Pulsix AI Research Program
Data centers, grid, and community optimization
How can existing buildings, data centers, distributed batteries, and compute operate together more efficiently, while maintaining reliability and sharing value with owners and communities? Pulsix AI research answers that in four phases, starting from public data.
- SDVOSB certified (SBA)
- Texas VetHUB — VID 21829539
- SAM.gov UEI J6NATC3XXUK5
- CAGE 12LP3
- TDLR licensed — TACLA00169073C
From roofs to coupled infrastructure
- 01
Phase I · Public data: observe the Texas data-center fleet
"The Invisible Tax" examined whether aging roof performance creates a cooling, water, and peak-demand burden that conventional planning overlooks, tracking roof geometry and nine years of satellite spectral history across a cohort of 403 Texas data centers.
- 02
Phase II · Modeled: model the facility as a system
Two facilities with similar roofs can respond differently. Phase II links weather and envelope behavior to cooling demand, fans and pumps, heat rejection, electrical losses, facility power, and productive IT capacity, with carbon and water attached through explicit boundaries.
- 03
Phase III · Owner data required: replace assumptions with owner data
Equipment schedules and 60 to 90 days of read-only BAS/BMS history establish which configuration actually applies. Predictions are locked before an intervention, then compared with normalized measured results.
- 04
Phase IV · Owner data required: learn across facilities
Repeated cases build a structured record (configuration, conditions, predicted and measured response, uncertainty) toward a hybrid physical-infrastructure intelligence model.
Every case is labeled by its evidence
Cases start with public information and become more specific as authorized owner data replaces assumptions. We never present a screening result as a verified one.
| Tier | Typical inputs | Appropriate outputs |
|---|---|---|
| Public screening | Satellite imagery, weather, public grid and price data, geometry, engineering references | Observable exterior state, configuration hypotheses, scenario ranges |
| Owner-enriched | As-builts, equipment schedules, maintenance history, tariffs, operating limits | Corrected topology, equipment-specific baselines, a targeted measurement plan |
| Operationally calibrated | BAS/BMS, meters, IT load, battery or node telemetry | Normalized residuals, constraint-aware operating opportunities |
| Intervention-validated | Locked prediction, treatment event, controls, post-event measurements | Verified physical response, persistence, attributable value |
How a case works
- 01
Register
Register the site, systems, and owner.
- 02
Agree
Agree data rights, access, and publication scope.
- 03
Ingest
Ingest public observations and authorized owner records.
- 04
Reconstruct
Reconstruct the physical asset graph.
- 05
Baseline
Build a versioned expected-state baseline.
- 06
Evaluate
Evaluate interventions and operating scenarios.
- 07
Rank
Rank opportunities by value, confidence, and risk.
- 08
Scope
Issue an approved scope or bounded recommendation.
- 09
Measure
Measure the result against the locked prediction.
- 10
Update
Update the evidence, publish the approved case, monitor persistence.
Research controls
- Every material result carries an evidence label; missing or stale sources stay visible.
- Predictions are locked before treatment and compared with normalized measurements.
- Equipment or control changes start a new operating regime instead of rewriting history.
- Negative results and corrected assumptions remain part of the record.
- Satellite correlations are treated as associations; savings claims require intervention evidence.
Featured case study: Crusoe Abilene (Lancium Clean Campus)
Our principal example of moving from remotely observed exterior conditions to a whole-system model, built entirely from public data: satellite spectral and thermal observations, weather, ERCOT pricing, manufacturer baselines, and explicitly labeled configuration assumptions. The model points to the cooling plant, not the roof, as where the most efficiency headroom is likely to be. Pulsix AI is not affiliated with the campus owner or operator. Figures in the case study are public-data screening and modeled values, not measured facility performance.
The distributed energy and compute track
Beyond individual facilities, we are researching a community digital twin: homes, commercial buildings, batteries, compute nodes and the campus interface, each with its own baseline, telemetry, constraints and economic ledger. This work is R&D in progress. These collaborations are in active discussion; no agreements are formalized.
- Distributed compute: workload requirements, measured node power, availability and economics for pairing compute with suitable sites.
- Residential energy storage: a household battery and fleet pathway for home-to-community flexibility.
- Campus power and grid: campus-scale power orchestration, interconnection limits and grid-coordination constraints.
- Mechanical and water treatment: cooling-loop chemistry and maintenance evidence for measured-facility pilots.
Questions
- What is a Pulsix AI data center evaluation?
- We start from public data, then show exactly which owner data would turn screening results into a verified plan. Access is read-only, and nothing is published without your approval.
- Are Pulsix AI case studies endorsed by the facilities they study?
- No. Public case studies use public data only, label every figure as observed, modeled, or owner-required, and Pulsix AI is not affiliated with the owners or operators of the campuses it studies.