The world is witnessing an unprecedented race to build data centres. Artificial intelligence, cloud computing, digital services and the explosion of data are driving demand for computing infrastructure at a pace never seen before. Yet despite billions of dollars being invested, supply is struggling to keep up.
According to a recent Jefferies research report titled “All the Data Center Demand in the World, Still Not Enough Supply”, the biggest challenge for the industry is no longer demand. It is the ability to build enough capacity quickly enough.
Jefferies estimates that global hyperscalers including Amazon, Microsoft, Google, Meta and Oracle are on track to spend nearly $770 billion on capital expenditure in 2026, a 74 per cent increase over the previous year. Since 2023, hyperscaler spending has increased nearly five-fold as companies rush to build AI infrastructure.
Demand is exploding
The numbers are staggering. Jefferies estimates that AI accelerator chip shipments alone imply nearly 30 GW of global AI power demand in 2026. Around 19 GW of that demand is expected to come from North America. However, only about 10.3 GW of new data centre capacity is expected to come online during the same period. In other words, demand is running almost twice as fast as supply.
The imbalance has become increasingly visible. In 2025, data centre operators signed leases for roughly 15.6 GW of capacity in North America, while only 3.4 GW of non-hyperscaler capacity actually came online. The result was a deficit of more than 12 GW. Jefferies notes that the gap has widened sharply from 2.9 GW in 2023 and 4.3 GW in 2024.
Why supply cannot keep up
Contrary to popular belief, the bottleneck is not money. The bottleneck is physical infrastructure. Jefferies identifies six major constraints that are limiting data centre expansion. The most immediate challenge is the shortage of engineering, procurement and construction labour. Skilled electricians, contractors and specialised workers are simply not available in sufficient numbers to meet demand. Cooling systems are another major hurdle and are expected to become the industry’s biggest bottleneck after 2028.
Other constraints include power availability, electrical equipment such as switchgear and transformers, land availability and utility connections. As AI workloads become more power intensive, modern facilities require significantly more cooling equipment and electrical infrastructure than traditional cloud data centres. This is stretching supply chains across the world.
The AI effect
Artificial intelligence has fundamentally changed the economics of data centres. According to Jefferies, AI accelerator deployments imply North American AI power demand of 19.2 GW in 2026, rising to more than 39 GW in 2027 and over 63 GW by 2028. By comparison, annual data centre capacity additions are expected to be only 10.3 GW in 2026 and 13.1 GW by 2028. The result is a structural shortage that could persist for years.
Large AI model developers are already scrambling to secure capacity. Jefferies highlights that hyperscalers are increasingly signing leases years in advance because they fear future shortages.
Winners in a supply-constrained market
The shortage is creating winners across the ecosystem. Data centre operators are benefiting from rising occupancy and rental rates. Vacancy levels in major markets have fallen to between 1 per cent and 3 per cent, giving operators strong pricing power. Equipment suppliers are seeing order books grow much faster than revenues as customers compete for scarce hardware.
Jefferies believes that companies involved in cooling systems, electrical equipment, power infrastructure and engineering services stand to benefit the most from the prolonged supply crunch.
What it means for India
India is emerging as one of the fastest-growing data centre markets globally. The country’s digital economy, AI ambitions, cloud adoption and expanding internet user base are driving demand for local capacity. However, the lessons from global markets are clear. Building data centres is no longer just about securing capital. Access to power, skilled labour, cooling technology and supply chain resilience will determine who wins the next phase of the AI infrastructure race.
As artificial intelligence moves from experimentation to mass adoption, the demand for data centres is likely to remain insatiable. The question is no longer whether the world needs more data centres. The question is whether the industry can build them fast enough.
