From estimates to evidence:
a research agenda
What data exists, what's missing, and how Northeastern could lead on AI sustainability measurement — across all campuses.
What data exists — and where the gaps are
Every source mapped by availability and measurement precision. Green sources are ready to use today. Orange and red sources represent the research frontier.
Available Now
3 sourcesCampus Energy Consumption Data
EstimatedBuilding-level kWh from Facilities — provides baseline for any on-campus compute infrastructure
Potential: Baseline for campus-hosted compute; context for off-campus API usage share
Software License Counts
ExactGitHub Copilot, Microsoft 365 Copilot, and other enterprise AI seat licenses procured by NU
Potential: Hard lower bound on active user counts — useful for calibration
Vendor Carbon Disclosure Reports
UnknownOpenAI, Anthropic, Google, and Microsoft annual sustainability / GHG disclosure documents
Potential: Tool-level footprint context — but vendor figures are often incomplete or aggregated
Requires Survey
1 sourceStudent & Faculty AI Usage Surveys
EstimatedIRB-approved self-report surveys on tool adoption, query frequency, and use case breakdown
Potential: Currently the only source of use-case and adoption data at peer institutions
Requires IT Access
2 sourcesNU IT Network Traffic Logs
ExactPacket-level or flow-level data showing traffic to AI service endpoints (OpenAI, Anthropic, GitHub Copilot APIs)
Potential: Direct proxy for query volume — the gold standard for usage measurement
Cloud Compute Bills
ExactAWS, Azure, or GCP invoices for any NU-operated AI infrastructure (research clusters, hosted models)
Potential: Direct measurement of on-premises AI energy; includes GPU utilization data
Does Not Exist Yet
1 sourceNU Data Center PUE
UnknownPower Usage Effectiveness ratio for any NU-operated server rooms or co-location facilities
Potential: Critical multiplier for on-campus compute footprint; most universities lack this
Three questions the Incubator could own
Each is tractable in 6–12 months, publishable, and positions Northeastern as the institution that defined the measurement standard for AI sustainability in higher education.
“What is the actual AI-related energy consumption at Northeastern, and how does it compare to peer institutions?”
Proposed methodology
- 1
Deploy anonymized API traffic monitoring on NU's network edge in partnership with OIT — captures query volume without content inspection
- 2
Combine with software license data (Copilot seats, Microsoft 365 AI features) to establish hard lower bounds on user counts across colleges
- 3
Run parallel survey (n ≥ 500, stratified by college and role) to capture unmanaged tool usage (personal API keys, consumer ChatGPT accounts)
Expected outputs
- Peer-reviewed baseline report
- Reusable measurement framework
- Comparison dataset across 3–5 R1 peers
“Which AI use cases on campus generate the most value per unit of carbon — and which are low-value, high-energy waste?”
Proposed methodology
- 1
Pair usage data with academic outcome proxies (grade distributions, submission rates, research output) to model value per use case at the aggregate level
- 2
Conduct qualitative interviews (n = 40) with students and faculty across high- and low-adoption departments to surface task taxonomies and perceived value
- 3
Apply a carbon-cost-effectiveness lens borrowed from health economics: cost-per-unit-of-impact (QALY equivalent for academic productivity)
Expected outputs
- AI use-case carbon efficiency rankings
- Policy memo for NU Provost
- Framework for responsible AI procurement criteria
“How do students and faculty think about the environmental cost of the AI tools they use daily?”
Proposed methodology
- 1
Mixed-methods survey (quantitative scales + qualitative probes) measuring awareness, concern, and willingness to change behavior — administered at semester start and end
- 2
A/B test: show a random half of survey respondents this NUCarbon dashboard before answering — measure whether data exposure shifts reported attitudes or intended behavior
- 3
Ground findings in existing frameworks: Stern's Value-Belief-Norm theory, tech acceptance models, and campus sustainability behavior literature
Expected outputs
- First dataset on AI carbon awareness in higher ed
- Behavioral intervention design brief
- Journal article: Environmental Communication or Computers & Education
Northeastern's 24,000-person community is an extraordinary living laboratory
Distributed across 7 cities and 3 countries, using the same AI tools against different grid intensities, cultural norms, and academic contexts — this is a natural experiment that no single-site study can replicate.
A standardized measurement framework developed at Northeastern could become a model for universities worldwide. AASHE has 1,000+ member institutions actively looking for exactly this kind of replicable methodology. The first mover advantage is real, and the moment is now — before AI carbon accounting becomes regulated and the easy wins are gone.
NU Campuses — 7 cities, 3 countries
Each campus sits on a different regional grid with a different carbon intensity. Boston (NE-ISO: 0.386 kg/kWh) vs. Oakland (WECC: 0.214 kg/kWh) produces a natural control for grid effects on the same AI usage pattern — a ready-made comparative study.
A 12-month path to publication
Achievable with one researcher, one faculty advisor, and seed funding. Every milestone produces a standalone deliverable.
Data Audit
Audit existing data sources, interview IT and Facilities to map what's accessible and on what timeline
Survey Deploy
Deploy usage survey to 500-student stratified sample; simultaneously seek IRB approval for behavioral arm
Baseline Report
Publish open baseline report with full methodology; submit NUCarbon framework to Sustainability office for endorsement
Incubator Demo
Present findings at Sustainability Innovation Week; demo live dashboard to VP of Sustainability and peer institution contacts
Publication
Submit for peer review; share open-source measurement framework with AASHE member institutions and Second Nature GHG Protocol working group
This dashboard is a starting point, not the answer.
The methodology is documented, the gaps are mapped, and the research questions are tractable. What it needs now is institutional commitment and a researcher with the access to close the loop.