The global AI datacenter buildout is accelerating toward roughly 50 gigawatts of new capacity coming online in 2027 outside China, supported by an advancing pipeline of more than 24 gigawatts across about 40 campuses, with many sites already partially completed.

SemiAnalysis projects about 50 gigawatts of AI datacenter capacity in 2027, nearly doubling prior estimates and signaling an unprecedented surge in hyperscaler construction and leasing activity.

Forecasts indicate that roughly 50 gigawatts of new AI datacenter capacity will come online in 2027, approaching the industry’s total added capacity through 2024.

Developers are increasingly deploying their own power solutions on-site—fuel cells, refurbished turbines, and engines—to bypass grid delays and mitigate reliability issues.

Construction momentum is driven by record starts and over $1 trillion in signed binding lease value, with modular design and development techniques cutting timelines to under 10 months.

Despite challenges like labor shortages and supply chain delays, innovations in design, power, and modular construction are helping to accelerate rollout and mitigate bottlenecks.

At the same time, risks such as labor shortages, supply chain delays, fundraising hurdles, and grid limits are noted, though data still show accelerations in large-site completions and deliveries.

Leasing activity has broadened beyond Microsoft, with Oracle, Meta, AWS, Google, Nvidia, AMD, Anthropic, and OpenAI among the major players fueling debt financing for new builds.

Market sentiment shows mixed expectations, with Polymarket traders assigning about 9% odds to an AI industry downturn by year-end.

Nvidia chips are expected to remain in strong demand as developers navigate construction and electrical delays, supporting the AI data-center expansion.

Recent progress includes completed or near-complete facilities by Crusoe, AWS, QTS, Microsoft, and others, illustrating rapid on-site development and deployment of compute infrastructure.

A faster compute supply could lower AI costs, speed model progress, and require higher capital expenditure, potentially pushing yields and rates higher for investors and vendors.