目录
01 从"标准"走向"推进计划" 02 Vertiv —— 2026年下半年推出产品，5个月完成工程验证 03 富士康 —— 台湾高雄K-1数据中心成为率先落地基地 04 施耐德电气 —— 支持包括谷歌、Meta在内的超大规模云服务商 05 供应链的扩展与落地效益 06 参考事实卡 从"标准"走向"推进计划" NVIDIA主导的800VDC电源架构，在Computex 2025发布仅一年多后，就从标准发布阶段进入了具体的推进计划阶段。当机柜功率超过200kW，传统的400V三相交流或48/54VDC配电方式会在铜排电流容量上遇到物理极限。NVIDIA已明确表示，将从2027年起主导整个行业向800VDC过渡，目标是支撑1MW以上的IT机柜。
问题在于"何时、由谁"真正落地。将电源厂商与EMS（电子制造服务）巨头的一手信息对照后可以看到，推进计划已经有了具体的日期和地点。
Vertiv —— 2026年下半年推出产品，5个月完成工程验证 Vertiv于2025年5月宣布与NVIDIA的800VDC路线图保持一致，计划在2026年下半年推出800VDC电源产品组合。该产品组合由集中整流器、DC母线槽、机柜级DC-DC转换器、DC后备系统四大类构成。
值得关注的是速度。Vertiv在2025年5月宣布合作后，仅5个月后的同年10月，就宣布与NVIDIA合作在800VDC架构上达成工程就绪（engineering readiness）。这一产品线的推出时间，是为了配合2027年展开的NVIDIA Rubin Ultra平台（Kyber机柜）的落地节奏。
2025年5月
Vertiv宣布与NVIDIA 800VDC路线图战略对齐
2025年10月
Vertiv达成工程就绪／施耐德电气宣布800VDC承诺／富士康与NVIDIA在OCP Global Summit宣布落地实施
2026年上半年
富士康旗下台湾高雄K-1 AI数据中心的800VDC率先落地推进
2026年6月
施耐德电气与富士康宣布追加战略合作
2026年下半年
Vertiv推出800VDC电源产品组合
2027年
NVIDIA Rubin Ultra版Kyber机柜（每机柜576颗GPU）开始出货
富士康 —— 台湾高雄K-1数据中心成为率先落地基地 EMS全球最大厂商富士康（鸿海精密工业）于2025年10月14日在OCP Global Summit 2025上宣布，将与NVIDIA合作把800VDC架构落地到AI工厂。此次合作指定的首个实施地点，是富士康旗下台湾高雄的"K-1"AI数据中心项目，该基地被定位为展示集团AI服务器、数据中心、可再生能源整合能力的示范据点。
富士康子公司Ingrasys Technology在OCP Global Summit 2025上，展示了将NVIDIA GB300 NVL72平台与列间CDU（冷却液分配单元）整合的方案。GB300 NVL72每机柜搭载72颗GPU、功耗142kW，这本身也是800VDC过渡前"最后一代过渡机型"功率密度的实例。
此外，2026年6月15日，施耐德电气与富士康宣布追加面向新一代AI数据中心的战略合作。将富士康的计算平台、AI机柜整合能力、全球制造能力，与施耐德的电力、冷却、能源管理技术相结合，双方目标是打造跨区域强调"速度、效率、可预测性"的一体化、即装即用方案。富士康董事长刘扬伟表示，"AI基础设施的设计、构建与交付模式的革新，是行业当下所需"。
施耐德电气 —— 支持包括谷歌、Meta在内的超大规模云服务商 施耐德电气于2025年10月13日宣布，将加强对NVIDIA新一代GPU的800VDC电源基础设施支持。其正在开发的800VDC侧车（sidecar）设计上支持包括谷歌、Meta在内的主要超大规模云服务商，并搭载"Live Swap"功能，可在不断电的情况下完成部件更换。该公司数据中心业务CTO Jim Simonelli将800VDC的过渡定位为"随着计算密度提升而来的自然演进"。
供应链的扩展与落地效益 NVIDIA的800VDC生态圈已有Analog Devices、英飞凌、Innoscience、MPS、纳微半导体（Navitas ）、安森美（OnSemi）、瑞萨（Renesas ）、罗姆（ROHM ）、意法半导体（STMicroelectronics ）、德州仪器（Texas Instruments）等主要半导体厂商参与。面向新一代Vera Rubin NVL72机柜服务器，已有超过50家MGX（NVIDIA模块化GPU参考设计）合作伙伴推进适配。
NVIDIA公布的落地效益数据显示：端到端效率最高提升5%，维护成本最高降低70%，总拥有成本（TCO）最高降低30%。主要原因是电力转换级数的减少和导体用量的降低。
从采购、电源设计、投资决策的角度看，800VDC已不再是"未来构想"。Vertiv与施耐德电气两大主力电源厂商都将在2026年内把产品推向市场，富士康也已在实际基地推进率先落地。面向2027年Kyber机柜的出货，设施侧的配电设计与设备采购提前期，已经是现在就需要纳入考量的变量。
参考事实卡
Vertiv plans to launch its 800 VDC power portfolio in H2 2026, aligned with NVIDIA's 800 VDC roadmap
On May 19, 2025, Vertiv (NYSE: VRT) announced its strategic alignment with the 800 VDC architecture roadmap NVIDIA published for next-generation AI data centers. The company's 800 VDC power portfolio is scheduled for release in the second half of 2026, ahead of the rollout of the NVIDIA Kyber and NVIDIA Rubin Ultra platforms.
NVIDIA to lead the industry's shift to 800 VDC data center power infrastructure starting in 2027
To support IT racks of 1MW and above, NVIDIA will lead the industry's shift to the 800 VDC architecture starting in 2027. The conventional 54 VDC distribution scheme was designed for kW-scale racks and cannot meet the requirements of MW-scale AI factories. Major data-center power system vendors such as Eaton, Schneider Electric, and Vertiv, and component makers such as Delta, Flex Power, and LiteOn, are aligning with the transition.
Vertiv reaches engineering readiness on the 800 VDC power architecture in collaboration with NVIDIA (October 2025)
On October 13, 2025, Vertiv (NYSE: VRT) announced it had reached engineering readiness on the 800 VDC power architecture for AI factories in collaboration with NVIDIA. Building on the strategic partnership announced in May 2025, the effort moved from the concept stage to the implementation-design stage. The product portfolio is scheduled for release in H2 2026.
Vertiv's 800 VDC products are timed to support the 2027 rollout of the NVIDIA Rubin Ultra platform
Vertiv's 800 VDC power portfolio, planned for release in H2 2026, is timed to align with the 2027 rollout of the NVIDIA Rubin Ultra platform. This is designed so that the next-generation GPU platform and the power infrastructure for AI factories come online in sync.
Vertiv reached engineering readiness five months after its May 2025 partnership announcement
Vertiv announced its strategic partnership with NVIDIA in May 2025, and on October 13 of the same year (about five months later) announced it had achieved engineering readiness on the 800 VDC architecture. This pace reflects how rapidly the industry is responding to AI infrastructure's power demands. Component specifications and platform design are now complete.
NVIDIA's Rubin Ultra-based Kyber rack is scheduled to begin shipping in 2027
The NVIDIA Kyber rack, equipped with Rubin Ultra GPUs, is scheduled to begin shipping in 2027. Schneider Electric plans to bring its sidecar to market ahead of that, providing engineers and data center operators with technical specifications and reference designs in advance.
Foxconn is building a 40MW 800VDC data center, "Kaohsiung-1," in Kaohsiung, Taiwan
Foxconn disclosed details at the OCP Global Summit of the 40-megawatt data center "Kaohsiung-1" under construction in Kaohsiung, Taiwan. The facility is designed for 800VDC (high-voltage DC) infrastructure and adopts a next-generation AI factory architecture for the gigawatt era. CoreWeave, Lambda, Nebius, Oracle Cloud Infrastructure, and Together AI are also involved.
Schneider Electric's 800 VDC sidecar is designed to support major hyperscalers including Google and Meta
The 800 VDC sidecar Schneider Electric is developing conforms to NVIDIA's emerging standard and is designed to support hyperscalers including Google and Meta. It also features a "Live Swap" capability aimed at greatly simplifying maintenance and shortening repair time.
Schneider Electric commits to 800 VDC power systems for next-generation GPUs (October 2025)
On October 13, 2025, in Boston, Schneider Electric announced expanded support for 800 VDC power infrastructure for NVIDIA's next-generation GPUs. As high-density rack architectures spread, 800 VDC is being adopted industry-wide as a key requirement for next-generation data centers. The company offers an integrated, system-wide approach spanning power conversion, protection, and metering.
Schneider Electric's data center CTO frames 800 VDC as a natural evolution driven by rising compute density
Jim Simonelli, the company's CTO for data centers, said the shift to 800 VDC is "a natural evolution as compute density increases," and pledged to help customers transition safely and reliably. He said the company's strength lies in understanding the entire power ecosystem from the grid to the server and designing integrated solutions.
NVIDIA's 800 VDC ecosystem includes 12 major semiconductor makers such as Analog Devices, Infineon, and STMicroelectronics
At the silicon-provider layer, Analog Devices, Infineon, Innoscience, MPS, Navitas, OnSemi, Renesas, ROHM, STMicroelectronics, and Texas Instruments are participating. At the power-system-component layer, Delta, Flex Power, Lead Wealth, LiteOn, and Megmeet are involved, and at the data-center power-system layer, additional major vendors participate.
More than 50 MGX partners are preparing to support NVIDIA's Vera Rubin NVL72
NVIDIA unveiled the open-architecture rack server specification for the Vera Rubin NVL72 MGX generation at the OCP Global Summit. More than 50 MGX system and component partners are working on support, with NVIDIA offering its compute-tray and rack innovations as an open standard through the OCP consortium.
The 800 VDC architecture improves end-to-end efficiency by up to 5% and reduces TCO by up to 30%
NVIDIA's 800 VDC architecture improves end-to-end efficiency by up to 5% by minimizing energy loss and reducing the number of PSUs required. Maintenance costs fall by up to 70%, cooling costs are also reduced, and total cost of ownership (TCO) is expected to drop by up to 30%. The core design principle is suppressing conversion loss by reducing the number of power-conversion stages.
Once rack power exceeds 200kW, 54 VDC distribution hits a physical limit as copper busbar current capacity breaks down
In current AI factories, power distributed at 54 VDC is delivered from rack-mounted power shelves to compute trays via large copper busbars. Once rack power exceeds 200kW, increasing current while staying at 54 VDC hits a physical limit in busbar weight, volume, and heat (since power equals voltage times current, low voltage maximizes current). This "copper overload" problem underpins the engineering necessity of the shift to 800 VDC.
Foxconn and NVIDIA jointly implement the 800 VDC power architecture for next-generation AI factories
On October 14, 2025, Hon Hai Technology Group (Foxconn) announced it would implement the 800 VDC power architecture in AI factories in collaboration with NVIDIA. The architecture was announced at Computex 2025 and detailed at the OCP Global Summit 2025. It is designed for high-density AI workloads with a modular, scalable structure.
The 800 VDC power architecture's first deployment site is the K-1 AI data center project in Kaohsiung, Taiwan
The new power architecture will first be implemented at Foxconn's K-1 AI data center project in Kaohsiung. The site serves as a demonstration facility for the group's AI server, data center, and renewable energy integration capabilities, and aims to reinforce Taiwan's strategic position in the global AI supply chain.
Schneider Electric and Foxconn (Hon Hai) announce a strategic collaboration for next-generation AI data centers (June 15, 2026)
Schneider Electric and Hon Hai Precision Industry (Foxconn) announced a strategic collaboration on June 15, 2026 to accelerate the construction and deployment of next-generation AI data centers. Combining Foxconn's compute platforms, AI rack integration, and global manufacturing capacity with Schneider's power, cooling, and energy management technology, the companies will offer an integrated, ready-to-deploy solution, with production planned to begin within 2026.
Foxconn Chairman Young Liu: "The industry needs to reinvent how AI infrastructure is designed, built, and delivered"
Foxconn Chairman Young Liu said that keeping pace with the speed of AI's evolution requires transforming the existing infrastructure delivery model. He explained that combining Foxconn's systems-integration strength with Schneider's power and energy expertise opens a path for customers to deploy AI capacity faster, smarter, and more sustainably.
