SKN CBBA -
SKN CBBA
Cross Border Banking Advisors
SKN | Citi Sees Data Center Expansion Continuing as Higher Costs Reshape AI Infrastructure Financing

Investors

SKN | Citi Sees Data Center Expansion Continuing as Higher Costs Reshape AI Infrastructure Financing

By Or Sushan

•

November 15, 2026

Key Takeaways:

  • Citi expects data center construction to continue despite rising costs and regulatory delays, supported by sustained demand for artificial intelligence computing capacity.
  • The bank forecasts AI token production to grow at an annual rate of 236% over five years, with generative AI workloads increasing from 26% of total data center demand in 2026 to approximately 82% by 2031.
  • More demanding permitting requirements and higher financing costs could disadvantage speculative developments while strengthening the position of completed facilities, established operators and projects with reliable power access.

The expansion of artificial intelligence infrastructure is entering a more demanding phase, with regulatory scrutiny, construction expenses and financing conditions increasingly influencing project viability. In its latest research assessment, Citi argues that these pressures are unlikely to stop the data center construction cycle because demand for computing capacity continues to expand. Instead, the bank expects a more selective development environment in which access to financing, power and regulatory approvals becomes increasingly important.

AI Demand Remains the Foundation of Citi’s Infrastructure Outlook

Citi strategist Steph Choe identifies sustained demand for semiconductors, computing capacity and supporting infrastructure as the principal reason the construction pipeline should remain active. The bank’s equity research forecasts AI token production, a measure of output generated by artificial intelligence systems, to increase at an annual rate of 236% over the next five years.

The composition of data center workloads is also expected to change substantially. Generative AI is projected to account for approximately 26% of total data center demand in 2026, increasing to around 82% by 2031. This shift would make generative AI a dominant driver of infrastructure requirements, supporting continued investment in computing systems, semiconductor capacity and the facilities needed to operate them.

For Citi, the scale of expected demand provides a foundation for continued development even as the economics of individual projects become more challenging. However, strong industry-wide demand does not guarantee that every proposed facility will be completed or achieve acceptable returns.

Regulatory Reviews Could Separate Viable Projects From Speculative Developments

Following the US midterm elections, data center developers face a more demanding regulatory environment. Additional permits, local reviews and restrictions could extend construction schedules and increase expenses for companies developing and operating facilities. Some of these costs may ultimately be passed on to tenants, potentially affecting commercial agreements and the economics of computing capacity.

Choe’s assessment suggests that recent regulatory actions and construction moratoriums are intended, at least in part, to distinguish financially viable developments from more speculative proposals. This could change the composition and timing of investment rather than bring the broader construction cycle to an end.

The implications are significant for project developers. Facilities that have already secured approvals, financing and power connections may hold an advantage over projects still exposed to permitting uncertainty. Longer development timelines can also increase the risk of construction-cost overruns and delay the point at which a facility begins generating operating revenue.

Financing Costs and Existing Infrastructure Become More Important

Higher financing expenses represent another constraint. Citi considers an increase in borrowing costs for new data center developments inevitable, particularly as investors demand greater compensation for construction and execution risks. The bank also observes that high-yield bonds issued in 2025 to finance completed and powered facilities have significantly outperformed securities issued this year for projects still under construction.

This divergence highlights how credit markets are beginning to distinguish between operational infrastructure and developments carrying greater completion risk. Citi expects high-yield data center bond issuance to decline in 2027 as risk premiums increase. However, Choe cautions that reduced activity in this segment should not automatically be interpreted as a contraction in overall financing. Funding may migrate toward other capital markets, while investment-grade hyperscaler backing and dedicated power supplies could help reduce borrowing costs.

For institutional investors, the distinction between demand growth and financing availability is becoming increasingly important. A strong long-term market opportunity does not remove the risks associated with construction delays, capital requirements or uncertain tenant commitments.

Infrastructure Investment Enters a More Selective Cycle

Citi’s assessment points toward a continued expansion of AI infrastructure, but with greater differentiation between projects. Developers with established customers, committed financing, reliable electricity access and secured regulatory approvals may be better positioned than speculative projects facing unresolved execution challenges.

Investors should monitor changes in high-yield issuance, borrowing spreads, construction schedules, tenant commitments and access to dedicated power. The next stage of the data center cycle may be defined less by the volume of announced projects and more by the ability to deliver operational capacity on time and within budget. Citi’s central message is that higher costs may reshape the market’s financing structure and competitive landscape without eliminating the underlying demand for AI computing infrastructure.

Leave a Reply

Your email address will not be published. Required fields are marked *

More like this