Cooling instruments and software for AI data centers
Know when a liquid-cooled chip is running out of cooling margin.
BCP Technologies is building a sensor-plus-software instrument for cold plates in liquid and two-phase cooled data centers. It records each plate's thermal health at commissioning, tracks how it changes in service, and is designed to give early warning before a two-phase plate reaches boiling crisis.
Early stage. The instrument is in design and its first bench test is planned for 2027. Our open-source evaluation software is released today.
The problem
AI chips are projected to pass 4,000 W each by 2029 (Dell'Oro Group, January 2026), and data centers are moving to liquid cold plates to keep up. Each plate is installed on trust. There is no routine way to check how well an installed plate is removing heat, so a plate that is partly blocked, badly mounted or short of flow often goes unnoticed until the chip throttles.
Two-phase cold plates cool by boiling, which removes far more heat than single-phase flow. Boiling also has a hard limit, called critical heat flux. Past it, vapor blankets the surface and temperature runs away. Today, operators learn about that margin from formulas fitted to old lab data, from lab tests run to failure, or from temperature alarms that fire after trouble starts.
Machine learning models are now being proposed to predict this behavior. Many are scored in ways that let near-identical data leak between training and testing, so they look more accurate than they are.
The product
Two parts that work together: an instrument at the cold plate and a software platform that turns its readings into decisions.
BCP Plate Instrument
Hardware (working name)
- A compact sensor module and measurement electronics that fit at the cold plate, running our own firmware.
- Portable version for commissioning and acceptance testing: records a health baseline for every plate.
- Embedded version, later, built into cold plates or coolant distribution units (CDUs) for continuous monitoring.
- Designed for single-phase and two-phase cold plates. The boiling crisis warning is specific to two-phase.
BCP Margin
Software platform (working name)
- A per-plate health record that starts at commissioning and follows the plate through its service life.
- Alerts when a plate drifts from its own baseline, so degraded plates are found before chips throttle.
- Critical heat flux risk flags for two-phase plates, with an honest confidence level on every warning.
- Model check: tests thermal AI models on held-out conditions, checks for data leakage, and reports when a model is outside what it has seen.
- Exports to CDU controllers, building management and DCIM systems. Built on our open-source CHF-Watch toolkit.
The instrument measures each plate in place and writes a baseline into the acceptance record.
The platform tracks every plate against its own baseline and flags change.
For two-phase plates, it gives an early warning trend as a plate approaches boiling crisis.
| Plate | Type | Margin | Trend | Status |
|---|---|---|---|---|
| Plate 01 | Two-phase | Normal | ||
| Plate 02 | Two-phase | Normal | ||
| Plate 03 | Single-phase | Normal | ||
| Plate 04 | Two-phase | Early warning | ||
| Plate 05 | Single-phase | Drift | ||
| Plate 06 | Two-phase | Normal |
Illustrative mockup. All plates and values are invented.
Cooling margin
Example value
Trend, last 30 days Baseline at commissioning
Margin trending down from this plate's own baseline. Confidence: medium.
Suggested check: coolant flow and plate mounting.
Who it is for
Cold plate and CDU makers
A per-plate quality and health signal built into their product.
Data center operators
AI cluster teams running direct liquid or two-phase cooling.
Commissioning agents and test labs
A repeatable acceptance check for liquid cooling.
Thermal AI teams
Builders of digital twins, cooling controls and prediction models who need honest testing.
Business model
Our plan for how BCP Technologies makes money.
BCP Technologies is a product company. We do not sell consulting, engineering or commissioning services.
Instrument sales
The portable instrument is sold per unit to commissioning firms, test labs and cold plate makers.
Software subscription
BCP Margin is licensed per monitored plate or per site, billed annually.
OEM licensing
The embedded sensor design and firmware are licensed to cold plate and CDU makers with a per-unit royalty.
Open core
The CHF-Watch evaluation toolkit stays open source. The hosted platform, alerts, integrations and reports are paid.
We have no customers or revenue yet, and pricing is not set. We plan to raise a pre-seed round to fund the first lab test campaign and the pilot instrument.
Roadmap
Released work, then planned milestones. Dates for planned work are targets, not results.
- Sep 2026Released
CHF-Watch v0.1.0, our open-source evaluation software, archived on Zenodo.
- Oct 2026Released
Four open preprints on leakage-safe evaluation and critical heat flux modeling.
- Q1 2027Planned
Platform prototype: per-plate health record, alerts and model check running on public and simulated data.
- H1 2027Planned
First bench test on a low-power boiling bench built in house, with pass and fail criteria written before the test is run. Not yet run.
- Late 2027Planned
Partner lab campaign: repeated cold plate tests at an established two-phase boiling facility, with an independent reference measurement.
- 2028Planned
Pilot instrument: portable prototype placed with early partners from the waitlist.
- 2028 and laterPlanned
Embedded version for flow boiling cold plates, with an OEM integration path.
Every milestone has written pass and fail criteria. If a test fails, we publish the result and narrow the product to what the data supports.
Open research
We publish our evaluation methods openly so others can check them.
CHF-Watch: Leave-One-Surface-Out Evaluation of Multimodal Boiling Heat-Flux Learning, Uncertainty Calibration, and Research Alarms
Leakage-Safe Evaluation for Multimodal Boiling Machine Learning: A Domain-Shift Baseline on BoilingBench
Campaign-disjoint evaluation of physics-structured critical heat flux models: gains, failure modes and data-basis sensitivity
Observation contracts for multimodal boiling: temporal support, spatial topology and limits of physical inference
CHF-Watch software, version 0.1.0
Team
Brad Moore
Founder and CEO
Brad has worked in HVAC controls, commissioning and data center monitoring since 2006. He has installed and programmed building automation systems, led analytics rollouts across more than 350 buildings, and engineered monitoring for Tier IV data centers. He designs the instrument's electronics and firmware himself. BS, Mechanical and Energy Engineering, University of North Texas. Certified Energy Manager.
Join the pilot waitlist
We are looking for cold plate makers, CDU makers, operators and test labs to shape the first pilot instrument. Tell us about your cooling setup and we will contact you when pilot units are ready.
Join the waitlistBCP Technologies LLC, Hutto, Texas. brad@bcptech.dev