At Computex 2026, we saw an important technology for the future of data centers. The Wiwynn booth always showcases great technology ahead of its time, making it one of my favorites to visit. This year, we are seeing a transition to a new generation of AI accelerators and, therefore, a new level of power density. As a result, there is a move to 800V DC power in the data center. Not only is the industry putting more power through an 800V busbar, but that busbar is being liquid-cooled. In the booth, we got to see the TE Connectivity solution for this new era. That sent me down a bit of a rabbit hole learning about 800V DC.
Starting from the Top with Big AI Accelerators and Delivering Power in a Server
Wiwynn and TE Connectivity were showing an example server with mock-ups of next-generation accelerators. In a fun twist, it was only partly to show off the server design. Instead, a major motivation for showing this is the power cable running down the middle.
Just for some frame of reference, let us start at the front with what looks like a normal ORv3 server node.
We see the management module on the left side and four E1.S SSDs.
In the center, there are dummy optical network cages, so we know this is just a mock-up of a next-gen server (but usually Wiwynn’s mock-ups are very informative)
Moving to the rear, we start to see differences.
There is a new liquid-cooling nozzle and connectors for scale-up networking within the rack.
In the center is the busbar connector which is what is starting the power chain that this entire demo is showing.
Just for some frame of reference, this is showing the \TE Connectivity HVDC IT GEAR TO ELCON MICRO+ Power Cable Assembly that has an 800V/75A HVDC IT gear end and a 400V ELCON MICRO+ end, all in an assembly.
Here are the two 100+cm2 accelerator dummy modules with the power assembly in the middle. A big question will be how to deliver power to the chassis in a manufacturable way. The industry is moving to more robotic assembly, so building the power cable assembly is not really practical on the line. Instead, the entire thing is meant to drop in during manufacturing.
Here is a close-up of how it connects to the internal server power distribution.
Here is another look including the channel that this slots into between the accelerators.
Of course, this will all be liquid-cooled, so expect large internal liquid cooling blocks to be on everything, but this is a really neat exercise. Again, for those who want to see next-gen servers, Wiwynn often has mock-ups that are very directionally informative as to where things are going.
