Main Facts: The New Frontier of AI Infrastructure
As the global appetite for artificial intelligence continues to accelerate, the physical infrastructure supporting these compute-intensive workloads is facing a power crisis. Modern data centers are no longer just storage facilities; they are hyper-scale engines of computation, driven by massive GPU clusters that demand unprecedented amounts of electricity. To address this, Infineon Technologies AG has unveiled a high-performance 27kW three-phase power supply unit (PSU) reference design, specifically engineered to meet the stringent demands of OCP Open Rack V3 (ORV3) specifications and next-generation AI server architectures.
The transition toward 800 VDC or ±400 VDC power architectures is no longer a luxury but a necessity for modern server ODMs and OEMs. By shifting to these higher voltages, data centers can minimize transmission losses and manage the extreme rack densities required by modern AI training and inference. Infineon’s new solution serves as a critical bridge, allowing developers to bypass long, iterative design cycles and deploy high-density, thermally efficient power systems that are ready for the global stage.
Chronology: The Evolution Toward High-Density Power
The journey toward this 27kW solution is the culmination of years of power management research at Infineon. For decades, the industry operated under standard 12V or 48V architectures, which were sufficient for traditional CPU-based server loads. However, the emergence of GPU-heavy AI clusters necessitated a paradigm shift.
- Pre-2020: The industry standard for data center power delivery remained largely stagnant, focusing on incremental efficiency gains.
- 2021-2023: As generative AI gained mainstream traction, the power requirements per rack surged from single-digit kilowatts to 50kW and beyond. Infineon began aligning its semiconductor portfolio—specifically its Silicon Carbide (SiC) and Gallium Nitride (GaN) technologies—to support higher power densities.
- Early 2024: Infineon identified the "thermal and space bottleneck," where the physical footprint of power supplies began competing with compute space within server racks.
- Late 2025: Following a year of refinement and internal testing, Infineon officially launches the 27kW reference design, integrating its advanced microcontroller (MCU) and sensor technologies to provide a complete, modular ecosystem for data center OEMs.
Supporting Data: Engineering Excellence in a Compact Footprint
The technical specifications of the new 27kW reference design are designed to shatter current performance ceilings. By leveraging a five-level ANPC (Active Neutral Point Clamped) PFC topology combined with a three-level LLC converter, Infineon has achieved a power density of 116 W/in³. This figure significantly outperforms the ORV3 baseline requirement of 94 W/in³, providing designers with roughly 23% more power density in the same physical volume.
Efficiency and Thermal Performance
Efficiency is the cornerstone of this solution. The design boasts a peak efficiency of over 98% at an input voltage of 480 VAC and a 50% load. In the context of hyper-scale data centers, a 2% improvement in efficiency translates to millions of dollars in electricity savings and a drastic reduction in the carbon footprint over the lifetime of the hardware.
Integrated Intelligence
The design’s "intelligence" is driven by a suite of Infineon’s proprietary components:
- Power Semiconductors: 650V CoolSiC™ and CoolMOS™ MOSFETs provide the backbone for high-frequency, high-efficiency switching.
- Digital Control: The PSOC™ Control C3 Performance Line microcontroller utilizes a Programmable Power Control Accelerator (PPCA). This allows the system to respond to AI-driven load transients—which can fluctuate in milliseconds—with extreme precision.
- Sensors and Drivers: EiceDRIVER™ gate drivers and XENSIV™ current sensors provide the real-time telemetry necessary to maintain power factor stability (above 0.99) and minimize total harmonic distortion (iTHD).
One of the most notable engineering feats is the integrated energy buffer. By providing a 20-millisecond hold-up time during grid disturbances and smoothing out rapid GPU load changes, the design effectively eliminates the need for an external capacitor bank unit (CBU). This not only saves cost but also frees up valuable rack space, allowing OEMs to allocate more area to compute and storage.
Official Responses and Strategic Vision
"The challenge of AI is that it doesn’t just require more power; it requires smarter power," noted a spokesperson for Infineon Technologies during the launch. "Our goal with this 27kW reference design was to simplify the complex. By providing a modular, scalable solution, we are empowering server architects to focus on their core mission—delivering the compute power needed for the next wave of AI innovation—without worrying about the underlying power delivery hardware."
Infineon’s approach underscores its broader corporate strategy: decarbonization and digitalization. With 57,000 employees and €14.7 billion in revenue for the 2025 fiscal year, the company is positioning itself as the primary semiconductor partner for the infrastructure of the future. The integration of planar magnetic structures in this design, which enables a compact, high-frequency transformer, demonstrates the company’s focus on vertical integration and modular design.
Implications: The Future of Data Center Design
The ripple effects of this technology will be felt across the entire data center value chain. As AI workloads continue to push rack densities toward the 100kW threshold, the industry must move away from custom, hand-built power subsystems toward standardized, high-density reference designs.
Reducing Time-to-Market
For OEMs and ODMs, the primary implication is a drastically shortened product development lifecycle. By adopting a pre-validated reference design that meets OCP ORV3 standards, manufacturers can bypass the long R&D cycles required to test thermal dissipation, electromagnetic interference (EMI), and transient load response.
Grid Stability and Sustainability
The design’s ability to operate across a three-phase input voltage range (311 to 528 VAC) ensures that it is "grid-agnostic," suitable for global data center deployments from North America to Europe and Asia. Furthermore, the high power factor and low iTHD ensure that the data centers of tomorrow do not become sources of grid instability, a growing concern as AI infrastructure expands in power-constrained urban environments.
The Shift to Modular AI Infrastructure
Finally, the elimination of the separate CBU signals a trend toward greater integration. In the world of high-density AI servers, every square inch of the rack is a commodity. By integrating energy buffering and digital control into the PSU itself, Infineon is setting a new standard for how server racks will be organized. The future is one of "Power-as-a-Module," where infrastructure can be swapped, scaled, and updated as efficiently as a software patch.
As we look toward the remainder of the decade, the ability to pack more compute into less space while consuming less energy will be the defining factor for AI competitiveness. Infineon’s 27kW reference design is not merely a component upgrade; it is a blueprint for the sustainable, hyper-scale data centers of the future. By merging advanced silicon materials with sophisticated digital control, the industry is taking a vital step toward ensuring that the digital revolution remains as sustainable as it is transformative.
