This article by technology analyst Matthew Fall examines the impact of these innovations on AI data centers. How 800V EV innovations are redefining AI data centers. Over the past decade, the electric vehicle industry has become the dominant application area for power electronics. Most power electronics players have enjoyed the automotive industry as their primary source of revenue for years, and power electronics innovations have primarily focused on improving the efficiency and performance of electric drive systems. However, over the past two years, major power electronics manufacturers have seen their revenue from the automotive industry gradually stagnate, amid reduced government incentives and increasingly fierce competition from highly vertically integrated Chinese EV manufacturers.

Although IDTechEx continues to forecast long-term growth in the electric vehicle market, many players are exploring alternative and growing application areas for power electronics. Data centers, and especially AI data centers, represent a rapidly growing and innovating market, with increasing adoption of broadband semiconductors such as SiC and GaN to support increasingly powerful and energy-intensive AI training models. According to IDTechEx's forecasts in the market report "Power Electronics Market 2026-2036: Data Centers, Electric Vehicles, and Renewables," the power electronics market for data centers is expected to grow 2.5 times by 2036.

In the coming years, a paradigm shift will occur in data center power architecture, with the move towards HVDC (800 VDC) data centers, resulting in an order-of-magnitude increase in rack power. This overhaul of data center power architecture is directly influenced by the power electronics of 800 V electric vehicles, from the materials used to the high-voltage safety and protection mechanisms.

AI data centers are facing a bottleneck and are looking to electric vehicle power electronics for inspiration. In November 2022, ChatGPT brought artificial intelligence from science fiction into the real world. Since then, AI has become increasingly mainstream, and leading AI software companies, particularly in the US and China, have invested billions of dollars in developing more complex and powerful AI models. The GPU industry has rapidly evolved to accommodate more advanced training models.

Each new generation of servers requires more power, and the current data center rack architecture simply can't support Nvidia's next-generation "Vera Rubin Ultra" servers. Rack power levels have skyrocketed, from around 20 kW before ChatGPT to 100 kW today, and rack power is expected to exceed 1 MW by the end of the decade. With today's silicon power electronics and 480/54 V data center architecture, a 1 MW rack would require dedicating the entire rack space to power conversion and would need more than 200 kg of copper.

Data center efficiency was and remains a key design consideration. Data center power density—the amount of power processed per unit volume—will become an increasingly critical metric for AI data center design. In “Power Electronics Market 2026–2036: Data Centers, Electric Vehicles, and Renewables,” IDTechEx analyzes two key data center advancements that will facilitate this shift: the adoption of broadband semiconductors and 800V DC data center architecture. Both have been made possible directly by technological innovations in electric vehicles.

The 800V architecture for electric vehicles, facilitated by wideband semiconductors (WBG), enables a radical change in powertrain efficiency

The 800V architecture for electric vehicles began commercialization in 2019 with the Porsche Taycan and has been steadily adopted in high-performance models and, increasingly, in budget-friendly models. The two main drivers of the adoption of an 800V platform are related to increased charging speeds and reduced wiring weight and I2R losses, which improve the overall efficiency of electric vehicles. The move to 800V has been supported by improvements and the commercialization of wideband semiconductors (WBGs), particularly silicon carbide (SiC). At 800V, the efficiency improvements of SiC compared to existing silicon IGBT and silicon MOSFET technologies are significant and offset the higher cost of silicon carbide power semiconductors.

Given that electric vehicles account for the majority of the power electronics market share, innovation in this field has focused on electric vehicles and electric powertrains for both SiC and GaN technologies. This has led to significant commercialization and cost reductions for both broadband materials, as well as demonstrating their long-term reliability in harsh environments and their superiority over silicon in high-voltage applications. Innovations in GaN have been significant due to the strong interest surrounding its potential future adoption in electric vehicles. However, the adoption of GaN in electric vehicles has been slow, as the material is better suited for low-voltage, high-frequency applications.

Innovations in broadband semiconductors and HVDC are future-proofing data centers for next-generation AI training and inference

Thanks to technological innovations and the commercialization of broadband semiconductors, largely driven by electric vehicles, the adoption of broadband technology in data center power electronics has accelerated. Reference designs for 8 kW and 12 kW power supply units (PSUs) from leading power electronics players such as Infineon and Navitas incorporate SiC for high-voltage conversion and power factor correction (PFC) stages, and GaN for low-voltage conversion stages.

With lower voltage conversion and the possibility of much faster switching (down to MHz), the advantages of GaN in data centers are highly desirable; GaN is likely to become more widespread in data centers than in electric vehicles. The shift to WBG semiconductors in data centers leads to increased efficiency, but more importantly, it also brings a significant increase in power density, with smaller SiC and GaN devices and a reduction in the size of passive components such as capacitors and inductors.

Just as WBG semiconductors enabled the shift to 800 V in electric vehicle architecture, they will also facilitate the transition to 800 V DC/HVDC AI data centers. With the precedent of the 800 V architecture and the development of the necessary safety assessments and processes—which can be adapted from electric vehicles to data centers—the development of HVDC data center architecture can be accelerated by leveraging existing electric vehicle power architectures. The transition to 800 V DC is expected to simplify data center architectures and future-proof them.

With fewer power conversion stages and lower I2R losses, data center efficiency will increase. Fewer points of failure reduce the likelihood of downtime. Most importantly, the 800 VDC data center architecture enables the delivery of significantly higher power levels to server racks, supporting future generations of GPUs and AI training models. The interconnectedness of power electronics applications complicates the market. Innovations in one application can directly influence large-scale changes in another. "Power Electronics Market 2026-2036: Data Centers, Electric Vehicles, and Renewables" analyzes cross-sector innovations in materials, manufacturing, and device architecture, providing a clear overview of the power electronics industry as a whole.

Author: Matthew Fall, Technology Analyst