Prototype in test · bench validation underway

Every watt of heat
is on a journey.
We choose when
it travels.

Bordem AI is a phase-change thermal buffer for AI infrastructure — a stationary store that holds heat when moving it is expensive and releases it when it's cheap, carrying a data center's cooling load away from its costliest hours.

Safar — the journey of heat, charted and controlled
SCROLL
The problem

AI is outgrowing the way we move heat.

Compute demand keeps compounding, rack densities have jumped an order of magnitude, and cooling still runs as a rigid, always-on load — indifferent to weather, grid, or price. That inflexibility is becoming the real constraint on how fast infrastructure can scale.

0%
of a data center's energy goes to cooling — the biggest line item after compute.
projected growth in global data-center power demand by 2030.
0
per rack in modern AI deployments — up from the 4–10 kW air was built for.

Figures are widely-cited industry estimates for context, not Bordem AI measurements — they vary by facility and region.

How it works · the journey of heat

A stationary waystation between the racks and the sky.

Heat leaves the servers, rests in the buffer when the outside world is hot and expensive, and travels on when conditions are kind — timed by a controller reading the road ahead.

GPU racksheat begins Coolant loopseparated PCM waystation18–21°C rest stop Rejection3-layer stack Regeneratereset off-peak
i.

Separated coolant loop

Server heat rides a clean, isolated loop and is handed to the buffer through a heat exchanger — nothing new inside the rack, no new failure mode.

ii.

The PCM waystation

A stationary tank of capric–lauric eutectic melts across an 18–21°C band, holding heat at a near-constant temperature until it's cheap to move on.

iii.

Three-layer rejection

A dry cooler takes the easy hours, the buffer absorbs the peaks, and an adiabatic trim stage covers the worst-case envelope.

iv.

The map-reader

Model-predictive control forecasts load, weather, and price and decides when heat travels — the defensible core of the system.

PCM 18–21°C a cutaway of the waystation
The physics · interactive

Why a phase change makes such a good buffer.

Drag energy into the material and watch its temperature. While it melts, it swallows a huge amount of heat while barely warming — that long flat stretch is the whole trick. Bordem AI parks a data center's heat right on that plateau.

Material temperature
5.0°C
Solid — warming up
Melt fraction

Illustrates the physics of a phase-change material — not a product measurement. Bordem AI uses a capric–lauric eutectic tuned to melt at 18–21°C.

Performance · modeled

What the physics says — before the hardware says it.

A v0 thermal model of a 1 MW facility in a hot climate. Sending heat rejection toward the cool, efficient hours flattens the chiller's electrical draw. The headline is peak reduction — the demand-charge lever. These are modeled figures; bench validation is underway.

Chiller electrical draw · a day, chartedv0 model
Baseline chillerWith Bordem AI buffer
~19%
Electrical peak shave

The worst-hour draw that sets demand charges and chiller sizing.

~1.6 MWh
Thermal buffer / MW IT

Sized to hold roughly a full afternoon of shifted heat.

Off-peak
Rejection shifted to cool hours

Cooling work moves to cheaper, cooler hours — time-of-use upside on top.

Modeled, not measured

These come from a first-pass thermal model for engineering illustration — not lab measurements. A bench prototype is in test; measured charge/discharge curves will replace these projections as data arrives.  Download the modeled dataset (CSV) →

Applications

Wherever a cyclic heat load meets an expensive peak.

The same buffering physics travels across sectors. We prove it where pilots move fastest, then carry the data up the road to AI data centers.

i.

AI data centers

The destination — where density and cooling bills make the buffer most valuable at scale.

ii.

BESS

Battery sites need tight thermal control in a small envelope — the quickest path to a pilot.

iii.

EV fast-charging

High-power stacks spike hard and briefly — a textbook case for buffering the peak.

iv.

Telecom & edge

Distributed, cooling-constrained sites where avoided peak has outsized value.

v.

Semiconductor fab

Process-critical thermal stability where excursions are costly.

vi.

Cloud providers

Flatten cooling demand across fleets and shift to cheaper hours.

vii.

Research & HPC

University and lab clusters with bursty, schedule-driven compute.

viii.

Enterprise

On-prem rooms retrofitting toward denser, hotter hardware.

Roadmap

From bench to first field pilot.

Three waypoints, kept honestly apart: reached, under way, and ahead. Nothing here is drawn as a place we've arrived before we have.

Reached done

Architecture & thermal model

Stationary bulk-PCM, separated-loop architecture defined; v0 thermal model built and documented.

Under way now

Bench prototype & measurement

Building and instrumenting a bench buffer to capture real charge/discharge and thermal-curve data against the model.

Ahead planned

Pilot-scale unit

A field-deployable buffer for a first design-partner site in a fast-moving market.

Ahead vision

Data-center deployment

Scale the proven buffer into AI data-center cooling, carrying operating data from earlier pilots.

We're charting the thermal layer the AI buildout forgot to plan for — the quiet infrastructure the world will run on, and never have to think about.

Founders

Builders who start from first principles.

Each founder keeps a personal site in the same warm, editorial spirit. Open one below — the full site loads inline, right here, without leaving Bordem AI.

Co-founder & CEO

Vedant Malhotra

Owns the technical thesis and company direction. Previously took Bordem AI from zero to paying customers.

Co-founder & COO

Dharmyash Jain

Runs execution — the prototype and bench-test program, mechanical build, and supply chain.

Co-founder & CSO

Aayan Gupta

Owns strategy — where the company plays, who it serves first, and how it gets there.

Founders' network ·
The founder's full personal site, loaded inline as guest content — a deliberately distinct visual identity.
Contact

Building the cooling layer for AI infrastructure.

We're raising and talking to design partners to take the buffer from bench prototype to first field pilot. If you back deep tech, climate, or infrastructure — or run a facility fighting its cooling peak — let's talk.

Founders
Vedant · Dharmyash · Aayan
Location
India
Status
Prototype in test · raising