Skip to main content
True Precision Machining

Industries · Super Computing

When the chip dissipates 1 kW, the cold plate has to be perfect.

High-performance computing and AI training pushed thermal density past what air-cooling can handle. Direct-to-chip liquid cooling demands cold plates flat to 0.0005″ over a foot. We machine cold plates, heat exchangers, optical packaging, and chassis hardware for HPC, hyperscaler, and exascale-class customers.

Flatness class
0.0005″
Materials
OFHC · 6061 · MIC-6
Surface Ra
8–16 µin
Volume range
10 → 5K

01 · Why HPC needs precision machining

Heat is the whole problem.

A modern AI training accelerator dissipates 700 W to 1.2 kW per chip. A rack of them dissipates 80–120 kW. Air-cooling capped out years ago — direct-to-chip liquid cooling and immersion are the only paths forward. The cold plate sitting on the chip is the single most precision-critical interface in the rack.

That cold plate has to hold 0.0005″ flatness over its full footprint, achieve 8–16 µin Ra surface finish for thermal-interface-material wetting, and survive vibration and pressure cycling. We machine those plates in OFHC copper and 6061-T6, validate flatness on the Hexagon Global S CMM, and provide CMM scan data with each lot.

02 · Components we make

From the chip up.

Liquid-cooling cold plates

Precision-flat OFHC copper or 6061-T6 with internal flow channels. Direct-to-chip cooling for 1+ kW density rack hardware. Flatness 0.0005″ over 12″.

Custom heat exchangers

Micro-finned aluminum, copper-skived, brazed plate-fin assemblies. Thermal-test fixtured. CFD-validated geometries machined to spec.

Optical-transceiver housings

Silicon photonics packaging — fiber-array housings, lens-mount alignment fixtures, hermetic sub-mounts. 0.0001″ alignment-critical features.

Server chassis & sleds

1U/2U/4U sleds, GPU-mount backplanes, NVMe carriers, cable management. 5052/6061 with anodize.

Test & metrology fixtures

Wafer probes, package-test sockets, thermal-shock fixtures, bond-pull testers. 0.0001″ positional, validated to gauge R&R.

Vibration-isolation hardware

Damped mounts, pneumatic-isolation interfaces, precision spring carriers. Spec to picometer-class metrology environments.

ORNL Frontier exascale supercomputer

03 · Customer profile

Hyperscalers, national labs, AI infrastructure.

Our HPC work spans hyperscaler data-center hardware, DOE national-lab exascale systems, AI training infrastructure, and silicon-photonics packaging. The thermal-and-precision discipline we built for aerospace transfers directly — same Hexagon Global S CMM, same SensFlo process monitoring, same DFM rigor.

Imagery: ORNL Frontier exascale supercomputer (Carlos Jones / ORNL / DOE, public domain).

04 · Off our floor

Aluminum fin-array work — actual production parts.

Precision-machined fin geometry for cooling and heat-exchanger applications. Tight pitch, consistent fin thickness, uniform surface finish across the array — what direct-to-chip thermal interfaces actually require in production.

Customer-specific dimensions, materials, and end-application held under NDA.

Precision-machined aluminum fin array — TPM production work

Building HPC hardware that needs precision metalwork?

Cold plates, heat exchangers, photonics housings — send drawings or a STEP file with thermal/optical requirements. Production-volume capable.

Quotes in 24 hours