Artificial intelligence (AI) is getting into businesses faster than the physical infrastructure built to support it. Even after massive spending by hyperscalers like Google, Microsoft, Meta, and Amazon on data centres and related infrastructure, demand for AI-ready data centres is still outpacing supply. Data center construction, power access, chip supply and skilled labor are all becoming bottlenecks rather than background details.
According to a report by Jefferies, global demand for data centers is significantly outpacing the available supply, resulting in a structural shortage that is expected to last for years as the adoption of AI accelerates.
Jefferies research says that hyperscaler spending remains strong, AI demand remains strong, infrastructure shortages are likely to persist, and data centre operators may continue to benefit from favourable supply-demand dynamics.
[Also Read: SoftBank Plans 5 GW AI Data Center Network in France, Starts With 1 GW Bosquel Project]
AI models need power, chips, an efficient cooling system and space. Those four inputs are not expanding at the same speed as AI usage. Big Tech’s planned AI infrastructure spending for 2026 was about $635 billion, up from $383 billion in 2025 and $80 billion in 2019, but S&P Global still warned that rising energy costs and geopolitical instability could force cutbacks.
Companies can commit capital, but they still have to wait for grid connections, permits, equipment and labor. Even declared AI projects often run into real-world constraints such as energy capacity, logistics and construction delays.
The supply problem is not just about chips
A lot of the public discussion around AI shortages has focused on GPUs. That is only part of the picture. Morgan Stanley warned this week that memory-chip prices have increased sixfold in a year, creating what it called “chipflation,” as suppliers prioritize high-margin data center demand over consumer devices.
But the more major bottleneck is power supply. AI data centers are highly copper-intensive and depend on grid connectivity, electrical equipment, and construction capacity. That makes them vulnerable to a long list of supply-side delays: substations, transformers, backup power systems, and the workers needed to install them.
The scale of spending is already record-setting
The spending is large enough to explain why investors are still upbeat, but also why shortages persist. U.S. data center construction spending hit a record $40 billion annualized rate in June 2025, up 30% year over year after a 50% rise in 2024.
Global data center dealmaking reached a record in 2025, with more than 100 deals worth nearly $61 billion by November. That was higher than the previous record of $60.81 billion in 2024.
Even that pace is not closing the gap quickly. Alphabet said in February that its 2026 capital spending could rise to $175 billion to $185 billion, up sharply from $91.45 billion in 2025. Big Tech companies were also spending heavily as AI demand kept climbing.
Why demand is still outrunning supply
There are 3 reasons the market is still short of AI capacity.
First, enterprise adoption is broadening. AI is no longer limited to chatbots and basic-level applications. Now it is being used in search, coding, customer service, security and internal automation. That creates continuous infrastructure demand rather than one-off testing.
Second, the hardware stack is getting heavier. Morgan Stanley said memory chips are now taking a much larger share of hyperscaler AI spending, with memory costs rising sharply as data-center demand expands.
Third, build time is long. Reuters has noted that power and infrastructure projects can take years, not months, to come online. Even when firms have cash and chip orders in place, the physical buildout lags behind demand.
[Also Read: Google and Blackstone Form AI Cloud Venture as Demand for AI Data Centers Surges ]
