Mitul Mehta

Founder @ Xtherma | Ultra-conductive solid-state heat spreaders for hotspot reduction + fast heat spreading in tight packages/modules. AI/HPC + power.

[Mitul Mehta @xtherma]

The most expensive AI chip in the world is useless if it throttles.

I believe the next AI race is not only about compute.

It is about how fast heat can escape from a hotspot only millimetres wide.

We are building Xtherma for that race.

[Mitul Mehta @xtherma]

Thermal teams don’t approve new solutions because of one impressive number.

They approve them when evaluation risk is low enough.

At Xtherma, we’re building around:
- clear boundary conditions
- real-stack integration
- comparison baselines
- qualification path

[Mitul Mehta @xtherma]

At Xtherma we don’t start with “what’s the conductivity?”

We start with:
– hotspot behavior
– spreading across the stack
– real interfaces under real conditions

Because that’s what actually limits systems.

Benchmark: >8°C lower hotspot vs baselines, ~½ weight, passive.

[Mitul Mehta @xtherma]

Follow-up: we turned the “not yet” into an Evaluation Kit.

Includes: protocol + boundary conditions, 3 reference stacks (AI/HPC, power modules, high-reliability), baseline comparison method, integration notes, and a qualification pathway.

[Mitul Mehta @xtherma]

Founder lesson this week: stop selling “samples.” Sell an evaluation packet.

We got a polite “not yet” from a Tier-1, not because the need isn’t real, but because big teams can’t absorb uncertainty.

What do you wish startups delivered before asking for a PoC?

[Mitul Mehta @xtherma]

Liquids solve temp but add qual + maintenance + new failure modes.

We’re building passive drop-in spreaders/modules to cut hotspots w/o pumps/fluids.

What’s your limiter?
A Hotspot
B Spreading
C Interface
D Air-side
E Serviceability

[Mitul Mehta @xtherma]

Liquid cooling reduces temperature.
It also adds pumps, maintenance & complexity.

Sometimes you don’t need liquid.
You need better spreading.

We build passive drop-in spreaders to flatten hotspots while keeping systems simple.

Would you replace a cold plate if passive worked?

[Mitul Mehta @xtherma]

“High conductivity” ≠ best cooling.

Real performance depends on:
– Spreading resistance
– Interfaces
– Pressure
– Thickness limits

We design for real stack behavior, not datasheet numbers.

What matters more in your system:
Hotspot delta or average temp?

[Mitul Mehta @xtherma]

Air cooling isn’t usually the problem.
Hotspots are.

When one region runs 10°C+ hotter, you get derating, bigger sinks & redesign.

We build passive drop-in spreaders to flatten peak temps without pumps or fluids.

Bench: >10°C hotspot reduction vs metal baselines.

[Mitul Mehta @xtherma]

Hotspot problems aren’t always “more conductivity.” Often it’s stress at the interface.

We bias compliance: softer at hotspot regions, stiffer elsewhere. Controls where heat/stress goes first and improves stability under cycling.

#thermalmanagement #reliabilityengineering

[Mitul Mehta @xtherma]

My quick filter for thermal claims: if the boundary conditions aren’t stated, it’s not a claim.
Same part can look great or useless depending on interface, heat injection, airflow, and what you’re optimizing (peak vs gradient).
Always ask: what changed vs baseline?
#electronics

[Mitul Mehta @xtherma]

When airflow + heatsink volume are capped, the lever that still moves the needle is often inside the footprint.

That’s what we’re building at Xtherma: compact, solid-state heat spreading to reduce hotspots + flatten gradients #thermalmanagement #HPC #powerelectronics #aihardware

[Mitul Mehta @xtherma]

A packaging engineer said: “We fixed airflow/heatsink. The hotspot still decides everything.”

Often the sink isn’t the gate, the spreading inside the footprint is. In dense stacks I look for 2 wins: drop the peak + flatten the gradient. #thermalmanagement #HPC #AIhardware

[Mitul Mehta @xtherma]

Something I keep seeing in evaluations: we improve airflow/heatsink…and the reliability questions don’t disappear. Often it’s not the average temp, it’s the gradient inside the package/module. Xtherma= Lower peak + flatten gradient. #thermalmanagement #HPC #aichip #gpu #Power

[Mitul Mehta @xtherma]

Hotspots don’t always respond to a bigger heatsink.

Often the bottleneck isn’t “getting heat out.” It’s spreading heat inside the footprint early enough.

With capped airflow + volume, peak temp + gradients decide everything.

Xtherma is built for that.