Marex and Oxford Smith School find multi-year gap in AI data centre capacity delivery
Updated
Updated · marex.com · Apr 27
Marex and Oxford Smith School find multi-year gap in AI data centre capacity delivery
11 articles · Updated · marex.com · Apr 27
The research highlights delays of 2–3 years in favourable regions and over 7 years in constrained areas, with hyperscalers planning $100–$200 billion annually in data centre spending.
These structural bottlenecks are intensifying local power price spikes, driving off-grid gas generation, and causing commodity markets to price in undelivered capacity, risking overpricing and future corrections.
Operators increasingly adopt on-site generation, creating unreported natural gas demand, while the misalignment between announced and actual capacity raises concerns about stranded assets and underutilization if AI adoption lags.
Who will ultimately pay the price for the massive power AI data centers demand?
Will Big Tech's AI ambitions be powered by a 'shadow' fossil fuel grid?
Is the AI infrastructure boom a bubble built on 'bragawatts' and hype?
How can regulators manage the AI power surge without stifling innovation?
Can the physical world's infrastructure ever keep pace with AI's digital speed?
Is the AI revolution causing a 'great consumer chip winter' for electronics?