Recommendations on Powering Artificial Intelligence and Data Center Infrastructure (SEAB, July 30, 2024)

Overview

This report was prepared by the Working Group on Powering AI and Data Center Infrastructure of the US Department of Energy's Secretary of Energy Advisory Board (SEAB) and presented to the Secretary of Energy on 30 July 2024. It is divided into three tracks: the efficiency and power dynamics of large language model training and inference, an operational-flexibility framework, and generation and storage options.

One recommendation in the first track is the creation of a data-center-scale AI testbed within DOE, distinct from existing high-performance computing facilities (Track 1, recommendation 4). The report's overall recommendations also include better load data, a flexibility taxonomy, support for novel backup power and model tariffs. These topics form a policy frame for how the relationship between data centres and the grid might be managed.

The second track recommends developing a flexibility taxonomy and framework, with a starting taxonomy in Appendix A (Track 2, recommendation 2). The same track recommends technical, business-model and permitting support for novel backup power strategies (recommendation 3), and suggests revisiting Federal Power Act 202(c) authorities to leverage data center backup capacity (recommendation 6). Backup generation is thus treated as a resource for grid support, not only for the facility's own security.

The third track notes that data centers largely rely on diesel generation for backup, with little experience using alternatives (Track 3, finding 1). The limit of the report is that it is a policy-recommendation document: it sits in the US energy market context and is not a design standard. It gives no numerical design values. It cannot be applied directly to projects in Turkey or other regions, but is a context source for the discussion of backup power options and grid flexibility.

Key points

1. Presented to the Secretary of Energy on 30 July 2024.
2. Three tracks: LLM training/inference efficiency and power dynamics, operational-flexibility framework, generation and storage options.
3. Recommends a data-center-scale AI testbed within DOE, distinct from existing HPC facilities.
4. Recommends developing a flexibility taxonomy and framework; a starting taxonomy is in Appendix A.
5. Recommends technical, business-model and permitting support for novel backup power strategies.
6. Suggests revisiting Federal Power Act 202(c) authorities to leverage data center backup capacity.
7. Data centers largely rely on diesel generation for backup, with little experience using alternatives.
8. Further recommendations cover better load data, model tariffs and a flexibility framework.

Using it in design

• Offer alternatives to diesel (gas, BESS, fuel cells) in backup-power design.
• Capture grid-support capability (curtailable load, backup export) as attributes for flexibility studies.
• Add the permitting and regulatory conditions for using backup generators for grid support to project notes.
• Use the report's flexibility taxonomy as a starting point for defining flexibility fields in your own design tools.

Who should read it

Data centre power designers, engineers planning grid connection and backup power strategy, and energy policy specialists.

Publication details

Authors

SEAB Working Group on Powering AI and Data Center Infrastructure

Publisher

U.S. Department of Energy, Secretary of Energy Advisory Board (SEAB)

Year

2024

Document language

English

Source type

Government / public

Access

Free to read

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation