Pith. sign in

REVIEW 2 major objections 5 minor 21 references

Turning AI Data Centers into Grid-Interactive Assets: Results from a Field Demonstration in Phoenix, Arizona

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper reports the first field demonstration of a software-only platform that cut a 256-GPU cluster's power by 25% for three hours during peak grid events while preserving AI quality-of-service guarantees.

desk verdict A real first-of-its-kind field demo of software-only demand response on a production AI cluster, but the grid-benefit headline overreaches because only GPU power was measured, not facility-level load. read the letter →

arxiv 2507.00909 v1 pith:SWHYHNW4 submitted 2025-07-01 cs.DC cs.AIcs.PFcs.SYeess.SY

Philip Colangelo , Ayse K. Coskun , Jack Megrue , Ciaran Roberts , Shayan Sengupta , Varun Sivaram , Ethan Tiao , Aroon Vijaykar
show 7 more authors
Chris Williams Daniel C. Wilson Zack MacFarland Daniel Dreiling Nathan Morey Anuja Ratnayake Baskar Vairamohan
This is my paper · ORCID
classification cs.DCcs.AIcs.PFcs.SYeess.SY
keywords demandresponseAIdatacentersGPUpowercappingworkloadorchestrationgridflexibilityqualityofservicefielddemonstrationenergysystems
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports a field demonstration in a 256-GPU cluster inside a commercial hyperscale data center in Phoenix: a software platform called Emerald Conductor reduced cluster GPU power by 25% for three hours during two utility-designated peak grid events and during a reenacted emergency event, while every job stayed inside its agreed performance envelope. The authors' central claim is that AI data centers can act as grid-interactive assets using only workload orchestration—power capping, job pausing, and resource reallocation—without hardware retrofits or energy storage. The demonstration is offered as the first validation that sustained, accurate power reductions can come from workload scheduling alone on a production AI cluster. A companion simulator predicted cluster power with about 4.5% root-mean-square error, which is what lets the controller commit to a grid response ahead of time. If the claim holds, utilities could treat AI compute as flexible demand and accelerate interconnection of new data centers without building as much new generation or transmission.

What carries the argument

The load-bearing mechanism is Emerald Conductor, a centralized scheduler that reads real-time grid load forecasts, consults the Emerald Simulator—a trained system-level model of job power and throughput behavior—and issues control commands to each compute node. The control knobs are GPU power capping through frequency scaling, pausing jobs at checkpoint boundaries, and changing the number of GPUs allocated to a job; jobs are tagged Flex 0 through Flex 3 by how much throughput degradation their service agreement tolerates. The simulator's 4.52% RMSE in power prediction is what makes the approach trustworthy: the controller can pre-commit to a grid reduction and then execute it without breaching job SLAs.

What would settle it

Install a facility-level power meter on the same 256-GPU cluster and repeat a utility peak event: if whole-data-center demand does not fall by roughly the same percentage as the GPU-level measurement during the three-hour window, the claim that the cluster acts as a grid-interactive asset would be falsified at the point of grid interconnection.

Watch

Extended reading notes

Core claim

The paper's core claim is that GPU-driven AI workloads contain enough operational flexibility that a scheduler can turn a production AI cluster into a controllable grid resource. In the Phoenix trial, the cluster met utility-set targets of a 25% power reduction sustained for three hours with graceful 15-minute ramps, and matched a two-step emergency curtailment profile, all with zero SLA violations across more than thirty experiments and over two hundred jobs. The authors attribute this to workload tagging into flexibility tiers (no slowdown, up to 10%, 25%, or 50% throughput reduction), an offline-trained simulator that predicts power and performance, and a controller that applies DVFS power capping, job pausing, and GPU reallocation according to greedy or fair policies. They conclude that this is the first validation of software-only, sustained demand response from an AI cluster, positioning data centers as flexible assets rather than static loads.

Load-bearing premise

The load-bearing premise is that a 25% cut in GPU-cluster power, measured on the GPUs themselves, translates into a meaningful reduction at the data center's grid meter; the trial did not measure facility-level loads such as cooling, power distribution, and networking, so the actual relief seen by the grid could be substantially smaller.

Editorial extensions

If this is right

  • Existing AI clusters can participate in demand response without capital investment, since only software changes are needed.
  • Utilities can set concrete, verifiable targets for AI data center flexibility—sustained reduction, ramp rate, duration—and expect them to be met.
  • Flexibility tiers give data center operators a contract language for AI SLAs that preserves strict jobs while allowing power management on tolerant jobs.
  • If broadly deployed, the approach could unlock on the order of 100 GW of new AI data center capacity in the U.S. on existing infrastructure, per the cited headroom estimates.
  • The same platform can respond to emergency events with stepwise curtailments, not just pre-scheduled peak events.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because power was measured only at the GPU level via the system management interface, the grid-level benefit remains untested; a whole-facility meter would be needed to confirm that cooling and other overheads do not erode the 25% reduction.
  • The tiered-SLA model presumes customers will accept measurable throughput loss in exchange for cheaper or faster compute; that market mechanism, not the technology, is likely the bottleneck to scaling.
  • Latency-sensitive inference (Flex 0) is excluded here; shifting workloads across geographic zones, rather than slowing them, is a natural extension that could capture flexibility without any SLA impact.
  • Simulator accuracy of about 4.5% RMSE suggests the approach could be extended to day-ahead capacity markets, where a data center bids a fixed load reduction in advance using simulator predictions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. This paper reports a field demonstration of Emerald Conductor, a software orchestration platform that curtails power consumption of a 256-GPU A100 cluster at an Oracle Cloud data center in Phoenix, Arizona, in response to grid signals. The trial, run with SRP, APS, and EPRI, claims to have achieved a 25% reduction in cluster power for three hours on May 1 and May 3, 2025, plus a synthetic CAISO-style emergency event, while preserving AI QoS. The paper also reports a 4.52% RMSE for the Emerald Simulator's power predictions and describes flexible SLA tiers (Flex 0-3), control knobs (DVFS, job pausing, resource reallocation), and orchestration policies (Greedy, Fair). The central claim is that software-only workload orchestration can turn AI data centers into grid-interactive assets without hardware retrofits or storage.

Significance. If the headline result holds, this is an important first field demonstration that GPU AI workloads can provide sustained demand response through software alone, and the involvement of utility partners (SRP, APS) and EPRI's DCFlex initiative lends practical credibility. The utility-set targets and real grid-event timing are strengths, and the paper provides useful detail on control knobs and workload flexibility tiers. However, the significance is materially limited by the measurement boundary: the 25% reduction is measured on GPU power only, not at the facility or grid meter, and the paper itself concedes in Section 3.5 that full data center telemetry is needed for system-level validation. The absence of raw data, a precise baseline definition, and error bars also prevents independent verification of the quantitative claims.

major comments (2)
  1. [Section 2.2] The baseline definition is not precise enough to verify the claimed 25% reduction. The text says the reduction was measured with respect to 'the average base load during the peak demand period,' but it does not specify the averaging window, whether the baseline is pre-event, same-time previous day, weather-adjusted, or controlled for the workload mix that was running. Figures 2 and 3 show power traces but no raw data or error bars, so the reader cannot assess whether the reduction is robust to baseline choices. A rigorous, explicitly defined baseline is load-bearing for the central result and must be provided.
  2. [Section 2.2 / Section 4.4] The claims that 'every experiment performed as expected' and that there were 'zero SLA violations' are unquantified. Section 4.4 reports that compliance thresholds were fully maintained and zero SLA violations occurred across 33 experiments and 212 jobs, but no definition of an SLA violation is given (e.g., tolerance for throughput degradation versus the Flex-tier limits), no per-experiment or per-job results are tabulated, and no audit trail or telemetry extracts are provided. These strong universal claims need a concrete metric definition and supporting data, or they should be substantially qualified.
minor comments (5)
  1. [Methods 4.4] The text contains a typo: 'NVDIA-SMI' should be 'NVIDIA-SMI.'
  2. [Figure 6] Figure 6 shows throughput versus power cap for eight workloads but includes no error bars or indication of run-to-run variability, which makes reported differences between workloads hard to interpret.
  3. [Section 3.4] The phrase 'received demand response credits in capacity or ancillary markets' should be 'receive demand response credits in capacity or ancillary markets' (verb form inconsistency).
  4. [Table 1] The column header '# Nodes' is followed by values such as '8' and '6'; clarifying that each node contains 8 A100 GPUs in the table header or a footnote would improve readability.
  5. [References] Reference [9] (Sivaram, 'Taming the Sun') is a general book on solar energy and does not appear to support the specific statement about GPU-driven AI workloads containing operational flexibility; consider replacing it with a directly relevant citation.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the 25% reduction result is externally benchmarked and independently measured; the simulator RMSE is an in-sample consistency metric but is not load-bearing.

full rationale

The central field-trial claim is not circular. The 25% power-reduction target was set externally by APS and SRP ('we set stringent targets... Each event required the cluster to reduce power by 25%'), and compliance was measured from NVIDIA-SMI GPU power telemetry (Methods 4.4) against a baseline, not computed from the controller's own setpoints. The QoS claim is checked against pre-agreed flexible SLA tiers (Flex 0-3) defined in Section 2.1, so zero SLA violations is a contractual compliance check rather than a redefinition of the outcome. The Emerald Simulator's 4.52% RMSE is an in-sample consistency metric: Methods 4.3 says jobs were profiled in advance and those estimated power-performance relationships drove runtime control decisions, and Methods 4.4 evaluates RMSE on the same experiments. This limits the generality of the simulator validation, but the 25% reduction result does not rest on out-of-sample predictive power. The self-citations ([5], [6]) are background on earlier HPC demand-response work and are not load-bearing for the present demonstration. The Limitations section's statement that 'full data center telemetry' is needed for system-level impacts is a measurement-scope caveat, not a circular step. On balance, the derivation chain is self-contained with no definitional or self-citation circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central result is an empirical measurement, so the ledger is dominated by measurement scope and representativeness assumptions rather than derived free parameters. The simulator's fitted models are the main hidden parameters, and the flexibility tier choices are design variables that influence feasibility.

free parameters (2)
  • Emerald Simulator power-performance model parameters = not disclosed
    Profiled each job in advance (Sec. 4.3) and used estimated power-performance relationships; the model is trained on the same cluster and used for control and accuracy evaluation.
  • Flexibility tier thresholds (Flex 1, Flex 2, Flex 3) = 10%, 25%, 50% allowed throughput reduction
    Chosen in consultation with partners (Sec. 2.1); they define acceptable QoS degradation and thus bound the achievable power reduction. These are design choices, not fitted to data directly, but they materially affect the result.
assumptions (3)
  • domain assumption GPU power measured via NVIDIA-SMI is a valid proxy for grid-level data center demand reduction.
    Methods 4.4 measures GPU power only; the paper frames results as grid-interactive, implicitly assuming facility loads scale proportionally or are negligible.
  • domain assumption The selected workload ensembles and flexibility tiers are representative of real production AI workloads.
    Methods 4.1 states ensembles were designed with partners; there is no evidence they reflect actual customer traffic or that customers would accept such throttling.
  • ad hoc to paper Checkpointing overhead is negligible for demand response events.
    Sec. 3.3 asserts this based on training jobs running for days; it is not measured in the trial and could affect pause/resume efficiency.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Turning AI Data Centers into Grid-Interactive Assets: Results from a Field Demonstration in Phoenix, Arizona." pith.science (2026). https://pith.science/paper/SWHYHNW4

@misc{pith2026250700909,
  author       = {Pith},
  title        = {Pith review of: Turning AI Data Centers into Grid-Interactive Assets: Results from a Field Demonstration in Phoenix, Arizona},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SWHYHNW4}},
  note         = {Machine review of arXiv:2507.00909}
}
read the original abstract

Artificial intelligence (AI) is fueling exponential electricity demand growth, threatening grid reliability, raising prices for communities paying for new energy infrastructure, and stunting AI innovation as data centers wait for interconnection to constrained grids. This paper presents the first field demonstration, in collaboration with major corporate partners, of a software-only approach--Emerald Conductor--that transforms AI data centers into flexible grid resources that can efficiently and immediately harness existing power systems without massive infrastructure buildout. Conducted at a 256-GPU cluster running representative AI workloads within a commercial, hyperscale cloud data center in Phoenix, Arizona, the trial achieved a 25% reduction in cluster power usage for three hours during peak grid events while maintaining AI quality of service (QoS) guarantees. By orchestrating AI workloads based on real-time grid signals without hardware modifications or energy storage, this platform reimagines data centers as grid-interactive assets that enhance grid reliability, advance affordability, and accelerate AI's development.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 17 canonical work pages

  1. [1]

    https://doi.org/10.17226/29101

    National Academies of Sciences, Engineering, and Medicine.Implications of Artificial Intelli- gence–Related Data Center Electricity Use and Emissions: Proceedings of a Workshop(The National Academies Press, Washington, DC, 2025). https://doi.org/10.17226/29101

    work page doi:10.17226/29101 2025

  2. [2]

    Data center power and energy management: Past, present, and future.IEEE Micro44, 30–36 (2024)

    Ricardo Bianchini, Christian Belady & Anand Sivasubramaniam. Data center power and energy management: Past, present, and future.IEEE Micro44, 30–36 (2024). https://doi.org/10.1109/ MM.2024.3426478

    work page arXiv 2024

  3. [3]

    Electricity 2025: Analysis and forecast to 2027

    Eren C ¸ am, Marc Casanovas & John Moloney. Electricity 2025: Analysis and forecast to 2027. IEA (2025). https://www.iea.org/reports/electricity-2025

    work page 2025

  4. [4]

    Powering intelligence: Analyzing artificial intelligence and data center energy consumption

    Jordan Aljbour, Tom Wilson & Poorvi Patel. Powering intelligence: Analyzing artificial intelligence and data center energy consumption. Electric Power Research Institute (EPRI) (2024). https: //www.epri.com/research/products/000000003002028905

    arXiv 2024

  5. [5]

    Paschalidis & Ayse K

    Yijia Zhang, Daniel Curtis Wilson, Ioannis Ch. Paschalidis & Ayse K. Coskun. HPC data center participation in demand response: An adaptive policy with QoS assurance.IEEE Transactions on Sustainable Computing7, 157–171 (2022). https://doi.org/10.1109/TSUSC.2021.3077254

    work page arXiv 2022

  6. [6]

    Wilson, Ioannis Ch

    Yijia Zhang, Daniel C. Wilson, Ioannis Ch. Paschalidis & Ayse K. CoskunA data center demand response policy for real-world workload scenarios in HPC. In2021 Design, Automation & Test in Europe Conference & Exhibition (DATE), 282–287 (2021)

    work page 2021

  7. [7]

    Jiali Xing, Bilge Acun, Aditya Sundarrajan, David Brooks, Manoj Chakkaravarthy, Nikky Avila, Carole-Jean Wu & Benjamin C. Lee. Carbon responder: Coordinating demand response for the datacenter fleet (2023). arXiv:2311.08589

    arXiv 2023

  8. [8]

    Carbon-aware computing for datacenters.IEEE Transactions on Power Systems38, 1270–1280 (2023)

    Ana Radovanovi´ c, Ross Koningstein, Ian Schneider, Bokan Chen, Alexandre Duarte, Binz Roy, Diyue Xiao, Maya Haridasan, Patrick Hung, Nick Care, Saurav Talukdar, Eric Mullen, Kendal Smith, MariEllen Cottman & Walfredo Cirne. Carbon-aware computing for datacenters.IEEE Transactions on Power Systems38, 1270–1280 (2023). https://doi.org/10.1109/TPWRS.2022.3173250

    arXiv 2023

Show all 21 references
  1. [9]

    ISBN: 978-0-262-53707-0

    Varun Sivaram.Taming the Sun: Innovations to Harness Solar Energy and Power the Planet(The MIT Press, Cambridge, Massachusetts, 2018). ISBN: 978-0-262-53707-0

    2018

  2. [10]

    Using cost–benefit analyses to identify key opportunities in demand-side mitigation.Nature Climate Change14, 1158–1164 (2024)

    Jie-Sheng Tan-Soo, Ping Qin, Yifei Quan, Jun Li & Xiaoxi Wang. Using cost–benefit analyses to identify key opportunities in demand-side mitigation.Nature Climate Change14, 1158–1164 (2024). https://doi.org/10.1038/s41558-024-02146-4

    2024 doi

  3. [11]

    Rethinking load growth: assessing the potential for integration of large flexible loads in US power systems

    Tyler Norris, Timothy Profeta, Dalia Patino-Echeverri & Adam Cowie-Haskell. Rethinking load growth: assessing the potential for integration of large flexible loads in US power systems. Nicholas Institute for Energy, Environment & Sustainability (2025). https://hdl.handle.net/1...

    2025

  4. [12]

    Final root cause analysis: Mid-august 2020 extreme heat wave

    Elliot Mainzer, Marybel Batjer & David Hochschild. Final root cause analysis: Mid-august 2020 extreme heat wave. California Independent System Operator (CAISO) (2021). https://www.caiso. com/Documents/Final-Root-Cause-Analysis-Mid-August-2020-Extreme-Heat-Wave.pdf

    2021

  5. [13]

    ChienExploding AI power use: an opportunity to rethink grid planning and management

    Liuzixuan Lin, Rajini Wijayawardana, Varsha Rao, Hai Nguyen, Emmanuel Wedan GNIBGA & Andrew A. ChienExploding AI power use: an opportunity to rethink grid planning and management. InProceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems, ...

    2024

  6. [14]

    In Proceedings of the 2023 ACM Symposium on Cloud Computing, SoCC ’23, 588–596 (Association for Computing Machinery, New York, NY, USA, 2023)

    Dan Zhao, Siddharth Samsi, Joseph McDonald, Baolin Li, David Bestor, Michael Jones, Devesh Tiwari & Vijay GadepallySustainable supercomputing for AI: GPU power capping at HPC scale. In Proceedings of the 2023 ACM Symposium on Cloud Computing, SoCC ’23, 588–596 (Association for...

    2023

  7. [15]

    The online pause and resume problem: Optimal algorithms and an application to carbon-aware load shifting.Proc

    Adam Lechowicz, Nicolas Christianson, Jinhang Zuo, Noman Bashir, Mohammad Hajiesmaili, Adam Wierman & Prashant Shenoy. The online pause and resume problem: Optimal algorithms and an application to carbon-aware load shifting.Proc. ACM Meas. Anal. Comput. Syst.7(2023). https://d...

    2023 doi

  8. [16]

    Grid flexibility needs and data center characteristics

    Anuja Ratnayake, Irene Danti Lopez, Baskar Vairamohan & Eamonn Lannoye. Grid flexibility needs and data center characteristics. Electric Power Research Institute (EPRI) (2025). https: //www.epri.com/research/programs/063638/results/3002031504

    2025

  9. [17]

    Electricity rate designs for large loads: Evolving practices and opportunities

    Andrew Satchwell, Natalie Mims Frick, Peter Cappers, Sanem Sergici, Ryan Hledik, Goksin Kavlak & Glenda Oskar. Electricity rate designs for large loads: Evolving practices and opportunities. Lawrence Berkeley National Laboratory (2025). Retrieved from https://escholarship.org/...

    2025

  10. [18]

    Chien & Sangwon Suh

    Jiajia Zheng, Andrew A. Chien & Sangwon Suh. Mitigating curtailment and carbon emissions through load migration between data centers.Joule4, 2208–2222 (2020). https://doi.org/10.1016/ j.joule.2020.08.001

    2020

  11. [19]

    Supporting power grids with demand response at google data centers

    Varun Mehra & Raiden Hasegawa. Supporting power grids with demand response at google data centers. Google Cloud Blog (2023). https://cloud.google.com/blog/products/infrastructure/ using-demand-response-to-reduce-data-center-power-consumption

    2023

  12. [20]

    Survey of U.S

    Zhi Zhou, Todd Levin & Guenter Conzelmann. Survey of U.S. ancillary services markets. Argonne National Lab (ANL) (2016). https://www.osti.gov/biblio/1236451

    2016

  13. [21]

    Whitepaper retrieved from https://www.ercot.com/files/ docs/2024/10/07/ERCOT-Ancillary-Services-Study-Final-White-Paper.pdf

    ERCOT ancillary services study (2024). Whitepaper retrieved from https://www.ercot.com/files/ docs/2024/10/07/ERCOT-Ancillary-Services-Study-Final-White-Paper.pdf. 10

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.