AI Data Center Scalability Has Stalled on Power Limits
Experts identify critical power and cooling constraints for the future of hyperscale AI infrastructure development.
Updated on Sept. 29, 2026 in Data Centers
Live Poll
Do you worry that large-scale AI data centers will negatively affect energy availability in your area?
Large AI data centers have hit power constraints exceeding 100 MW per facility, shifting industry priorities toward integrated physical infrastructure. Global data center spending is now projected to exceed $650 billion for 2026 as firms prioritize power access over land.
Why it matters
Every watt of power consumed by AI accelerators generates heat, forcing engineers to balance cooling needs with power delivery limits. These constraints have turned power and memory bandwidth into the most strategic hurdles in the global semiconductor supply chain.
Large AI data centers currently require upwards of 100 MW of power to function at scale. Worldwide spending on data centers is forecast to surpass $650 billion throughout 2026.
The players
Ang Wee Seng
He is a leading expert who identified power and cooling as the primary constraints for AI data center scalability.
The details
Infrastructure design has moved toward integrated, rack-scale systems that combine accelerators, networking, and power delivery into single units. To manage these loads, facilities utilize advanced power electronics including silicon, silicon carbide, and gallium nitride technologies to convert grid electricity for rack-level shelves.
Timeline
July 2026 marked the original publication of these findings in an EE Times ASEAN e-guide.
September 2026 was when Ang Wee Seng detailed these infrastructure trends at EE Power Asia.
The Big Picture
This shift follows patterns established by the Semiconductor Industry Association's regional reports regarding hardware integration. The move toward rack-scale architecture marks a departure from traditional processor-centric design by prioritizing power delivery as the primary limit on AI growth.
The push for more efficient power electronics may eventually lead to improvements in hardware energy density and reduced operational costs for large-scale cloud services. These infrastructure changes ensure that growing AI demands do not compromise the stability of existing digital services.
The takeaway
Industry leaders must now solve the physical bottleneck of power cooling to maintain the pace of AI expansion. Investors and firms should watch how integration of gallium nitride and silicon carbide technologies dictates the feasibility of future hyperscale projects.
Further reading
For more information on the evolving physical requirements of modern computing, visit the Data Centers section.
Live Poll
Do you worry that large-scale AI data centers will negatively affect energy availability in your area?
