From GPU to data center power revolution: NVDA.US (NVDA.US) explores another growth point and joins hands with Onsemi (ON.US) to deploy 800V DC

The Zhitong Finance App learned that the stock price of chip giant Ansemi Semiconductor (ON.US) for the automotive and industrial sector rose more than 6% before the market on Tuesday. Earlier, the chip company announced a partnership with the “AI chip hegemon” Nvidia (NVDA.US) to accelerate the promotion of 800 volt DC (or 800V DC) power solutions for next-generation artificial intelligence data centers.

At the core of this transformation is a new power distribution system that must deliver large amounts of electricity with minimal loss per voltage conversion. Ansemi Semiconductor said that its intelligent power solution is the most critical part of supplying power to next-generation AI data centers, and can achieve high-efficiency and high-power density power conversion at every stage.

As artificial intelligence applications such as ChatGPT, Claude, and DeepSeek take the world by storm, it can be said that the electricity demand for large-scale AI data centers around the world is becoming more and more huge. The scale of high-energy AI data centers, which are expanding exponentially with the intense demand for AI computing power infrastructure such as AI chips, is inseparable from the core foundation of power supply. This is also the origin of the market view that “the end of AI is electricity.” According to a forecast report from the International Energy Agency (IEA), global data center electricity demand will more than double by 2030, reaching about 945 terawatt-hours (TWh), which is slightly higher than Japan's current total electricity consumption. Artificial intelligence applications will be the most important driving force for this growth; it is expected that by 2030, the overall electricity demand for data centers focusing on artificial intelligence will more than quadruple.

The power consumption of AI data centers around the world is rapidly jumping from the traditional 20-30 kW per rack to 500 kW or even 1 MW. To transport several times the current within the same cabinet space without letting the copper bars, cables, and converters heat out of control, the industry is moving from a 48V/400V DC stage to an 800V high voltage direct current (HVDC) architecture.

The 800V DC solution jointly announced by Ansemi Semiconductors and Nvidia is this evolution. By increasing the bus voltage of the data center room from 48V or 380/400V DC to 800 V DC, current and I²R losses can be reduced to 1/10 at the same power, and the number of copper rows, cables, and conversion stages can be significantly reduced, thereby improving overall power supply efficiency by at least 5 percentage points, and leaving power space and cooling margin for Nvidia's “AI factory” with a single frame of 500 kW-1 MW in the future.

According to information, at this stage, hyperscale cloud computing and most AI training/inference cluster computer rooms are still dominated by 48V DC (OCP Open-Rack, etc.) or 380/400V DC parallel power architectures. 800V is a next-generation high-voltage DC (HVDC) solution that has just begun to be piloted. Nvidia, Ansemi Semiconductors, etc. are leading the demonstration in 2025-2027.

Currently, the mainstream power system is still 48V and some 380V DC; 800V can be said to be a future paradigm introduced in the early stages, targeting 1MW racks. Nvidia's single AI GPU server is close to 1kW, fully equipped with the Nvidia NVL AI server cabinet; Nvidia plans to mass-produce a 1MW AI Factory AI server rack cluster in 2027. Only 800 V high voltage busbars can control the busbar size and heat loss within the usable range, and 800V busbars are also easier to cooperate with immersive/cold plate liquid cooling, mainly due to characteristics such as low current and low heating point.

For Nvidia, the world's highest-capitalized tech giant and the first “A-chip hegemon” with a market capitalization exceeding $4 trillion, new revenue generation channels may soon be added — after all, 800V DC HVDC distribution will probably be the new power backbone for future megawatt AI data centers. This also means that after AI GPUs, InfiniBand, a high-performance networking solution for data centers, and NVIDIA DRIVE Thor, an automotive-grade high-performance SoC platform, Nvidia will usher in a new channel of performance growth with extremely strong demand.

For Ansemi Semiconductor, the company is expected to provide 800V grade SiC MOSFETs, solid state transformers, and high-density DC-DC modules, and is the core power device/power supply OEM for the complete solution. With Nvidia's exclusive large-scale ecosystem platform deployment, Ansemi Semiconductor can be described as directly targeting long-term large orders and strengthening its competitive advantage over competitors such as Infineon (Infineon) in the high-voltage SiC power supply market.