Nvidia's next-generation 800 volt HVDC power system, a new way of delivering electricity inside AI server racks that cuts energy loss and lets racks scale to far higher power levels, may not reach large-scale production until 2028 or later, a year later than the 2027 timeline the market had expected, according to research firm SemiAnalysis. HVDC, short for high-voltage direct current, replaces the many small power converters inside a server rack with fewer, more efficient ones, which matters because AI racks are increasingly limited by how much power and cooling they can pack in, not by chip performance. But semiconductor analyst Jukan, citing supply chain sources, argues the market has misread both the cause and the timing of the delay. The issue, these sources say, is a validation problem with first generation product designs, not a change to Nvidia's technology roadmap. Power cabinets, battery backup units, and battery packs have all been resubmitted for validation testing, a normal part of qualifying new hardware rather than a sign of a deeper problem. Taiwan's Delta Electronics, a major supplier of data center power systems, is described as positioned for a second-quarter ramp once validation clears. The system is meant to support Nvidia's upcoming Vera Rubin chip platform, and any slip in power hardware readiness could delay how fast data centers can actually deploy Rubin at scale, even if the chips themselves ship on time.
Why it matters · AI racks are now bottlenecked by power delivery, not chip speed, so a delay in the 800V system could slow how fast Nvidia's next platform actually gets deployed even if the chips ship on schedule.
Worth asking · SemiAnalysis says the 800V power rollout slips a year, supply chain sources say it's just a validation delay with the roadmap intact. Which read should power-supply-chain investors trust?