A new framework from the Electric Power Research Institute (EPRI) creates a classification system for data center demand response capabilities to help promote grid flexibility and, in turn, support the continued growth of digital infrastructure.
Internal efforts to maximize token use, combined with changes in LLM pricing structures, have rapidly increased enterprise AI spending. Yet there is often limited visibility into whether such expenditures are creating value.
Vendors and consultants expect broad and rapid adoption of AI in data center management, but they may be overestimating operators' appetite for change.
Traditional air gap security presents a barrier to applications requiring live IT-OT telemetry data. Rising interest in real-time monitoring and AI-driven operations requires a rethink of outdated, inflexible cybersecurity approaches.
Operators in most geographies should reconsider hastily-set commitments to reach 24x7 carbon-free energy consumption. Instead, they should adopt more realistic goals.
AI workloads are forcing operators to rethink cooling infrastructure: as power constraints intensify, thermal energy storage is gaining renewed attention as a way to shift cooling demand, reduce peak load and increase usable IT capacity.
The early years of scale DLC deployment came with complexities in system design, coolant chemistry, resiliency design and maintenance practices. Operator experiences suggest consensus is beginning to form, if imperfectly.
Benchmarks may produce impressive energy-per-token metrics, but real-world AI workloads are bursty; when throughput drops and GPUs sit idle, joules per token can increase. Do not size AI infrastructure for lab conditions — plan for demand.
An alert from the North American grid connection authority shows that data centers will be treated similarly to generation assets when requesting power connections, requiring operators to share more information and permit operational monitoring.
An exclusive focus on densification and DLC (as if they were inevitable) risks becoming tunnel vision that ignores costs and alternative choices. For IT infrastructure not fully transitioning to DLC, keeping densities moderate may make more sense.
When choosing whether to develop a brand new LLM or fine-tune an existing one, the second option often makes more sense. It can be more cost-effective and requires fewer IT and facility resources.
By integrating new natural gas electricity generation with carbon capture, operators can safeguard net-zero targets threatened by a reliance on fossil power — but initial adoption will be costly and limited to specific geographic locations.
AI applications are becoming critical to enterprise operations, but service availability still varies sharply across providers. Inference services should be evaluated not only on model capability, but on operational maturity.
As US state legislatures face difficulties passing local bans on data centers, opponents are increasingly turning to new regulatory approaches. Data center operators will need to navigate this fragmented policy landscape to stay ahead of compliance.
A new generation of vendors is entering the market to offer prefabricated modular data centers — betting that surging demand for AI will give these facilities a competitive edge.
Dr. Tomas Rahkonen
Dr. Owen Rogers
Max Smolaks
John O'Brien
Peter Judge
Jay Dietrich
Dr. Rand Talib
Jacqueline Davis
Daniel Bizo
Rose Weinschenk