Infosys: Data Center Efficiency Can Ease AI Power Constraints, but Not Replace Grid Investment
Редакция ЦОД:Главное
Infosys argues that data center efficiency measures and the scheduling of flexible compute workloads can create short-term power headroom for AI demand. The company’s analysis, covering AI energy demand through 2030, maintains that these measures must be paired with investment in generation, transmission and grid modernization.
- Efficiency, cooling and peak-load optimization may provide short-term power headroom for AI workloads.
- Training, retraining and batch workloads are more suitable for time shifting than real-time inference or customer-facing services.
- Load flexibility requires workload-specific SLAs, demand-response incentives and tariffs for non-urgent compute.
- Infosys frames demand management as a complement to, not a replacement for, generation, transmission and grid investment.
Infosys is advocating a two-track response to rising AI-related power demand: extract more usable capacity from existing infrastructure while accelerating investment in new generation, transmission and grid upgrades.
In an analysis of AI energy demand through 2030, the company identifies AI-managed energy operations, improved cooling and peak-load optimization as measures that could make capacity available more quickly than building new generation, transmission assets or grid infrastructure. The objective is to create near-term headroom for AI workloads where power availability is constrained.
The operational potential depends on the workload mix within a given data center. Infosys identifies model training, retraining and batch processing as workloads that may be flexible enough to schedule around grid conditions. By contrast, real-time inference and customer-facing applications generally are not suitable for time shifting.
For operators, this distinction makes workload-aware power management central to any flexibility strategy. Broader use of demand-side flexibility would require service-level agreements that reflect workload type, demand-response incentives and tariff structures that support the movement of non-urgent compute work.
Infosys does not present a quantitative estimate of the AI capacity that could be unlocked through efficiency improvements, cooling upgrades or flexible load management. Its conclusion is instead that these measures can provide temporary relief, while durable AI infrastructure expansion will still depend on the pace of new generation, transmission development and electricity-grid modernization.
For data center operators, demand-side flexibility may provide a near-term way to accommodate some AI load where power capacity is constrained. Its value is limited by workload composition: training, retraining and batch processing may be time-shifted, while real-time inference and customer-facing applications generally cannot. Efficiency and load management therefore complement, rather than substitute for, new power and grid infrastructure.
- — / 22 октября 2026 г. · АнглийскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.