Google’s Reported Long-Term Constellation Power Agreement Highlights AI Infrastructure Constraints
DataNews Editorial Team
Former Google Taiwan head Jan Li-feng said Google recently reached a long-term electricity-purchase agreement with Constellation Energy in the United States, citing it as an example of the urgency of securing power for AI infrastructure. He argued that competition for sites with water, power and cable infrastructure is tightening.
- Google was reported by Jan Li-feng to have reached a long-term electricity-purchase agreement with Constellation Energy in the United States.
- The comments place water, power and cable infrastructure among the key constraints on AI data-center site development.
- Li contrasted a 20–30-year new-nuclear construction timeline with a stated one-to-two-year period to raise output at an existing plant by 33%.
- Pre-secured infrastructure sites and long-term power procurement could become more valuable if competition for AI capacity continues.
- The underlying material also flags the risk that AI demand growth may slow if enterprise deployment becomes more cautious.
Jan Li-feng, former head of Google Taiwan, said Google recently reached a long-term agreement with Constellation Energy to purchase electricity in the United States. He presented the reported agreement during a discussion of AI and digital-transformation trends, framing it as evidence of the pressure technology companies face to secure energy for AI data-center infrastructure.
Li said competition is intensifying for sites that combine water availability, electrical supply and cable infrastructure. In his assessment, such sites could become unavailable for purchase within one to two years. This is an assessment by the speaker rather than a confirmed market forecast, but it underscores the importance of site readiness and grid-accessible power in AI campus development.
He also contrasted the lead time for new nuclear generation with efforts to increase output from existing nuclear plants. According to his remarks, building a new nuclear power plant can take 20–30 years, while increasing generation at an existing plant by 33% could take one to two years. The source does not identify the plant concerned or establish that the 33% figure is connected to the reported Google–Constellation agreement.
The comments suggest that technology companies view AI data-center investment as urgent despite pressure on cash flow. For data-center operators and hyperscalers, the implication could be a higher premium for sites with secured power, water and network infrastructure, alongside greater interest in long-term power procurement and accelerated additions from existing generation assets rather than relying solely on new-build capacity.
The same material notes a demand-side risk: expected AI demand could moderate if companies take a more cautious approach to deployment. No contracted energy volume, delivery terms, date of signing, data-center capacity, CAPEX or associated infrastructure-project status was disclosed.
Power availability, water access, cable infrastructure and site readiness are increasingly potential constraints on AI data-center expansion. If this competition persists, pre-secured powered sites and faster-to-access generation from existing assets could command a premium, while hyperscalers and operators may place greater emphasis on long-term power contracts.
- TechNews 科技新報 / ChineseSource →
This article was prepared by the DataNews editorial team based on the sources listed above.