Odaily Planet Daily reports that SemiAnalysis, a research institute specializing in semiconductors and AI infrastructure, has released a new report stating that two key technological pathways for AI data centers have experienced significant delays, including the mass production timelines for NVIDIA’s native 800V DC architecture and CPO (co-packaged optics) technology, both of which are behind market expectations.
The report indicates that NVIDIA’s planned large-scale adoption of 800V DC power architecture has been delayed until 2028, rather than the previously market-expected timeline. Meanwhile, CPO shipment volumes in 2027 are expected to fall below market projections, with mass production likely delayed until 2028–2029, primarily due to challenges in optical engine yield, ASIC integration complexity, and cost factors.
SemiAnalysis believes that the 400V DC solution will still ramp up as planned starting in 2026, with some NPO projects potentially accelerating. However, the power supply, switch, and optical module supply chains dependent on new platforms such as Rubin Ultra and Kyber will face delays. The report maintains a relatively positive outlook on Amphenol, Vertiv, and Legrand, while adopting a more cautious stance toward companies such as Lumentum, Himax, Navitas, and Wolfspeed.