from China rose sharply (from under about 1,500 units in 2022 to more than 8,000 in 2025), and at the same time new tariff policy and trade uncertainty have added price and sourcing risk for those imports. (bloomberg.com, taxfoundation.org) Hyperscalers are responding by changing project and power strategies: some are buying or investing in generation and storage pipelines to secure energy directly (Google’s purchase of Intersect Power’s pipeline and Amazon’s investments in solar and batteries are examples), while others are trimming or reprioritizing leases to favor sites with more stable local grid plans — moves that shift where and how AI capacity gets built. (sightlineclimate.com, semafor.com)
Key numbers
- (sightlineclimate.com) One high‑scale example: an eight‑building campus in Abilene, Texas that will be used by OpenAI is expected to draw about 1.2 gigawatts of power — roughly the amount needed to run nearly 1 million U.S.
- homes — at the same time that cloud companies plan to spend more than $650 billion on AI infrastructure this year, putting unprecedented strain on supply chains.
What happens next
- Analysts at Sightline Climate warn that 30–50% of the data‑center capacity announced for 2026 may be delayed or canceled, leaving a large portion of the roughly 16 gigawatts of announced capacity at risk and only about 5 gigawatts actually under construction.
- (sightlineclimate.com) One high‑scale example: an eight‑building campus in Abilene, Texas that will be used by OpenAI is expected to draw about 1.2 gigawatts of power — roughly the amount needed to run nearly 1 million U.S.
- homes — at the same time that cloud companies plan to spend more than $650 billion on AI infrastructure this year, putting unprecedented strain on supply chains.
Quick answers
What happened in AI data-centre build-out hits equipment snag?
America’s push to build more AI data centres is slowing after shortages of critical electrical equipment — much of it sourced from China — raised supply risks and cost uncertainty. The bottleneck, coupled with tariff and trade worries, is forcing firms to treat capacity and hardware availability as genuine business constraints rather than distant engineering problems. That changes system-design conversations about regional failover, capacity planning and cost-aware rollouts. (firstpost.com) (cnbc.com)
Why does AI data-centre build-out hits equipment snag matter?
Analysts at Sightline Climate warn that 30–50% of the data‑center capacity announced for 2026 may be delayed or canceled, leaving a large portion of the roughly 16 gigawatts of announced capacity at risk and only about 5 gigawatts actually under construction. (sightlineclimate.com) One high‑scale example: an eight‑building campus in Abilene, Texas that will be used by OpenAI is expected to draw about 1.2 gigawatts of power — roughly the amount needed to run nearly 1 million U.S. homes — at the same time that cloud companies plan to spend more than $650 billion on AI infrastructure this year, putting unprecedented strain on supply chains. (bloomberg.com) The immediate bottleneck is in electrical hardware: transformers (machines that raise or lower voltage so transmission lines can feed a site safely), switchgear (control and protection panels that connect and isolate electrical circuits), and batteries (devices that store energy for short‑term backup); lead time — the delay between placing an order and receiving the equipment — for large transformers has stretched from roughly two years before 2020 to as long as five years in some cases, while many AI campuses expect to deploy inside a 12– to 18‑month window. (eepower.com, sightlineclimate.com) Because U.S. factories cannot meet that surge, builders have turned to global suppliers; Bloomberg reports imports of high‑power transformers from China rose sharply (from under about 1,500 units in 2022 to more than 8,000 in 2025), and at the same time new tariff policy and trade uncertainty have added price and sourcing risk for those imports. (bloomberg.com, taxfoundation.org) Hyperscalers are responding by changing project and power strategies: some are buying or investing in generation and storage pipelines to secure energy directly (Google’s purchase of Intersect Power’s pipeline and Amazon’s investments in solar and batteries are examples), while others are trimming or reprioritizing leases to favor sites with more stable local grid plans — moves that shift where and how AI capacity gets built. (sightlineclimate.com, semafor.com)
