AI data centers are racing ahead, but the grid isn't keeping up: operators need to rethink how they connect to, generate and manage energy to unlock expansion without overwhelming already strained generation and transmission systems.
New flagship AMD and Intel servers with high core counts push performance boundaries while improving efficiencies. This report breaks down and visualizes improvements based on an analysis of SERT data.
In AI model training, idle GPUs — not high prices — are the biggest driver of cost, with poor utilization quietly burning tens of thousands of dollars in wasted compute capacity.
Cooling systems with similar energy performance can vary significantly in water consumption. An analysis of facility data confirms that evaporative cooling's efficiency advantage may be limited and dependent on unacceptable levels of water use.
Choosing whether to train a model from scratch or fine-tune an existing one comes down to the use case and cost — with hardware utilization remaining an important cost factor.
Although cloud platforms often offer the lowest cost for AI inference, on-premises deployment may be preferable due to application architecture, data locality and control requirements.
Data4 needed to test how to build and commission liquid-cooled high-capacity racks before offering them to customers. The operator used a proof-of-concept test to develop an industrialized version, which is now in commercial operation.
Competition for grid power is increasing; data center operators need to use reserved grid power responsibly — to support business objectives, maintain strong relationships with authorities and avoid negative publicity.
The proposed Scope 2 Guidance updates radically alter current accounting methodologies. These changes will complicate Scope 2 offset markets (e.g., EACs, RECs and GOs), adding unnecessary complexity and resource demands to Scope 2 accounting.
The public wants to understand data center resource use and performance metrics. Data center operators need to propose a label format that shares appropriate performance metrics while protecting confidential information.
The projected tripling of data center capacity calls into question the industry's commitment to sustainability. The growth creates an opportunity to build out energy- and water-efficient infrastructure and increase carbon-free energy use.
Serverless container services enable rapid scalability, which is ideal for AI inference. However, inconsistent and opaque pricing metrics hinder comparisons. This report uses machine learning to derive clear guidance by means of decision trees.
Underground hot rocks are emerging as a source of firm, low-carbon power for data centers, with new techniques expanding viable locations. Compared with nuclear, geothermal may be better positioned to support planned data center growth.
Cybersecurity has traditionally not been a key focus of attention for data center operators. But cyber incidents are on the rise and concerns are growing. Unaddressed vulnerabilities leave operators at increasing risk from evolving threats.
This briefing report identifies and describes several de facto standards and laws used in the field of data center sustainability and efficiency (for convenience, we use the term "standards" for all).
Jay Dietrich
Dr. Tomas Rahkonen
Dr. Owen Rogers
Dr. Rand Talib
Peter Judge
John O'Brien
Seb Shehadi