The global build-out of artificial intelligence (AI) infrastructure will require up to $1.6 trillion over the next five years, with access to power, not capital, now the primary constraint on growth, according to Knight Frank.
The global property consultant said in a statement on Thursday that global data center capacity is forecast to expand from 62GW in 2025 to more than 110GW by 2028, as hyper scalers race to scale AI infrastructure.
Meanwhile, AI-related capacity will more than triple from 8GW to 27GW by 2028, increasing its share of global workloads from 12.9 percent to 24.5 percent.
The shift is turning data centers into one of the most capital-intensive and energy-dependent sectors in the global economy.
Knight Frank highlighted that by 2030, AI-related deployment could reach 60GW, requiring $676 billion to $780 billion in development costs alone, with tenants expected to spend an equivalent or greater sum on information technology (IT) infrastructure.
Total global data center spending is projected to reach $3.2 trillion by the end of the decade.
According to the statement, capital expenditure by Microsoft, Amazon Web Services, Google and Meta is expected to exceed $650 billion in 2026, up 73 percent from $376 billion in 2025, as companies accelerate investment in AI compute.
AWS and Google are leading the expansion, with annual capex forecasts of $200 billion and $185 billion respectively.
Microsoft is expected to deploy $110 billion to $130 billion, following a record surge in late 2025, when capital expenditure rose 54% quarter-on-quarter.
Knight Frank opined that the impact is already visible in market activity.
Net colocation take-up reached a record 15.8GW in 2025, with 37 percent driven by AI demand, while AI-specific leasing volumes approached 6GW in 2025, three times higher than in 2024.
Knight Frank identified electricity infrastructure, particularly grid capacity, as the key bottleneck to scaling AI globally over the next 24–36 months.
It noted that grid capacity across core data center markets is largely reserved through to 2030, with connection delays extending to nearly 10 years in parts of London.
Vacancy rates have also tightened to record lows in several markets, including: Frankfurt (below 1 percent); Ashburn, Virginia (below 1 percent); Singapore (2.2 percent); London (3.1 percent).
According to Knight Frank, global wholesale vacancy has fallen to 8.1 percent, while new-lease rental rates are forecast to rise by 8 percent to 12 percent annually.
The scale of grid investment required is substantial: the United States may need $150 billion by 2030, the United Kingdom $107 billion to rebalance its power network, and China as much as $3.8 trillion by 2050.
This is driving a structural shift in site selection, with investors prioritizing “power-first” strategies focused on deliverable megawatts rather than location, it noted.
With grid access constrained, it also said major technology companies are moving aggressively to secure energy directly.
Microsoft has signed a 20-year, 835MW agreement to restart the Three Mile Island nuclear facility.
Amazon has secured up to 1.9GW of nuclear capacity in an $18 billion deal with Talen Energy.
Google and Meta have also committed to large-scale nuclear and long-term power agreements.
At the same time, governments are tightening requirements. In the United States, major technology companies including Amazon, Google, Meta, Microsoft, OpenAI and Oracle have committed to funding grid infrastructure under the Ratepayer Protection Pledge.
Meanwhile, Dublin has lifted a moratorium on new data centers on the condition that developers provide 100 percent on-site backup generation and source 80 percent of energy from new renewables.
Knight Frank also noted that a growing ecosystem of specialized AI infrastructure providers, or “neocloud” operators, is reshaping the market.
Sector revenue is projected to rise from $23.9 billion in 2025 to $179.1 billion by 2030, with around 200 operators globally.
companies such as CoreWeave and Nebius are scaling rapidly, supported by multi-billion dollar GPU contracts and capital investment, but are increasingly reliant on debt-heavy funding models.
Despite strong near-term fundamentals, Knight Frank warned that the scale and speed of investment are beginning to expose structural risks.
A $300 billion cloud agreement between Oracle and OpenAI, one of the largest AI infrastructure deals to date, has already faced revisions, including the cancellation of a 600MW data center expansion in Texas in March 2026.
As part of the deal, OpenAI’s committed $60 billion annual cloud spend compares with $13.1 billion in annualized revenue, raising questions over whether AI infrastructure demand can be sustained at current levels.
Oracle is also raising $50 billion to support the program, with credit agencies flagging leverage and counterparty risk.
In the near term, constrained supply and hyper scaler demand are expected to keep market conditions tight and landlord-favorable.
However, Knight Frank identified two critical variables that will determine the sector’s trajectory beyond 2028: whether AI workloads can generate sustained, large-scale revenues; whether power generation and grid infrastructure can catch up with demand.
It noted failure to rapidly monetize AI could lead to an overbuild scenario, whilst failure to scale power infrastructure will exacerbate pipeline constraints.
According to the statement, the rapid expansion of data centers is driving a parallel surge in electricity demand, with consumption expected to approach 945TWh by 2030, close to 3 percent of global demand.
This is accelerating investment across nuclear, renewables and grid infrastructure, and forcing governments and operators to rethink how energy systems are financed, built and allocated.
This marks a shift from reliance on public grids to vertically integrated energy strategies, it added.
“AI is scaling faster than the infrastructure that supports it. Capital is available, but power is not. Across every major market, access to electricity is now the gating factor for growth, shaping where data centers can be built and how quickly they can be delivered,
“The result is a shift to power-led strategies, with investors competing for scarce capacity in what is becoming one of the defining infrastructure challenges of the global economy,” said Stephen Beard, Head of Global Data Centers at Knight Frank.
Jiya Agarwal, Analyst, Data Centres, Asia-Pacific, Knight Frank, added Asia-Pacific is at the epicenter of the global AI infrastructure build-out.
According to her, the region already accounts for 41 percent of AI-focused colocation leasing activity, yet demand continues to run well ahead of available supply.
“Power has become the defining currency and increasingly, it is power delivery timelines, not capital, that are shaping investment decisions. In high-demand markets such as Tokyo and Seoul, grid connection queues are stretching to seven to ten years, pushing developers and hyper scalers alike to look beyond established hubs,
“Gigawatt-scale sites have become highly sought-after, and markets such as India, Malaysia and Indonesia are seeing a significant uptick in interest precisely because they offer the land, power availability, and pace of delivery that AI-driven demand now requires,” she added.
Fitch flags AI-driven credit risks in technology, media, and cloud sectors