STMicroelectronics announced three new power conversion architectures designed to make AI data centers more efficient and scalable. The company expanded its 800 VDC (volt direct current) power portfolio with two new solutions: 800 VDC to 12V and 800 VDC to 6V converters, complementing an earlier 800 VDC to 50V offering. These components, developed according to NVIDIA's 800 VDC reference design, address a critical infrastructure challenge as hyperscalers like Microsoft, Google, and Amazon race to build massive AI training facilities .

Why Does Power Conversion Matter for AI Infrastructure?

Most people think of AI data centers as a computing problem, but the real bottleneck is often power delivery. Modern AI accelerators (graphics processing units, or GPUs) require precise voltage levels at different stages of the power distribution chain. Traditional multi-stage conversion systems waste energy through resistive losses and require more copper wiring, which adds cost and complexity. The 800 VDC architecture represents a shift toward higher voltage distribution that reduces these losses significantly .

The new converters enable what the industry calls "rack-level efficiency," meaning power reaches the GPUs with minimal waste. This matters because a single AI training cluster can consume 100 megawatts or more of electricity. Even a 1% efficiency gain translates to millions of dollars in annual operating costs and reduced strain on regional power grids .

What Makes These New Converters Different?

STMicroelectronics designed the new 12V and 6V conversion stages to eliminate intermediate power distribution steps. The 800 VDC to 12V converter enables high-efficiency distribution directly from rack-level power shelves to the voltage domains feeding advanced AI accelerators. The 800 VDC to 6V path allows manufacturers to move the conversion point closer to the GPU itself, reducing copper usage and minimizing resistive losses .

The key innovation is architectural flexibility. Different GPU generations, server heights, and cooling strategies require different power delivery topologies. By offering 50V, 12V, and 6V intermediate buses, STMicroelectronics enables data center operators to optimize for their specific hardware configuration rather than forcing a one-size-fits-all approach .

"As AI infrastructure compute scale continues to expand fast, it requires higher voltage distribution and greater density, which can only be achieved with system-level innovation for each of the different AI server form factors," stated Marco Cassis, President of Analog, Power and Discrete, MEMS and Sensors Group Head of Strategy, System Research and Applications, Innovation Office at STMicroelectronics.

Marco Cassis, President, Analog, Power and Discrete, MEMS and Sensors Group Head of Strategy at STMicroelectronics

How Do These Converters Improve Data Center Operations?

• Efficiency Gains: The new architectures eliminate the traditional 54V intermediate stage, reducing conversion steps and system-level losses while achieving efficiency targets exceeding previous two-stage conversion paths.
• Reduced Copper Usage: Moving power conversion closer to the GPU minimizes the amount of copper wiring required, lowering material costs and simplifying integration for future GPU generations.
• Faster Response Times: The 800 VDC to 6V design minimizes IR drop (voltage loss across resistance) and improves transient performance, a critical differentiator for large-scale training clusters that experience rapid load changes.
• Scalability for Dense Configurations: The portfolio completes the topology options for servers with ultra-dense GPU configurations, enabling hyperscalers to pack more compute into existing rack space.

STMicroelectronics demonstrated the viability of this approach in October 2025 with a fully integrated prototype power delivery system. The prototype featured a GaN (gallium nitride) based LLC converter operating directly from 800 volts at 1 megahertz with over 98% efficiency and exceptional power density exceeding 2,600 watts per cubic inch at 50 volts, all in a smartphone-sized footprint .

What Technologies Enable These Solutions?

The three solutions combine multiple semiconductor technologies across power semiconductors (silicon, silicon carbide, and gallium nitride), analog and mixed-signal components, and microcontrollers. This integration approach allows STMicroelectronics to optimize each stage of power conversion for its specific voltage level and current requirements .

The use of GaN technology is particularly significant. GaN semiconductors can switch at higher frequencies and voltages than traditional silicon, enabling smaller, more efficient power conversion stages. This is why the prototype achieved such high power density in such a compact form factor .

Why Should Data Center Operators Care Now?

The timing matters because hyperscalers are planning multi-gigawatt AI infrastructure deployments over the next three to five years. Power efficiency directly impacts three critical factors: operating costs, environmental sustainability, and grid stability. A data center that wastes 5% of its power through inefficient conversion consumes significantly more electricity than necessary, driving up both utility bills and carbon emissions .

Additionally, regional power grids are already stressed by AI data center demand. More efficient power delivery means less total electricity consumption, reducing pressure on utilities and making it easier for data centers to secure power supply agreements. This is particularly important in regions where electricity capacity is limited .

STMicroelectronics' expanded portfolio signals that the industry is moving beyond one-off solutions toward a complete ecosystem for 800 VDC infrastructure. As more hyperscalers adopt this standard, component suppliers like STMicroelectronics, NVIDIA, and others will continue refining the technology to unlock additional efficiency gains and support increasingly dense AI compute configurations.