According to a research note released by Morgan Stanley on October 12, the market had previously been overly optimistic about the deployment timeline for 800-volt direct current (800VDC) power supply technology in artificial intelligence (AI) data centers, but recent concerns about delays in rolling out the technology have also been exaggerated. The bank believes that although the timeline for NVIDIA (NVDA.US) to fully adopt an 800V power architecture across its next-generation AI server racks may be pushed back to 2028, the overall direction of the technology transition remains unchanged. The Open Compute Project (OCP) Global Summit, held from October 12 to 15, is expected to further clarify the development roadmap for 800V technology and could serve as a positive catalyst for power management and analog semiconductor stocks.
Morgan Stanley analyst Joseph Moore noted in the report that the commercialization progress of 800V power technology has been a focal point for the semiconductor industry throughout this year. As the power consumption of AI servers continues to climb, traditional power architectures are facing growing challenges, and adopting higher-voltage DC power systems is seen as an important direction for improving data center power efficiency and meeting future high-power computing demands.
However, the actual pace of 800V technology deployment may be slower than the market previously expected. According to Morgan Stanley's analysis, NVIDIA originally planned to introduce native 800V power supply through its Kyber rack architecture in the Rubin Ultra series, scheduled for launch in the second half of 2027. However, Kyber is no longer the rack form factor planned for Rubin Ultra, and the related plans are being adjusted. NVIDIA's latest published technology roadmap shows that its third-generation server rack, with approximately 570 kilowatts of power, still uses 54V DC power distribution. Morgan Stanley believes this configuration is largely consistent with earlier expectations of a Rubin Ultra rack at roughly 600 kilowatts equipped with 144 GPUs. This means that NVIDIA's first server rack to clearly adopt a native 800V power architecture may not appear until the subsequent Feynman architecture generation, with power reaching approximately 1 megawatt, expected around 2028.
Morgan Stanley stressed, however, that the delay in deploying native 800V racks does not mean the entire 800V power technology transition has stalled. The bank noted that 800V external power supply modules (Sidecars) have already begun shipping on schedule, and NVIDIA's Power Center centralized power supply system, expected to launch in 2027, is still moving forward. Meanwhile, related technologies such as power conversion, power management, and voltage regulator modules (VRMs) are still expected to continue developing as the power density of AI computing chips increases.
For power semiconductors such as gallium nitride (GaN), the delay in deploying native 800V racks may postpone the release of demand for some high-frequency DC converters. But overall, the rollout of the 800V power architecture simply reflects a change in the pace of development across different segments, rather than a reversal in the technology's direction.
It is worth noting that Morgan Stanley believes market pessimism toward 800V-related semiconductor stocks may have already over-reflected the impact of the technology deployment delay. After NVIDIA's GTC conference in March this year unveiled a more detailed 800V technology roadmap, related semiconductor stocks were once sought after by investors. Data shows that the forward 12-month enterprise value-to-sales (EV/Sales) valuation multiple for 800V-related companies rose about 50% after the conference and peaked in mid-June. However, as news emerged that 800V technology deployment could be delayed, related stock valuations fell significantly. Currently, the sector's forward 12-month EV/Sales multiple has dropped to about 6.5 times, essentially returning to levels before the March GTC conference. At the same time, the valuation premium of the Philadelphia Semiconductor Index relative to 800V-related stocks has widened from about 12% to 25%, above the average of about 17% since April 2025.
Morgan Stanley pointed out that although related stock valuations have fallen noticeably, corporate fundamentals expectations have not deteriorated in tandem. Data shows that since the March GTC conference this year, the expected forward 12-month revenue growth rate for 800V-related companies has risen from about 12% to 27%. And since the first reports of technology deployment delays emerged, market revenue forecasts for the sector over the next 12 months have actually been revised up by a cumulative 11%, with no company seeing its revenue forecast cut. The bank believes this indicates that adjustments to the 800V technology deployment timeline have not yet had a significant impact on the short-term fundamentals of related companies. The overall recovery of the analog semiconductor industry and continued strong demand from AI data centers are still supporting corporate growth prospects.
The upcoming 2026 OCP Global Summit will be an important opportunity for the market to reassess the outlook for 800V technology. OCP is an open compute project jointly participated in by major cloud computing companies, equipment manufacturers, and component suppliers, dedicated to advancing next-generation data center infrastructure technology. Morgan Stanley noted that with the rapid growth in power supply demand for AI servers, the OCP Global Summit has significantly increased its focus on power and data center facility topics in recent years. According to the bank's tally, the total number of technical sessions at the 2026 OCP Summit reached 364, up from 267 in 2024, an increase of more than 30%. Among them, the number of sessions related to power and data center facilities nearly tripled compared with 2024, and their share of the total technical agenda has risen to about 16%. This year's summit will focus on topics including 800V DC power supply, high-voltage DC power distribution, rack power architectures, energy storage, and system safety. Technology companies including NVIDIA, Google (GOOGL.US), Microsoft (MSFT.US), and Meta (META.US) will all participate in related technical discussions. Among them, NVIDIA plans to present progress on the 800V DC power architecture's transition from technical standard coordination to actual deployment, while Google, Microsoft, and Meta will also participate in topics related to server power supply demand and technical standards.
Morgan Stanley believes the market will focus on whether 800V external power supply modules continue to progress as planned, the deployment arrangements for the 2027 Power Center system, and the specific launch timing of native 800V server racks. In addition, NVIDIA previously stated that more than 80 equipment manufacturers and infrastructure companies have joined the 800V power technology ecosystem. Morgan Stanley believes the growing number of industry participants also indicates that this technology transition is still advancing.
In terms of investment opportunities, Morgan Stanley noted that if the OCP Summit further confirms that the long-term direction of 800V power technology remains unchanged, related semiconductor stocks could see a positive market reaction. The bank specifically mentioned companies including Analog Devices (ADI.US), Texas Instruments (TXN.US), ON Semiconductor (ON.US), Wolfspeed (WOLF.US), and Navitas Semiconductor (NVTS.US), all of which are expected to benefit from growing demand for AI data center power management and power semiconductors.
However, Morgan Stanley believes the bigger impact of the 800V technology transition may be reflected in the utilization rates of mature-node semiconductor manufacturing capacity. The bank noted that, apart from the Chinese market and some of Texas Instruments' capacity, utilization rates at mature-node wafer fabs globally are generally at high levels. As demand for analog chips in AI data centers remains strong, if 800V power technology further expands demand for related chips, mature-node capacity supply could become tight in 2027. Morgan Stanley said it has already begun to anticipate possible supply shortages for some mature-node semiconductor products next year and therefore maintains a positive overall view on the analog semiconductor sector.