It's no longer just GPUs and CPUs getting the full immersion treatment: now even power busbars are going under liquid. At the Wiwynn booth, the latest generation of TE Connectivity 800 V DC components stood out precisely for this reason — liquid cooling is no longer an optional extra for chips, but is becoming an integral part of the electrical infrastructure feeding next-gen AI accelerators.

The Component in the Spotlight

The demonstration's centerpiece is a busbar capable of handling 800 V DC loads, designed to deliver increasing power to servers running AI workloads. The novelty lies in the liquid cooling system integrated directly into the busbar's body — a solution aimed at controlling the temperature of a component that is normally passive but, with hundreds of amps flowing through it, becomes a significant heat source itself.

TE Connectivity and Wiwynn are collaborating to bring this approach into ever-denser racks, where every watt counts and the power delivery chain must be as efficient as the processors it serves. No specific benchmarks were released, but the shift to 800 V distribution is already a telling indicator: higher voltages reduce currents and thus resistive losses, but require more sophisticated insulation and protection mechanisms.

Why Liquid Cooling Is Reaching Busbars

The push comes directly from AI accelerator roadmaps. Current compute models, both for training and large-scale inference, are driving single-card power draws well past 700 W, with next-generation projections exceeding a kilowatt. In a rack with tens of nodes, total power can reach tens of kW, and distributing that at low voltage (48 V or 12 V) would mean unmanageably high currents without massive, costly copper bars. That's why the industry is migrating to 400 V, 800 V, and beyond, borrowing solutions in part from automotive and heavy industry.

The snag is that even an 800 V busbar, when carrying thousands of amps, heats up. In a data center already saturated with heat, every degree matters. Liquid cooling the busbar is thus not a technological gimmick but a necessity to maintain efficiency, avoid hot spots, and ultimately enable power densities that air cooling simply cannot deliver.

Implications for On-Premise Deployments

For organizations evaluating large-scale on-premise AI infrastructure, signals like this carry real weight. Thermal management is one of the factors that most affects Total Cost of Ownership (TCO) and the very feasibility of a self-hosted training or inference cluster. Components such as liquid-cooled busbars promise to reduce footprint, simplify maintenance, and contain operational energy costs.

At the same time, they introduce engineering complexity that demands specialized skills and an ecosystem of suppliers capable of integrating power, cooling, and control into a single system. It is no coincidence that such solutions emerge from collaborations between connectivity specialists (TE Connectivity) and rack integrators (Wiwynn): it’s proof that next-gen AI is not just a silicon race, but depends on everything around it.

Beyond the Bar: Toward Fully Liquid-Cooled Racks

The adoption of liquid-cooled DC busbars is the latest piece of a broader trend: the shift from selective to total liquid cooling. Not long ago the discussion was about direct cooling only for CPUs and GPUs; now the frontier includes VRMs, memory modules, and even power distribution. The aim is temperature-controlled environments with minimal efficiency losses, which are also fundamental for data sovereignty: those managing sensitive workloads in-house need physical infrastructure that is reliable and predictable.

As we wait for these components to reach volume availability, the Wiwynn and TE Connectivity demonstration marks an important point: electrical infrastructure is no longer a forgotten commodity, but an innovation ground directly proportional to AI's hunger for compute.