AI-driven data centre demand far outstrips supply despite record hyperscaler spending: Jefferies
New Delhi, June 8
Global demand for data centres continues to significantly outpace available supply, creating a structural shortage that is expected to persist for years as artificial intelligence adoption accelerates, according to a report by Jefferies.
The brokerage said that despite aggressive capital expenditure plans by major technology companies and a rapid expansion in AI infrastructure, physical constraints across the supply chain are preventing the industry from delivering enough capacity to meet demand.
"Demand for data centers continues to outpace supply, with hyperscaler capex accelerating and chip volume forecasts implying GWs of capacity ahead of feasible data center delivery," the report said. It added that "Demand is not the issue, but the supply chain is."
According to Jefferies' proprietary data centre capacity analysis, only 8.9 gigawatts (GW) of capacity became operational in 2025, while demand reached nearly 21.1 GW, resulting in a deficit of about 12 GW.
The report highlighted that spending by hyperscalers such as Amazon, Microsoft, Google and Meta is rising sharply. Jefferies estimates hyperscaler capital expenditure will reach 770 billion US dollars in calendar year 2026, representing a 74 per cent increase year-on-year and nearly five times the 156 billion dollars invested in 2023.
At the same time, cloud service backlogs have surged to around 2 trillion US dollars, reflecting continued strong demand for computing infrastructure driven by AI applications.
Jefferies also noted that semiconductor shipment forecasts point to a substantial rise in AI-related power demand. Based on projected accelerator chip volumes, global AI data centre power demand could reach around 30 GW in 2026, with nearly 19 GW expected in North America alone.
The report warned that multiple supply-side bottlenecks are limiting the pace at which new data centres can be built. These include shortages of engineering and construction labour, cooling equipment, power infrastructure, transformers, electrical components and backup generators.
The brokerage believes the persistent gap between demand and supply is benefiting data centre operators, infrastructure developers and equipment suppliers. Vacancy rates remain extremely low across key markets, while leasing activity and rental rates continue to rise.
Jefferies further stated that from 2021 to 2025, the cumulative deficit of undelivered data centre capacity reached 20.4 GW and is likely to widen further as AI adoption drives sustained demand for computing infrastructure.
The report noted that hyperscalers may increasingly commit to capacity several years in advance as supply constraints continue to delay project delivery timelines.
— ANI
Reader Comments
$770 billion in capex by 2026 is mind-boggling! But with AI power demand reaching 30 GW globally, I'm concerned about the environmental impact. Where will all that electricity come from? In India, we're still struggling with coal dependency. Hope hyperscalers are investing in renewable energy to power these data centres, otherwise we're just trading one problem for another. 🤔
This is why we need to push for more semiconductor manufacturing in India. The supply chain bottlenecks Jefferies mentions - transformers, chips, cooling equipment - are all things we can potentially manufacture domestically. Our PLI schemes should focus on this entire ecosystem, not just assembly. Make in India for AI infrastructure! 💪
As someone working in cloud infrastructure in Bangalore, I can confirm this shortage is real. We've been waiting 18+ months for some GPU clusters. The Jefferies report is spot-on about the supply chain issues - even getting basic electrical panels is a nightmare right now. But it's also creating opportunities for Indian engineering firms that can step up and fill the gaps.
The numbers are impressive but I'm a bit skeptical. $2 trillion cloud backlogs? That sounds like companies are over-ordering capacity out of FOMO. Reminds me of the fiber optic boom in early 2000s. Hyperscalers should be careful not to overbuild, even if demand seems infinite today. AI adoption is real, but we've seen technology hype cycles before. Just saying... 🤷♂️
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.