Infineon and NVIDIA Drive the Next Generation of AI Data Centers with 800 VDC Power Architecture

October 13, 2025 – Infineon Technologies AG (FSE: IFX / OTCQX: IFNNY) announced its support for NVIDIA’s introduction of an 800-volt DC (VDC) power architecture for artificial intelligence (AI) infrastructure at Computex 2025 in Taipei.

 

The explosive growth of AI is quickly outpacing the capacity of today’s 54-volt data center power systems. Transitioning to a centralized 800 VDC architecture significantly reduces power losses while improving both efficiency and reliability. However, this next-generation architecture also introduces the need for new power conversion solutions and safety mechanisms to prevent potential hazards and minimize costly downtime during maintenance or service interruptions.

 

“There is no AI without power,” said Adam White, President of Infineon’s Power & Sensor Systems Division. “That’s why we’re working with NVIDIA to develop intelligent power systems that meet the massive energy demands of future AI data centers — while ensuring a maintainable architecture that minimizes system downtime. By enabling high-density, reliable, and safe 800-volt power delivery, we are redefining how power reaches AI server racks. Our mission is to extract the maximum value from every watt, paving the way toward a more efficient and sustainable AI ecosystem.”

 

Rethinking Power for the AI Era

Data center operators are investing heavily in AI-driven high-performance computing. Mega-scale projects, such as “AI Stargate”, require multibillion-dollar capital expenditures (CAPEX). Ensuring maximum uptime for AI server racks is therefore essential to achieving strong returns on investment. This calls for modular, scalable, and maintainable power architectures.

 

Infineon and NVIDIA are jointly addressing this challenge through advanced safety and serviceability features, such as hot-swap controller functions that allow future server motherboards to operate safely within an 800 VDC environment. These features enable controlled pre-charging and discharging of server boards on an active 800 VDC bus — allowing components to be replaced without shutting down the entire rack.

 

Infineon supports this capability with specialized silicon carbide (SiC)-based solutions, leveraging its CoolSiC™ JFET technology. This allows data center operators to replace or service motherboards while other servers continue running within the same rack, minimizing downtime and ensuring safe, continuous operation.

 

Building the Foundation for AI “Power Factories”

The migration to an 800-volt DC power architecture marks a critical step toward the future of large-scale AI “power factories.” AI server rack power consumption is projected to rise dramatically — from around 120 kW today to 500 kW, and even up to 1 megawatt by the end of the decade.

 

As announced in May 2025, Infineon and NVIDIA are co-developing advanced power conversion solutions for 800 VDC-powered accelerated computing platforms. Infineon is applying its expertise in intermediate bus converter (IBC) technology and GaN-based high-frequency switching solutions to accelerate the development of both two-stage and three-stage conversion topologies. Each stage is designed to deliver efficiencies of up to 98%, giving data center operators the flexibility to adopt architectures best suited to their specific performance and reliability goals.

 

Efficiency, Uptime, and Sustainability

With AI servers costing up to 30 times more than conventional servers, reducing energy consumption and maximizing uptime are vital to minimizing total cost of ownership and lowering the carbon footprint of AI data centers.

To meet these demands, Infineon provides a comprehensive system-level portfolio — including hot-swap controllers, robust protection technologies, and high-efficiency power conversion solutions based on Si, SiC, and GaN materials. Together, these technologies enable megawatt-scale AI server racks to operate safely, efficiently, and sustainably — powering the intelligence of tomorrow.