Skip to content

GridNinja evidence

Inspect the evidence behind a capacity decision.

Start with a versioned decision brief. Check its requested increment, modeled result, conditions, and unresolved question, then inspect the matching technical record. These synthetic examples explain the assessment method; they do not establish customer results or physical validation.

Published synthetic decision briefs

Four authored scenarios show how a request, its conditions, and missing evidence change the decision. Each brief has a matched PDF and technical record. These are illustrative assessments, not customer outcomes or operating permission.

Inspect the assessment and equipment relationships →

Research informing the direction

Peer-reviewed research helps frame the questions. It does not validate GridNinja's models, demonstrate customer outcomes, or confer operating authority.

These seven peer-reviewed journal articles inform different questions. Their methods, settings and limitations matter as much as their findings.

  1. Field demonstration

    AI data centres as grid-interactive assets

    Philip Colangelo; Ayse K. Coskun; Jack Megrue; Ciaran Roberts; Shayan Sengupta; Varun Sivaram; Ethan Tiao; Aroon Vijaykar; Chris Williams; Daniel C. Wilson; Brandon Records; Zack MacFarland; Daniel Dreiling; Nathan Morey; Anuja Ratnayake; Baskar Vairamohan.

    Nature Energy 11(2), 254–261 (2026). DOI: 10.1038/s41560-025-01927-1

    Cite the 2026 volume; the online version and DOI carry 2025.

    Setting
    Phoenix, Arizona, United States. One 256-GPU cluster with representative AI workloads. Three-hour peak-demand demonstration.
    Supported idea
    An external field trial shows that eligible AI workloads can provide bounded power flexibility while meeting the tested service requirements.

    Study limits

    • Selected cluster and workloads, not a campus-wide guarantee.
    • No GridNinja result or measured community benefit.
    • The external 25% result must not be attached to a GridNinja capacity metric.

    Interests and context: Emerald AI-led work; disclosed employment, equity and pending-patent interests. No GridNinja endorsement is implied.

    Review and access: Publisher abstract, metadata, affiliations and competing-interest statement. Public publisher material reviewed; unrestricted full-method access and commercial figure rights not established.

    Source review: . The Crossmark record reported this document current on that date.

  2. Power-system modeling

    Flexible data centers reduce power system costs but can increase emissions

    Juan Ramon L. Senga; Shen Wang; Christopher R. Knittel.

    iScience 29(7), 116497 (2026). DOI: 10.1016/j.isci.2026.116497

    PubMed indexes 2026-06-26 online and 2026-07-17 collection; exact online day not independently confirmed in accessible primary metadata.

    Setting
    Texas, Mid-Atlantic and Western Interconnection, United States. Regional investment and dispatch scenarios with varying temporal flexibility. Modeled 2030 scenarios.
    Supported idea
    In modeled U.S. power systems, temporal flexibility lowered system costs, while emissions outcomes depended on regional conditions.

    Study limits

    • Scenario modeling, not observed household bills or site economics.
    • Flexibility is treated as cost-free in the model.
    • Emissions increased in some modeled regional settings.

    Interests and context: No independence claim made; competing-interest details were not specifically audited.

    Review and access: Publicly indexed publisher summary, results, discussion and model-assumption text. Direct ScienceDirect/Cell retrieval returned 403; indexed publisher sections were accessible. Exact primary online date remains unverified.

    Source review: . The Crossmark record reported this document current on that date.

  3. Spatial footprint modeling

    The environmental footprint of data centers in the United States

    Md Abu Bakar Siddik; Arman Shehabi; Landon Marston.

    Environmental Research Letters 16(6), 064017 (2021). DOI: 10.1088/1748-9326/abfba1

    Online May 2021; June issue. Institution-hosted journal PDF supplies article history.

    Setting
    United States. Data-center cooling, electricity supply and watershed-level water stress.
    Supported idea
    Water responsibility requires considering both site cooling and the electricity supply in their local resource context.

    Study limits

    • Historical modeled estimates, not a 2026 inventory.
    • No site-specific hydrological conclusion or GridNinja water savings.
    • Do not reuse historical national percentages as current facts.

    Interests and context: No independence claim made; competing-interest details were not specifically audited.

    Review and access: Journal abstract/header and institutional publication record. Public institutional PDF available; complete geospatial inputs and calculations not reproduced.

    Source review: . The Crossmark record reported this document current on that date.

  4. Production-system evaluation

    Carbon-Aware Computing for Datacenters

    Ana Radovanović; 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.

    IEEE Transactions on Power Systems 38(2), 1270–1280 (2023). DOI: 10.1109/TPWRS.2022.3173250

    IEEE online date is 2022; issue citation is 2023. Do not substitute the 2021 arXiv date.

    Setting
    Google's distributed data-center fleet. Eligible temporally flexible workloads under forecast-informed hourly capacity limits.
    Supported idea
    Production research shows that temporally flexible computing can be scheduled around carbon signals while preserving a daily completion objective.

    Study limits

    • Completion requirement is probabilistic and limited to eligible workloads.
    • No general SLA guarantee, congestion relief or water benefit follows.
    • No evidence that GridNinja has implemented this system.

    Interests and context: Authors report on Google's production system; do not describe this as independent validation of GridNinja.

    Review and access: IEEE abstract and bibliographic metadata. University-hosted PDF indexed but direct retrieval failed; full methods and implementation not reviewed.

    Source review: . A Crossmark status record was unavailable; the update check used publisher-indexed content, deposited metadata and targeted notice searches.

  5. Survey experiment

    How sovereign control, decarbonization and energy costs shape public support for data centers

    Jonas Heering; Erik Voeten.

    Nature Communications 17(1), 9576 (2026). DOI: 10.1038/s41467-026-76500-9

    First published 10 August 2026; the current publisher page dates the version of record 8 September 2026.

    Setting
    Germany. Public vignette and conjoint survey experiments concerning data-center proposals. July 2025 survey.
    Supported idea
    In a German survey experiment, power prices, water strain, energy sources and operator ownership influenced preferences over data-center proposals.

    Study limits

    • Stated preferences in Germany, not U.S. community consent or permitting outcomes.
    • Ownership was especially influential; do not omit it when interpreting the findings.
    • Does not test GridNinja's message or establish conversion uplift.

    Interests and context: Publisher states that the authors declare no competing interests.

    Review and access: Publisher abstract, results, discussion, methods, metadata and interest declaration. Public full-text sections available; microdata, code and peer-review correspondence not reanalyzed.

    Source review: . The Crossmark record reported this document current on that date.

  6. District-heating modeling

    Utilizing data center waste heat in district heating – Impacts on energy efficiency and prospects for low-temperature district heating networks

    Mikko Wahlroos; Matti Pärssinen; Jukka Manner; Sanna Syri.

    Energy 140, 1228–1238 (2017). DOI: 10.1016/j.energy.2017.08.078

    Journal PDF gives online 29 August; Aalto's publication date is 1 December 2017.

    Setting
    Espoo, Finland. District-heating production model with actual plant/heat-demand data and an existing data-center load profile. 2013 and 2015 heat-system data.
    Supported idea
    A Finnish case study shows how data-center waste heat can contribute to district heating when the local system and commercial conditions fit.

    Study limits

    • Local demand, temperatures, equipment, network and counterparties determine applicability.
    • Do not transfer the study's economics into GridNinja examples.
    • Heat reuse is a conditional ecosystem opportunity, not a shipped capability or default assessment deliverable.

    Interests and context: Four authors confirmed by article PDF, Aalto record and Crossref; disregard a separate dissertation listing that adds Rinne. Competing interests not specifically audited.

    Review and access: Journal title page, abstract and introductory methodological sections; Aalto article record. Institution-hosted full PDF available; simulation not reproduced. CC BY-NC-ND does not authorize commercial figure adaptation.

    Source review: . The Crossmark record reported this document current on that date.

  7. Control theory and simulation

    A predictive safety filter for learning-based control of constrained nonlinear dynamical systems

    Kim Peter Wabersich; Melanie N. Zeilinger.

    Automatica 129, 109597 (2021). DOI: 10.1016/j.automatica.2021.109597

    Exact initial online day not verified. Use July 2021 journal citation, not the 2018 preprint date.

    Setting
    Not applicable: theoretical work and simulations. Constrained nonlinear control with pendulum and quadrotor simulation examples.
    Supported idea
    Safety-filter research separates proposed actions from constraint checks that can modify an action before execution.

    Study limits

    • Guarantees depend on model and uncertainty assumptions.
    • Not data-center field validation or GridNinja certification.
    • Supports the runtime-assurance development direction, not a claim of shipped autonomous control.

    Interests and context: No independence claim made; competing-interest details were not specifically audited.

    Review and access: Publisher abstract, introduction and conclusion preview. Some full content requires subscription access; full proof details and experiment files not reviewed.

    Source review: . The Crossmark record reported this document current on that date.

External research informs the direction. GridNinja currently offers bounded assessments using authorized historical inputs.

No correction or retraction notice was found in the records checked on 10 October 2026. This is a bounded source check, not a guarantee that no update exists. Full methods, data and code were not independently reproduced.

How to use this evidence

Read the scope with the result

The synthetic assessment has a versioned decision brief, PDF, and technical record. Other resources remain unavailable until their evidence, accountable reviewers, and publication approval are established.

Resources awaiting publication (4)

These resources are not yet published. Their status pages explain what is available today; they are not technical evidence.

Capacity assessment

Bring the operator question before the capacity claim

A bounded assessment reviews one capacity question, its evidence, constraints, and unresolved commercial decisions.

Scope an assessment