Datacentre Bottleneck Threatens Global AI Scaling as Half of Planned Projects Face Delays or Cancellation
Key Takeaways
- ▸Approximately 50% of 250 announced mega-datacentre projects (100MW+) will be delayed or cancelled, creating a critical infrastructure bottleneck for AI development
- ▸Google's cloud AI business is compute-constrained due to insufficient datacentre capacity, signaling acute pressure across the industry
- ▸Energy demands from new datacentres would consume 1.3% of global electricity in 2025, with supply chain disruptions, local opposition, and environmental concerns as primary obstacles
Summary
The global AI industry faces a critical infrastructure crisis as approximately half of 250 planned mega-datacentre projects are expected to be cancelled or delayed, according to the Uptime Institute. These projects, announced between 2021 and 2024, were meant to provide the computational power necessary for training and deploying increasingly sophisticated AI models. Major cancellations include Project Range in Arizona, Cyberjaya campus in Malaysia, and Virginia's Prince William Digital Gateway, which was halted due to proximity to a Civil War battlefield alongside environmental and energy concerns.
The delays pose immediate problems for leading AI companies that depend on datacentres to operate their services. Google has explicitly stated that its cloud AI business is now "compute-constrained" due to insufficient datacentre capacity. The infrastructure bottleneck creates a widening gap between the rapid advancement of AI models from companies like OpenAI, Anthropic, and Google and the ability to scale these models into production systems.
Multiple factors are stalling projects, including inexperienced developers without committed tenants, massive energy and water consumption requirements, global supply chain disruptions affecting chip sourcing, and mounting local community opposition. The planned projects would require unprecedented power consumption—new datacentres announced in 2024 alone would consume 1.3% of the world's projected electricity usage in 2025, nearly doubling current datacentre demand, with approximately 80% of new power demand concentrated in US projects.
The Uptime Institute has identified six mega-gigawatt projects (5GW+ each) in development, a scale equivalent to Ireland's entire peak energy demand. Without resolution of supply chain constraints and permitting challenges, the AI industry may face severe compute limitations that could slow model development, limit commercial AI service availability, and widen the competitive gap between well-funded incumbents and emerging players.
- Six mega-gigawatt (5GW+) projects are in development, but completion timelines remain uncertain despite urgent industry need
Editorial Opinion
The datacentre infrastructure crisis represents a fundamental threat to AI's trajectory. While AI model capabilities are advancing exponentially, the physical infrastructure to deploy them is lagging dangerously behind—a rare case where hardware, not algorithms, becomes the limiting factor. This bottleneck disproportionately favors tech giants with existing datacentre portfolios and capital reserves, potentially consolidating AI power in fewer hands. Without decisive action on energy infrastructure, supply chains, and environmental permitting, the next generation of AI innovation may be defined not by breakthrough research, but by rationed compute.
