The key link in the power supply of AI data centers is 800VHDC. Today, we are here at https://(x.com)/qinbank/status/2057685712645116279 to discuss the dismantling of the VR200NVL72 by Da Mo. It is mentioned that Da Mo expects 800V DC power to be widely adopted on the Rubin Ultra platform, which will be launched in the second half of 2027. In fact, 800VHDC is not a completely new concept. Since Nvidia announced last year that it would launch a DC power supply architecture, the market has paid considerable attention to it. The related targets in this sector have already exceeded expectations. But in fact, 800V DC power will only be widely adopted in the second half of the year, which is worth paying attention to. Let's talk about several issues: 1. What is the 800V HVDC architecture? The path of traditional high-density AI racks is roughly as follows: medium voltage AC in the power grid → transformer/America/PDU → 415/480VAC to the rack → PSU in the rack converted to 54VDC/12VDC → GPU core voltage. Dazi's 800VDC vision is to centralize the conversion of medium voltage AC to 800VDC at the data center boundary/power room, send it to the IT rack via 800V DC busway, and then use high ratio DC/DC conversion near the GPU. NVIDIA claims that the 54V architecture is starting to hit physical limitations above 200kW; If a 1MW rack continues to use 54V, a single rack copper bar may require up to about 200kg of copper. The 800V architecture aims to improve efficiency, reduce copper consumption, and free up rack space by reducing current, decreasing conversion levels, and reducing PSU inside the rack. It is not a simple voltage upgrade, nor is it "inventing DC power supply", but a platform level switch of AI data center power supply architecture. It is a systematic reconstruction of the entire power delivery architecture, aimed at solving the bottleneck of traditional 48V/54V rack power supply (limited space, copper cable overload, high multi-level conversion loss), supporting single rack power to jump from hundreds of kW to 1MW+, and paving the way for future GW level AI factories. 2. What is the significance and revolution of 800V HVDC? 1) Firstly, efficiency and spatial layout are naturally important Efficiency improvement: The conversion process from the power grid to GPU has been significantly reduced, and the overall energy efficiency can be improved from 90% to over 98.5%. The transmission loss has been significantly reduced, and the TCO (Total Cost of Ownership) has been reduced by up to 30% Space and density optimization: Reduce copper cable usage and power unit volume, increase computing space utilization in the rack by over 80%, and support higher density GPU clusters 2) 800V is not a single device upgrade, but an ecological reconstruction: central rectification, 800V DC busway, solid-state circuit breaker, hot swappable protection sidecar/power rack、BBU/CBU、 Supercapacitors/battery energy storage DC/DC、GaN/SiC、 Liquid cooling requires coordination. NVIDIA has also explicitly stated the need for organizations such as OCP to promote voltage ranges, connectors, and safety standards. If you have paid attention to the new energy vehicle industry chain, it should have an impact. In the past two years, the "fast charging" that domestic electric vehicle manufacturers have been promoting is basically 800V high-voltage DC charging. Now Dazi is making 800VDC a part of the next generation AI rack standardization route, so some key links in the charging industry chain for new energy vehicles are beginning to spill over to AI data centers. 3. How large is the space for 800V HVDC? To see the overall background, the AI data center market will grow from approximately $344 billion in 2025 to over $2 trillion in 2032 (CAGR 27.5%). Power infrastructure will become the core bottleneck and growth point of AI construction, and NVIDIA's standards will accelerate the adoption of hyperscalers, driving the explosion of sub markets such as solid-state transformers and GaN/SiC power devices. After 2027, the penetration rate of 800VDC or similar HVDC architectures will rapidly increase in AI zones with>300kW/rack, especially 400kW-1MW rack. If 30% -60% of the future new AI capacity adopts high-voltage DC architecture, and the core 800V power chain value corresponding to each MW is in the range of tens of thousands to millions of dollars, the cumulative space will enter the tens of billions to billions of dollars level. Of course, this prediction range is also very wide, as the reality depends on the shipping pace of Kyber/Rubin Ultra and the degree to which ultra large cloud factories accept NVIDIA 800V. 4. Composition of the 800V HVDC industry chain It is entirely NVIDIA's reference design led qualification certification, and the core partners previously listed by NVIDIA are divided into three categories with fierce competition. The share will depend on the certification progress, production capacity, and hyperscale contract. 1) Silicon wafer/power semiconductor suppliers (core components such as SiC/GaN MOSFETs, controllers for efficient conversion): Main players: Texas Instruments (TI, has released a complete 800V solution), STMicroelectronics (ST, 6-18kW power board), Infineon, ROHM (SiC devices) IFFNY、Navitas（GaN/SiC）、Analog Devices、onsemi、Renesas、Innoscience、MPS、AOS、EPC Wait. These are the core of NVIDIA's "Silicon Supplier" list, with TI/ST and other demonstrated reference devices 2) Suppliers of power system components/modules (power racks, sidecars, DC-DC converters, etc.): Main players: Delta Electronics (in deep collaboration with NVIDIA, releasing 800V solution), Flex, LITEON, Megmeet, Lead Wealth, Bizlink, etc. Chinese manufacturers such as Delta have obvious advantages and have already published white papers and implemented technologies; The stock prices of LITEON and others have significantly increased due to the expectation of 800V. 3) Data center power system/infrastructure suppliers (rack level power distribution, Sidecars, SST, busbars, etc.): Main players: Vertiv (Hopewind is a key sub supplier of its 800V system), Schneider Electric (developing 1.2) MW Sidecar）、Eaton、ABB、GE Vernova、Siemens、Hitachi Energy、Mitsubishi Electric Wait. Vertiv, Schneider, Eaton, and others are traditional powerhouses here. From a personal perspective 1) Vertiv, Schneider, Eaton, Delta, and ABB are the five companies most likely to have a significant share of early 800VDC related revenue; 2) LITEON, TI, ST, Infineon, and onsemi are the second group of beneficiaries with strong certainty; 3）Navitas、Power Integrations、MPS、BizLink、Megmeet、Innoscience（ InnoTech belongs to a group with greater flexibility but also higher validation/mass production/competition risks. From a personal perspective, what I am currently more optimistic about is, of course, this still needs to be dynamically iterated: nvts、IFNNY、 InnoTech vicr 5. Several important milestones for tracking the landing rhythm in the future 1) NVIDIA Kyber/Rubber Ultra 2027 rhythm: Is it clear that 800VDC is used as the default/main rack power architecture, and the shipping rhythm also drives the landing rhythm of 800V 2) OCP standard progress: Standardization of 800V connectors, safety, protection, PDB, BBU/CBU. 3) Focus on power management system components, disclosure of power semiconductor suppliers' orders, who has truly entered the backlog and production sockets; This is the most crucial factor that determines which supplier can get a large share of the market 4) Ultra large cloud factory route: Is the HPR split between 800V vs 400V/± 400V vs 50V. It determines the market's expectations and imagination space for 800V HVDC. 5) Single MW cost reduction curve: If 800VDC significantly improves the number of deployable GPUs, energy efficiency, and maintenance costs per MW, it will transition from NVIDIA's proprietary architecture to an industry standard.