Susquehanna Financial Group has warned that regulatory uncertainty following the 2026 US midterm elections could threaten approximately 12 gigawatts of planned artificial intelligence data centre capacity, representing 8% of projects scheduled for development between 2025 and 2030.
In a report released Friday, the firm identified potential delays affecting a much larger portion of planned infrastructure, although it suggested that near-term disruption should remain relatively limited.
Under its worst-case scenario, approximately 12 gigawatts of projects scheduled for 2028 through 2030 could be cancelled because of regulatory challenges.
Regulatory Approvals Put 40% of Planned Capacity at Risk of Delays
Susquehanna estimates that approximately 148 gigawatts of training-focused AI data centre capacity is planned for development between 2025 and 2030.
Of that total, around 60 gigawatts, equivalent to 40%, remains exposed to potential delays because regulatory approvals are still pending.
The firm calculates that the 12 gigawatts potentially facing cancellation would account for approximately 8% of total planned capacity and 20% of projects awaiting approval.
The analysis highlights growing uncertainty surrounding the expansion of AI computing infrastructure, particularly for developments requiring additional regulatory authorisation.
Most 2027 Projects Already Have Permits
Despite the potential disruption, Susquehanna expects the immediate impact on AI infrastructure development to be relatively contained.
Most projects scheduled for completion or installation in 2027 have already secured the necessary permits, reducing their exposure to new regulatory obstacles.
As a result, the firm continues to anticipate that 2027 will represent the peak year for new AI data centre installations.
The greatest risks are concentrated in the 2028-2030 period, when projects that have not yet obtained approval could face longer development timelines or possible cancellation.
AI Inference Demand Could Offset Infrastructure Shortfalls
Susquehanna noted that additional sources of semiconductor demand could help compensate for any reduction in planned AI training infrastructure.
Enterprise-driven AI inference workloads are expected to begin accelerating in 2027, creating demand for computing resources beyond those included in the firm’s 148-gigawatt estimate.
Inference refers to the processing required when trained AI models generate responses or perform tasks, rather than the computing power used to train those models.
The firm also identified a potential recovery in demand for semiconductors outside the AI sector as another factor that could partially offset weaker spending on data centre construction.
Semiconductor Revenue Could Decline 10% in Worst-Case Scenario
Susquehanna warned that the semiconductor industry could experience a 10% year-over-year revenue decline under its most pessimistic scenario.
The assessment comes after substantial weakness in semiconductor equities.
The PHLX Semiconductor Sector Index has fallen approximately 38% from its June peak and is trading near a historical low in its forward price-to-earnings valuation, at around 15 times expected earnings.
The decline suggests that investors have already adjusted valuations to reflect concerns about AI infrastructure spending and potential delays in new computing capacity.
Susquehanna Says Market May Already Reflect Regulatory Risks
Although the firm acknowledged that project-level risks have increased, it suggested that semiconductor share prices may already incorporate much of the potential negative impact.
With most 2027 developments permitted and additional demand expected from enterprise inference applications, Susquehanna sees possible factors that could limit the industry’s exposure to delayed data centre projects.
The firm’s analysis indicates that regulatory uncertainty remains a significant consideration for AI infrastructure investment, but current semiconductor valuations may reflect a more pessimistic outlook than the underlying project risks warrant.