U.S. Computing Share Declines on Power Shortages
A 32-gigawatt deficit through 2028 is driving hyperscale demand to Asia and Australia.
Updated on Oct. 16, 2026 in Data Centers
Live Poll
Is now a good time to prioritize energy-intensive computing developments in your local area?
Morgan Stanley has revised its forecast for the U.S. share of global computing capacity from 60 percent down to 55 percent. This shift is driven by a projected 32-gigawatt power shortfall in the U.S. between 2026 and 2028, leading to a migration of hyperscaler demand to overseas markets.
Why it matters
Domestic power constraints are forcing a fundamental reorganization of infrastructure strategy, as energy availability dictates where AI compute can scale. High-performance GPU architectures are creating electricity demands that current U.S. grid capacity cannot satisfy, forcing companies to look abroad.
Data center demand is increasing as systems like Nvidia’s NVL72 and NVL144 GPU architectures consume significantly more power than older 8U systems. Projections indicate 14 gigawatts of hyperscaler demand will shift to Asia by 2030.
The players
Morgan Stanley
An investment bank that provides financial market analysis and forecasts for global infrastructure and technology sectors.
Nvidia
A semiconductor company known for designing the high-power GPU architectures that power current AI training and data center expansion.
The details
Modern data center power demands are driven by high-density rack architectures, such as the NVL72 and NVL144, which cluster large numbers of GPUs (graphics processing units — chips specialized for parallel processing) into single, interconnected systems. These configurations require vastly more cooling and electricity per rack than previous 8U (standard modular chassis) designs. To manage these loads, companies are diversifying internationally, with plans such as Nvidia establishing 2 gigawatts of capacity in Australia by 2027.
Timeline
2026-2028: Period of the projected 32-gigawatt U.S. power shortfall.
2027: Planned completion of 2 gigawatts of capacity in Australia by Nvidia.
2030: Deadline for the migration of 14 gigawatts of U.S. hyperscaler demand to Asia.
The Tech Race
The U.S. is losing its dominance in global computing capacity as energy constraints force hyperscalers to expand elsewhere. This shift follows the trend identified by Morgan Stanley, where power availability now dictates the speed and location of infrastructure rollouts.
Developers and companies relying on U.S.-based hyperscale cloud infrastructure may face increased regional latency or costs as capacity migrates to Asia and Australia. Businesses should evaluate the geographical distribution of their compute resources to mitigate risks from domestic grid volatility.
The takeaway
The race to build AI capacity is no longer just about chip production but about securing reliable, massive power footprints. Investors and IT leads should track the 32-gigawatt shortfall projection to see if new energy sourcing—including coal—materializes in the U.S. by 2026.
Further reading
For more on how infrastructure constraints influence scaling, visit Data Centers.
Source note: This article includes information reported by NewKerala.
Live Poll
Is now a good time to prioritize energy-intensive computing developments in your local area?
