Cooling solutions provider AVC starts shipping double-sided cold plates for AI servers

VPD is not a power delivery tweak—it is a thermal architecture reset. When current flows vertically through the substrate, heat generation becomes dual-sided, forcing cold plates to evolve from top-cover designs into sandwich structures. This cascades upstream: CoWoS and Foveros packaging must redefine TIM thickness tolerances and thermal conductivity windows. Downstream, every OEM chassis airflow path and rack-level liquid loop gets redrawn from scratch. The compliance picture shifts materially. As liquid cooling penetration races toward near-universal adoption by 2027, data center infrastructure logic flips from air-plus-redundancy to liquid-plus-sealing. Fluorinated coolant procurement and disposal add regulatory overhead. More critically, dual-sided cold plate micro-channel machining capacity is concentrated among few players; a geopolitical disruption could stretch lead times from eight weeks past twenty. On competition: Asetek, CoolIT, and Boyd still anchor their SKUs on single-sided plates. AVC's first-mover lock on VPD specs is effectively a position grab in NVIDIA's B300/GB300 thermal BOM. Once OEMs complete thermal simulation sign-off, switching vendors carries NRE costs that dwarf the cold plate itself. Twelve to twenty-four months out, liquid cooling stops being a differentiator and becomes a gate. The real battleground migrates to pump power overhead (target under 1.5% of TDP) and phase-change cooling demands as HBM4 stacking pushes die heat density past 1000 W/in². The 2027 full-adoption timeline is likely conservative—HBM4 layer proliferation could force mandatory liquid cooling nodes as early as Q4 2026.