Three organizations — the National Electrical Manufacturers Association (NEMA), the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE), and the Pacific Northwest National Laboratory (PNNL) — have jointly published the AI Data Center Energy Performance Framework. The document consolidates technical standards and operational guidance spanning the full data center lifecycle, from initial site selection through commissioning, day-to-day operations, and future retrofits.
What the framework covers
The framework addresses a broad set of concerns that have grown more acute as AI workloads demand higher and more sustained power densities: energy sourcing and efficiency, thermal and water management, grid interaction, resiliency, and ongoing operational performance. It draws on more than a dozen existing NEMA standards covering equipment categories such as energy storage, microgrids, transformers, switchgear, uninterruptible power supply systems, and grounding infrastructure, while weaving in ASHRAE's thermal management expertise and PNNL's knowledge of energy management and grid integration.
The stated goal is to give project developers, engineers, and facility managers a single coherent reference rather than requiring each organization to assemble its own proprietary collection of guidance from multiple standards bodies. NEMA senior vice president Patrick Hughes told Data Center Knowledge that too many operators currently restart this assembly process from scratch on every project, adding time and uncertainty to an already compressed development cycle.
Why the AC-to-DC shift matters
One of the more substantive technical threads running through the framework concerns the industry's gradual move away from alternating-current power distribution toward high-voltage direct-current architectures. Hughes described the conventional AC pathway — from grid interconnection through a step-down transformer, into a UPS that converts to DC, back to AC, and finally to low-voltage DC at the server rack — as a chain of conversions that each carry efficiency losses. Eliminating intermediate steps via high-voltage DC distribution, such as 800 VDC systems, offers a meaningful efficiency gain.
The obstacle is that formal standards for these higher-voltage DC configurations are still being developed. In the interim, many operators are deploying adapted or proprietary solutions, which itself compounds the fragmentation problem. The framework attempts to address the gap at a systems level, providing enough connective tissue between existing standards that suppliers can begin designing and certifying equipment with greater confidence ahead of formal high-voltage DC standards reaching maturity.
Why it matters now
The release comes against a backdrop of intensifying pressure on data center developers to compress the timeline between project approval and operational capacity — sometimes called "speed to power." Hyperscalers, colocation providers, and independent developers are all exploring alternative generation, on-site storage, and microgrid arrangements to reduce dependence on utility interconnection queues.
Hughes noted that the codes and standards originally written for conventional commercial and industrial facilities were not designed with the energy consumption patterns of modern AI infrastructure in mind. Guidance connecting electrical and thermal systems to the specific demands of high-density AI compute has historically been scattered across multiple organizations, making it difficult to apply consistently. By consolidating input from NEMA, ASHRAE, and PNNL under a single document, the framework is intended to reduce that coordination overhead for teams designing new facilities or modernizing existing ones.
The framework is available through ASHRAE's technical resources webpage.
For professionals: Engineers and facility managers evaluating high-voltage DC architectures should note the framework's acknowledgment that formal 800 VDC standards remain incomplete; designs relying on adapted solutions carry certification risk that the framework partially mitigates at a systems level but does not fully resolve. Teams planning new AI data center builds can use the document to audit whether their internal standards libraries align with the consolidated NEMA, ASHRAE, and PNNL guidance.
Automated pipeline · Cloud & Infrastructure
Synthesized from 1 industry feed on 15 Jun 2026. Passed independent editor verification before publication. Style guide v1.2.
Sources
Decision trail
- Checking for duplicates — Duplicate story same-story cluster; write with candidate 18; cluster_primary=18
- Writing the article — Draft created article_id=42 slug=nema-ashrae-and-pnnl-release-unified-ai-data-center-energy-framework
-
Editor review — Approved
- Quote integrity — material: The article paraphrases Hughes as saying 'too many operators currently restart this assembly process from scratch on every project, adding time and uncertainty to an already compressed development cycle.' The source quotes Hughes verbatim as: 'Too many data center developers are working from proprietary internal standards. But it takes time for a data center developer to start from scratch on every project.' The article does not present this as a blockquote, so it is an attribution, not a quoted block — this is minor. However, the paraphrase adds 'adding time and uncertainty to an already compressed development cycle' which is not in the source quote or nearby context, slightly overstating what Hughes said.
- Factual grounding — minor: The article states the framework draws on 'more than a dozen existing NEMA standards covering equipment categories such as energy storage, microgrids, transformers, switchgear, uninterruptible power supply systems, and grounding infrastructure.' The source also lists 'wire and cable, electricity metering, fire and life-safety equipment, insulating materials' — these are omitted, which is acceptable compression, not an invented claim.
- Factual grounding — minor: The article states the AC pathway goes 'from grid interconnection through a step-down transformer, into a UPS that converts to DC, back to AC, and finally to low-voltage DC at the server rack.' The source describes this conversion chain in the same order and matches the article's description accurately.
- Style compliance — minor: Body word count is approximately 680 words, which is within the hard maximum of 750 but exceeds the target ceiling of 620 words. This is a minor style violation.
- Factual grounding — minor: The article says Hughes 'noted that the codes and standards originally written for conventional commercial and industrial facilities were not designed with the energy consumption patterns of modern AI infrastructure in mind.' The source quotes Hughes as saying 'data centers are using and managing energy in ways fundamentally different from what the codes and standards governing this type of infrastructure were designed to address.' The paraphrase is accurate and acceptable.
- Assigning hero image — Pexels pexels_id=37605910
- Linking related stories — Linked 4 relations from 26 candidates
- Linking related stories — Linked 4 relations from 26 candidates
- Publishing — Published nema-ashrae-and-pnnl-release-unified-ai-data-center-energy-framework

Discussion · coming soon
Be the first to join the thread when community discussion launches.