NVIDIA is reportedly preparing a major shift in cooling for its next-generation Rubin Ultra AI accelerators, exploring direct-to-chip microchannel cold plates to keep soaring power and thermals in check while maximizing performance per watt. As each new architecture raises the bar on compute, it also pushes thermal design power and heat density to levels that conventional liquid-cooling plates struggle to manage efficiently.
Industry chatter suggests the company has reached out to partners to develop microchannel cover plates tailored for Rubin Ultra, a move that would mark a meaningful evolution from typical cold plate designs used today. One report points to a rumored 2.3 kW TDP target for Rubin-class GPUs, underscoring why heat removal is becoming a first-order design constraint in AI servers and rack-scale systems.
What makes microchannel cover plates different is how they bring the coolant closer to the source of heat. Think of a precision copper plate etched with fine microchannels that route fluid directly over the hottest areas of the die. By increasing surface area and promoting local convection, these plates cut thermal resistance from silicon to coolant, unlocking higher sustained clocks, improved stability under heavy AI training loads, and better performance per watt. It’s conceptually similar to direct-die cooling popular with extreme CPU overclockers, but engineered for the reliability, serviceability, and scale that hyperscale data centers demand.
The push toward microchannel designs isn’t happening in isolation. The cadence from Blackwell to Rubin is aggressive, and power budgets are climbing. Even if developers would prefer to avoid major cooling overhauls, the physics of heat dissipation at these densities leaves little choice but to innovate. Recent industry talk also highlights potential concerns with microchannel plates—such as manufacturing complexity, pressure drop, and long-term reliability—yet the potential gains are compelling enough to keep development moving forward.
It’s claimed that NVIDIA has engaged Asia Vital Components, a Taiwanese thermal specialist, to help design the new plates. The approach was reportedly considered for the initial Rubin rollout, but tight timelines may have pushed full adoption to Rubin Ultra instead. In parallel, other players are experimenting with even more radical approaches. Microsoft recently showcased microfluidic cooling that goes in chip or on the silicon backside, signaling a broader industry shift toward advanced liquid cooling beyond standard cold plates and rear-door heat exchangers.
If Rubin Ultra adopts microchannel cover plates at scale, expect a ripple effect across the data center ecosystem. Direct-to-chip liquid loops, coolant distribution units, manifolds, and facility water infrastructure will all factor into deployment plans. With AI GPU and ASIC demand still climbing, analysts anticipate a meaningful uptick in liquid-cooling revenue around 2026 as next-gen accelerators enter volume production.
Bottom line: as AI compute races ahead, cooling has become a strategic battleground. Microchannel cover plates offer a credible path to tighter thermal control and higher efficiency for Rubin Ultra, even as the industry weighs the engineering trade-offs. Watch for more detail as hardware partners refine designs and the Rubin family nears launch.
