Morgan Stanley cuts US global computing share forecast to 55 pc amid power crunch
New Delhi, October 16
Morgan Stanley lowered its estimate of the US share of global computing capacity to 55 per cent from 60 per cent earlier as expanding data centre capacity overseas is unlikely to fully offset its projected power shortfall of 32 gigawatts between 2026 and 2028.
The report noted that US companies are increasingly exploring computing capacity overseas, citing Nvidia's plans to establish 2 GW of capacity in Australia by 2027, a market the company described as having "excess energy."
Morgan Stanley revised its estimate of the US share of global computing "down to 55% from 60%."
"We simultaneously revise our expectation of US Compute share down to 55% from 60%; these changes effectively balance out, leaving our total 2026-2028 power shortfall unchanged."
"While we expect international expansion to continue, we don't see it as a solution to the US power problem," it said.
Highlighting the challenges of shifting data centre operations overseas, the report said Europe has limited capacity to absorb additional demand due to power constraints and lengthy planning approval processes. Also, there is limited scope for further expansion in the region, while geopolitical instability could make the Middle East a challenging destination for new data centre investments.
"While there could be potential for expansion into the Middle East, we believe this is difficult given geopolitical instability," it said. Now, turning to Asia, Morgan Stanley said its projections already account for 14 GW of US hyperscaler demand shifting to the region through 2030. However, the US is still expected to face a power shortfall despite this anticipated expansion.
"...any incremental spillover would require tapping into energy sources like coal..." it said, stressing "we don't believe there's enough capacity available to absorb the US power shortfall."
The report further attributed the electricity deficit to the growing power requirements of advanced graphics processing unit (GPU) architectures, including Nvidia's NVL72 and NVL144 rack-scale systems, which consume significantly more electricity than existing eight-rack-unit (8U) systems.
"While more data centre capacity is shifting abroad, we don't believe there's enough capacity in ROW to absorb the US's net power shortfall (32 GW from '26-28)," it noted.
— ANI
Leave a comment