📌 The Test of Power Capacity’s Value Amid AI Expansion

The bottleneck in AI infrastructure is extending from procuring chips to whether power can be secured and actually used. UBS expects hyperscale cloud providers’ combined capital expenditures on AI infrastructure from 2026 to 2028 to total approximately USD 4.1 trillion, more than three times the investment of the previous six years. If this level of investment materializes, power capacity that can be delivered on schedule will directly affect whether data center and computing projects can take on orders.

This is also why some mining companies are being reassessed. Companies such as IREN, CIFR, and HUT collectively have approximately 32GW of planned capacity, with multiple companies’ HPC and AI deployments concentrated in Texas. For these companies, power resources originally serving mining may be redirected toward AI computing. But “planned capacity” does not equal capacity that can be leased out or generate cash flow; approvals, grid connection, construction, and customer contracting remain sequential execution hurdles.

CIFR’s case illustrates this difference: the near-term value of its approximately 1.2GW of approved capacity depends on leasing and construction progress; the larger-scale 4.4GW is planned to come online from 2028 to 2030. Nebius’s revenue expectations are likewise premised on converting 5GW of contracted power into long-term AI infrastructure contracts. Power resources are therefore more like a growth option, and the speed of realization determines whether they can become revenue and long-term contracts.

▌ Sources

• Cloud providers’ AI capital expenditure forecast by StockSavvyShay
• Overview of mining companies’ planned power capacity by StockSavvyShay
• Competition for AI power capacity in Texas by StockSavvyShay
• Milestones in Cipher’s capacity realization by StockSavvyShay
• Nebius’s path from power to contracts by StockSavvyShay