University of Michigan

Informed voices for infrastructure choices

The CIVIC Forum is building the civic discourse infrastructure that connects hardware measurements to policy frameworks, so communities can make evidence-based decisions about AI infrastructure that affects their energy, water, and future.

$1T US datacenter investment projected by 2030
$98B in projects blocked in Q2 2025 alone
2-10x variance in AI energy estimates

AI infrastructure decisions are failing everyone

When a datacenter requests grid connection, utilities conduct interconnection studies over 6 to 18 months at roughly $500K, producing reports comprehensible only to power engineers. What follows is predictable: utilities request billions for grid upgrades, consumer advocates resist ratepayer burdens, communities mobilize against "mystery power users," and datacenter operators promise economic benefits. This pattern is incapable of sustaining AI growth.

We have identified four systematic gaps preventing evidence-based deliberation:

GAP 01

Energy data relies on estimates

Energy estimates for AI systems vary 2 to 10x between calculations and direct measurement. Datacenter operators report figures without independent verification. Without verified measurements, stakeholders operate from incompatible assumptions.

GAP 02

Datacenters are treated as grid burdens

Interconnection studies treat datacenters as rigid loads, ignoring that AI workloads can shift training by hours or defer inference by seconds. This flexibility could reduce infrastructure costs by billions if coordinated.

GAP 03

Grid-AI futures rely on point forecasts

Current analyses assume a single future: this datacenter operating at this capacity for this duration. Yet uncertainties around datacenter counts, AI efficiency, and energy sources can swing infrastructure costs by billions.

GAP 04

Dialogue is technical and top-down

Technical outputs remain incomprehensible to non-experts. Residents cannot ask "how will this affect my electricity bill or water supply?" and get evidence-based answers. Discourse defaults to worst-case fears versus corporate reassurances.

The CIVIC Stack

Democratic infrastructure governance requires vertical integration: hardware measurements must connect to policy frameworks through validated layers that preserve technical accuracy while enabling public participation. Each layer builds on the one below it.

Layer 01

Hardware Measurements

Direct measurement of AI energy consumption and water use intensity establishes shared factual ground truth. Our measurement database provides the first independently verified, publicly accessible data for AI infrastructure impacts, building the factual foundation that evidence-based deliberation requires.

Chowdhury · Dvorkin · Love
Layer 02

Grid Impact Evaluation

Our testbed measures how temporal flexibility in AI training and inference workloads affects distribution infrastructure: voltage profiles, transformer loading, and power quality. By showing datacenters can be grid assets rather than burdens, we shift the conversation from "how much will this cost?" to "how should we coordinate this resource?"

Dvorkin · Mathieu
Layer 03

Scenario Exploration

Macro-energy system models explore thousands of potential grid-AI futures under deep uncertainty. Rather than single-point forecasts, communities can evaluate how decisions perform across scenarios, identify adaptive strategies, and understand which uncertainties matter most.

Craig · Rodríguez
Layer 04

Democratic Interface

AI agents synthesize outputs from all layers into interfaces where regulators, citizens, utilities, and datacenter operators explore tradeoffs between grid stress, electricity bills, water consumption, and emissions. Technical complexity becomes legible to all stakeholders through transparent uncertainty quantification.

Klass · Love · Ullman · Rodríguez · Chowdhury

Grounded in real decisions, built to scale

This is not a hypothetical exercise. Michigan communities are facing datacenter siting decisions right now, and the university has the interdisciplinary depth to address them.

Case Study

Ypsilanti Township

A proposed datacenter development prompted community concern about energy costs, water use, and environmental impact. Residents lacked access to verified data about actual resource demands, forcing deliberation based on estimates and competing narratives rather than evidence.

Case Study

City of Saline

Datacenter siting decisions here highlighted the gap between technical interconnection studies and public understanding. The community engagement process revealed how inaccessible technical outputs prevent meaningful democratic participation in infrastructure choices.

The University of Michigan uniquely combines ML systems research, power systems laboratories, energy planning expertise, water systems knowledge, AI capabilities, energy law scholarship, and community engagement experience. Existing partnerships with utilities, regulators, and community organizations provide the relationships needed to translate research into practice.

ML.ENERGY benchmark Power grid testbeds Macro-energy modeling Water systems assessment AI agent frameworks Energy law & regulation Community engagement (IES) Ginsberg Center partnership

An interdisciplinary team across six departments

Mosharaf Chowdhury
Lead PI
Computer Science & Engineering
ML systems, ML.ENERGY benchmark, energy measurement infrastructure
Michael Craig
Co-PI
Environment & Sustainability
Macro-energy systems, robust decision-making under uncertainty
Vladimir Dvorkin
Co-PI
Electrical & Computer Engineering
Power systems, grid testbed, distribution infrastructure analysis
Alexandra Klass
Co-PI
Law School
Energy law, regulatory frameworks, policy translation
Nancy Love
Co-PI
Civil & Environmental Engineering
Water systems, infrastructure assessment, environmental impact
Johanna Mathieu
Co-PI
Electrical & Computer Engineering
Grid coordination protocols, real-time control algorithms
Alexander Rodríguez
Co-PI
Computer Science & Engineering
AI for complex systems, agent frameworks, scenario modeling
Amanda Ullman
Community Engagement Lead
Institute for Energy Solutions
Community engagement, energy transition impacts, stakeholder processes

From Michigan proof-of-concept to national model

Within five years, we will demonstrate that democracies can govern planetary-scale AI infrastructure through legitimate deliberative processes, creating replicable models for infrastructure decisions nationwide.

Year 1

Measurement Foundation

Deploy verified energy and water measurement infrastructure. Establish ground truth database with independently verified AI consumption data. Launch community engagement with Ypsilanti and Saline stakeholders.

Year 2

Grid Testbed & Flexibility

Demonstrate datacenter workload flexibility on physical grid testbed. Publish first benchmark results showing coordination potential. Begin scenario model development with utility partners.

Year 3

Integrated Stack

Connect all four CIVIC Stack layers into working prototype. Pilot AI-powered stakeholder interfaces with Michigan communities. Launch certificate program for practitioners.

Years 4-5

Scaling & Adoption

Expand to multi-state regulatory contexts. Pursue federal funding alignment with DOE programs. Establish sustainable revenue through certificate programs and philanthropic partnerships.

Advancing Michigan's strategic priorities

Energy, Climate Action, Sustainability

Building the measurement and modeling infrastructure for evidence-based energy and water governance in the age of AI.

Advanced Technology

Creating verified benchmarks and open tools that establish Michigan as the national leader in AI infrastructure assessment.

Democracy, Civic & Global Engagement

Demonstrating that democratic processes can govern planetary-scale technology decisions through informed civic discourse.

Join the conversation

Whether you are a policymaker, community member, researcher, utility, or datacenter operator, the CIVIC Forum is building tools and frameworks for you.

The CIVIC Forum
University of Michigan
hello@thecivic.forum