# arXiv: Multi-timescale scheduling for shared storage at AI data center campuses [Canonical reading](https://escapemeridian.com/readings/arxiv-multi-timescale-scheduling-for-shared-storage-at-ai-data-59fac2b2) Authors Zihan Meng, Ruizhang Yang, Chenxi Hu and Francis Yunhe Hou argue that planning shared storage for AI data center campuses must account for sustained power variations, subsecond fluctuations, and delays between commanded and delivered power. Their paper proposes a multi-timescale scheduling model for a shared hybrid energy storage system (HESS). It models electrical pooling at the campus bus while constraining the HESS to a single injection at the point of common coupling (PCC). A two-layer model passes five-minute scheduled power, energy state and reserve capacity to a causal fast response model, which distinguishes commanded from delivered power. The authors say this addresses a delay that interval-based schedules cannot capture. In case studies using the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset, the authors report reductions in PCC and fast integral absolute error (IAE), while retaining most of the improvement available under ideal response. These are case-study results; the abstract gives no numerical reduction figures or evidence from operating campuses. The proposal makes delivery timing part of the scheduling model, but the reported results do not establish how it performs on other power profiles. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv eess.SY · Article Original source date: Oct 30, 2026 (2026-10-30). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T10:54:09.523Z. First observed: 2026-11-02T05:06:39.460Z. Retrieved: 2026-11-02T08:16:52.258Z. - [Multi-Timescale Deliverability of Shared Hybrid Storage for AI Data Center Campuses](https://arxiv.org/abs/2610.41249) ## Limits The reported results are from case studies using the National Laboratory of the Rockies GenAI Power Profiles Dataset; the abstract does not report tests on other profiles or operating campuses. The abstract reports reductions but provides no numerical results or details of the case-study setup. ## Open question How would the reported reductions compare in case studies using AIDC power profiles beyond the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset? ## Source-stated The authors say AIDC campuses have sustained power variations and subsecond fluctuations, and that interval-based schedules do not capture the delay between commanded and delivered power at the campus connection. > interval-based schedules cannot capture the delay between commanded and delivered power at the campus point of common coupling (PCC) - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated The paper proposes a multi-timescale scheduling model that pools electricity at the campus bus and constrains the shared HESS to a single PCC injection. > it models electrical pooling at the campus bus and constrains the shared HESS to a single PCC injection - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated The model passes scheduled power, energy state and reserve capacity to a fast response model that distinguishes commanded power from delivered power. > a two-layer model passes five-minute scheduled power, energy state, and reserve capacity to a causal fast response model that distinguishes commanded power from delivered power - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated In case studies using the named dataset, the authors report reductions in PCC and fast IAE while retaining most improvement available under ideal response. > Case studies using the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset show reductions in both PCC and fast integral absolute error (IAE), while retaining most of the improvement available under ideal response. - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source and contributor context Meng, Zihan, Yang, Ruizhang, Hu, Chenxi, Hou, Francis Yunhe · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 2ef8e279-ff9e-41df-8619-3a8cbbafb70b. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T10:54:09.523Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 831aec0bb4bb5bed58de622d1217c2202ba93ad5597932e400755d27f0c4ab22. Source text hash: b9eeb71e0f7a3cf7aadbb139a032baf4b993d575541ac0b28c708b8be0bea3b1. ## Revision and source-check history ### Version 1 — 2026-11-10T10:54:09.523Z Authors Zihan Meng, Ruizhang Yang, Chenxi Hu and Francis Yunhe Hou argue that planning shared storage for AI data center campuses must account for sustained power variations, subsecond fluctuations, and delays between commanded and delivered power. Their paper proposes a multi-timescale scheduling model for a shared hybrid energy storage system (HESS). It models electrical pooling at the campus bus while constraining the HESS to a single injection at the point of common coupling (PCC). A two-layer model passes five-minute scheduled power, energy state and reserve capacity to a causal fast response model, which distinguishes commanded from delivered power. The authors say this addresses a delay that interval-based schedules cannot capture. In case studies using the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset, the authors report reductions in PCC and fast integral absolute error (IAE), while retaining most of the improvement available under ideal response. These are case-study results; the abstract gives no numerical reduction figures or evidence from operating campuses. The proposal makes delivery timing part of the scheduling model, but the reported results do not establish how it performs on other power profiles. ## Original source arXiv eess.SY · Article Original source date: Oct 30, 2026 (2026-10-30). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T10:54:09.523Z. First observed: 2026-11-02T05:06:39.460Z. Retrieved: 2026-11-02T08:16:52.258Z. - [Multi-Timescale Deliverability of Shared Hybrid Storage for AI Data Center Campuses](https://arxiv.org/abs/2610.41249) ## Limits The reported results are from case studies using the National Laboratory of the Rockies GenAI Power Profiles Dataset; the abstract does not report tests on other profiles or operating campuses. The abstract reports reductions but provides no numerical results or details of the case-study setup. ## Open question How would the reported reductions compare in case studies using AIDC power profiles beyond the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset? ## Source-stated The authors say AIDC campuses have sustained power variations and subsecond fluctuations, and that interval-based schedules do not capture the delay between commanded and delivered power at the campus connection. > interval-based schedules cannot capture the delay between commanded and delivered power at the campus point of common coupling (PCC) - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated The paper proposes a multi-timescale scheduling model that pools electricity at the campus bus and constrains the shared HESS to a single PCC injection. > it models electrical pooling at the campus bus and constrains the shared HESS to a single PCC injection - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated The model passes scheduled power, energy state and reserve capacity to a fast response model that distinguishes commanded power from delivered power. > a two-layer model passes five-minute scheduled power, energy state, and reserve capacity to a causal fast response model that distinguishes commanded power from delivered power - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source-stated In case studies using the named dataset, the authors report reductions in PCC and fast IAE while retaining most improvement available under ideal response. > Case studies using the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset show reductions in both PCC and fast integral absolute error (IAE), while retaining most of the improvement available under ideal response. - [Supporting source](https://arxiv.org/abs/2610.41249) ## Source and contributor context Meng, Zihan, Yang, Ruizhang, Hu, Chenxi, Hou, Francis Yunhe · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 2ef8e279-ff9e-41df-8619-3a8cbbafb70b. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T10:54:09.523Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 831aec0bb4bb5bed58de622d1217c2202ba93ad5597932e400755d27f0c4ab22. Source text hash: b9eeb71e0f7a3cf7aadbb139a032baf4b993d575541ac0b28c708b8be0bea3b1. ### uncertain — 2026-11-10T10:54:09.523Z — anthropic/claude-opus-5.5 Publication: 2ef8e279-ff9e-41df-8619-3a8cbbafb70b attribution — pass: The reading correctly attributes the paper to Zihan Meng, Ruizhang Yang, Chenxi Hu and Francis Yunhe Hou. It frames the problem, method and results as the authors' claims and report. > Authors:Zihan Meng, Ruizhang Yang, Chenxi Hu, Francis Yunhe Hou summary — uncertain: The reading tracks the abstract faithfully but stays largely in specialist language. A general reader is not told in plain words that the campus bus is the shared internal power connection, that the PCC is where the campus connects to the wider grid, or what integral absolute error (IAE) measures. It also never says why smoothing AI campus power swings matters, for example for the grid connection. Smallest repair: add a plain-language lead saying shared batteries or other storage could smooth AI campus power swings at the grid connection, plus tooltips for PCC, IAE and bus. > Artificial intelligence data center (AIDC) campuses contain sustained power variations and subsecond power fluctuations, requiring both sufficient stored energy and fast response. date — pass: The originalPublished value of 2026-10-30 matches the arXiv v1 submission date. It is not the later feed date of 2026-11-02. > \[Submitted on 30 Oct 2026\] support — pass: Each claim matches the abstract. The limitations correctly note there are no numerical figures and no evidence from operating campuses. The closing sentence on generalization and the question about other power profiles follow from the case-study scope without adding unsupported benefits. > Case studies using the National Laboratory of the Rockies (NLR) GenAI Power Profiles Dataset show reductions in both PCC and fast integral absolute error (IAE), while retaining most of the improvement available under ideal response. media — pass: The selected media is the verified arXiv abstract page, which is the direct original source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/59fac2b2641597411ced485c752d4f73de05ef5ed193f8a9c0dff412b18ea5c3/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/arxiv-multi-timescale-scheduling-for-shared-storage-at-ai-data-59fac2b2.json)