GCM Corporation, a thermal management company based in Western Australia, has taken its VHD heat sink range from 21 products to 44, according to Data Center Dynamics. The sinks use a graphite-based composition, and GCM claims superior thermal conductivity to traditional copper or aluminum alternatives. Chief executive and managing director Clinton Booth said: "As AI drives chip power densities higher, thermal management has become a critical design consideration, with hyperscale operators increasingly relying on proven, specialist OEM providers." GCM has shipped product to customers in China, Europe, Korea, and the United States for testing and prototype development, and has begun design and manufacturing work with an established global liquid cooling OEM that has delivered to hyperscale operators. It named none of them. The same range also serves gaming PCs.
Copper moves heat at roughly the same rate in every direction, around 400 W/m·K. Engineered graphite does not, and the spread between its two axes is the whole engineering story. Annealed pyrolytic graphite reaches roughly 1,500 W/m·K in plane against single digits through plane. Graphite composites of the kind more likely to be machinable into a 44-SKU range typically land between 300 and 700 W/m·K in plane, at or modestly above copper. GCM has not disclosed which material class the VHD range uses, or its numbers.
The base of a heat sink does two jobs. It takes flux through its thickness from the die, and it spreads that flux sideways to reach the fins. Spreading resistance usually dominates on a small die under a large base, and spreading is an in-plane problem, which is exactly where anisotropic graphite is strong. Through-thickness conduction is where it is weak. So the lay-up decides the part: orient it correctly and graphite beats copper on spreading at a fraction of the mass, orient it wrong and it loses to the aluminum extrusion it was meant to replace. Interface resistance, thermal expansion mismatch against the package, machinability, and oxidation above roughly 400C decide the rest. GCM published no thermal resistance figures, no interface material assumptions, and no test conditions with the range expansion, which are the first items an OEM thermal engineer asks for.
In a direct-to-chip rack, cold plates take the accelerators, increasingly the CPUs, and in the newest designs the memory. Power supplies, voltage regulators, busbars, optical transceivers, drives, and most switch silicon stay on air inside that same cabinet. Liquid cooling fixed the GPU problem and storage started overheating. Cisco put direct liquid cooling inside the switch, and most of the network has not followed. Those stranded components are working in halls where air volume has been reduced to bank the fan power savings that justified the liquid retrofit, which raises local inlet temperature and shrinks their thermal budget at the same moment total rack power climbs. That is a real and growing market for a better air-side sink, and it is not the market GCM's announcement talks about.
GCM describes the program as being at an early stage of development, with samples out for prototype testing, one unnamed OEM engagement, no disclosed manufacturing capacity, and a product line that also sells into gaming. Server thermal design wins get decided by OEM qualification cycles measured in quarters, against skived and additively manufactured copper that is not standing still, on cost per watt removed at the OEM's own conditions, with a qualified supply chain behind it. On that clock, an OEM engagement announced in July 2026 reaches a shipping rack in late 2027 at the earliest, and only if the material data holds up. Treat the range expansion as a catalog move rather than a design win until GCM names the OEM.