Complete power solutions enable the leap in AI server computing power

With the explosive growth of generative AI and large language models, the computing power demands of AI servers are rising exponentially. As the physical bearer of computing power, AI servers not only require more powerful GPUs/CPUs, but also urgently need a stable, efficient, and high-density power supply system to support their operation. However, cabinet power consumption often reaching tens of kilowatts, drastically shortened product iteration cycles, and increasingly stringent energy efficiency regulations are putting traditional power architectures under strain. This article will analyze the power architecture requirements of AI servers, as well as the core products and application features of the power solutions launched by Murata, revealing how they empower AI servers to achieve a leap in computing power.

To address the high computing power and high power consumption requirements of AI servers, Murata has launched centralized power management solutions that comply with OCP standards. Its core products, include the MWOCES-191/211 series (such as MWOCES-191-P-D and MWOCES-211-P-D) Power Shelves, are designed to solve the high-current supply challenges and soaring cabinet power density issues faced by AI data centers.
 
Due to the enormous power consumption of AI servers, research shows that the average cabinet power density is expected to exceed 30kW by 2027. To cope with high power density, Murata's Power Shelves will be key infrastructure for addressing this trend. In addition, Murata's Power Shelves support mainstream AI architectures. The solutions can flexibly support mainstream AI server architectures such as OCP and NVIDIA MGX, can fit 19-inch and 21-inch racks, and have been successfully deployed in real-world applications. Murata also provides a complete power chain, offering comprehensive power solutions from the cabinet level (Power Shelf) to the board level (DC-DC converters), fully supporting high-load computing.
 
Taking the MWOCES-191-P-D (19-inch) and MWOCES-211-P-D (21-inch) as examples, the core parameters of the two products are basically the same. Their main features include ultra-high power density, delivering up to 33kW of total output power within a compact 1U (about 4.45cm) height, and a flexible redundant architecture that accommodates up to 6 hot-swappable power supply units (PSUs) and 1 remote management unit (RMU), supporting N+1 or N+N redundancy to ensure high system reliability.
 
In addition, these two products also support intelligent remote management. With the MWOC-RMU-3 module, they can realize remote ON/OFF control, status monitoring, and fault information reporting. They also have wide input compatibility, supporting various inputs such as three-phase Delta, three-phase Wye, single-phase AC, and HVDC (192-400 Vdc), adapting to different data center power environments.
 
On the other hand, the MWOCES-191-P-D and MWOCES-211-P-D both offer ultimate energy efficiency. The accompanying power modules (such as the MWOCP67-5500-B-RM) have peak efficiency greater than 97.5%, meeting the 80Plus Titanium or even higher 80Plus Ruby energy efficiency standards. They also feature high-grade safety and reliability, complying with multiple international safety certifications including UL, CSA, TUV, CQC, and CB, and come with a 2-year warranty.
 
Murata's OCP power solutions, with their high power density, flexible modular design, intelligent management capabilities, and top-tier energy efficiency, precisely meet the stringent requirements of AI servers for power systems. They are an ideal choice for building and expanding AI data center infrastructure.

To address the stringent power and efficiency demands of AI servers, Murata has also launched front-end power modules such as the D1UXXX series (such as the D1U74T-W-3200), providing critical underlying power support. Taking the D1U74T-W-3200-54-HB3C as an example, this module delivers an output power of 3200W, with an output voltage of 54 Vdc main output + 12 Vdc auxiliary output. Its efficiency level reaches 80 Plus Titanium, with efficiency exceeding 96% at 50% load, significantly reducing energy loss and cooling costs in data centers. It also features key characteristics such as hot-swap support, active current sharing, and a PMBus™ digital management interface. With a compact 1U design, it achieves 3200W high power output within a very small 1U space, effectively saving rack space.
 
This 3200W high-power module is an ideal choice for building high-density computing clusters and can adapt to mainstream AI server architectures. The module complies with the Intel CRPS standard, flexibly adapting to mainstream AI server architectures such as OCP and NVIDIA MGX. It also supports GPU-accelerated computing and is specifically designed to meet the high-load demands of GPU servers in AI training and inference.
 
The D1U74T-W-3200-54-HB3C offers high availability and flexibility, supporting N+1 redundancy and hot-swap functionality, with multiple modules operating in parallel, facilitating online maintenance and power expansion. It is also equipped with a PMBus™/I²C digital interface, enabling intelligent management functions such as real-time control, programming, and remote monitoring of the power supply. Additionally, this product features a wide input voltage range, supporting globally universal wide AC input voltage, adapting to power grid environments in different regions.
 
Murata's D1U74T-W-3200-54-HB3C front-end power module, with its 3200W high power, Titanium-level energy efficiency, intelligent management capabilities, and high-reliability design, precisely meets the stringent power requirements of AI servers. It is a key component for addressing the high-power consumption challenges brought by AI computing power and for building stable, efficient, and scalable AI data centers.

In a typical AI server power architecture, the Intermediate Bus Converter (IBC) is a key link connecting the front-end power supply and the back-end loads. In the AI server power architecture, the power modules (PSUs) in the AC-DC Power Shelf are responsible for converting AC power to 48V or 54V DC power. Then, the DC-DC conversion is taken over by the IBC, which efficiently converts the 48V/54V voltage to a 12V intermediate bus voltage. Next, the voltage regulator modules (VRMs) on the board convert the 12V voltage to the extremely low voltages (such as below 1V) required by core chips like CPUs and GPUs.
 
To address the power demands brought by high computing power in AI servers, Murata has introduced IBC products such as the MPQXXX series (e.g., MPQ800-12V67-L48NBMC), which play a critical role in the board-level power conversion architecture. Taking the MPQ800-12V67-L48NBMC as an example, it supports 800W output power, with an input voltage of 40V - 60Vdc, an output voltage of 12Vdc, a maximum output current of 125A, and efficiency up to 97%. Its physical dimensions are only 1/4 brick (58.42 x 36.83 x 15.0 mm), and it can operate over a temperature range of -40℃ to +125℃, with UL62368-1 safety certification.
 
The MPQ800-12V67-L48NBMC features ultra-high conversion efficiency of 97%, effectively reducing energy loss and heat generation, which is crucial for high-density AI servers. It also offers high power density, delivering 800W output in a compact 1/4 brick package, saving valuable PCB space. It supports digital control and intelligent management, with an optional PMBus™ interface for real-time monitoring, fault management, and output voltage trimming. It integrates comprehensive protection mechanisms, including overcurrent, overtemperature, and overvoltage protection, ensuring system safety under abnormal conditions. It also features high reliability, with 2250V isolation voltage providing excellent system safety and noise immunity.
 
Murata's MPQ800 IBC module, with its 97% ultra-high efficiency, 800W high power density, intelligent digital control, and complete protection features, precisely meets the stringent requirements of AI servers for board-level power conversion.

In the board-level power conversion architecture of AI servers, Charge Pump products play a vital role in the key "48V to 12V" voltage conversion stage. They share similar functions with traditional Intermediate Bus Converters (IBCs), but with their unique "lossless" technology, they can achieve higher efficiency and power density in specific scenarios.
 
Murata's charge pump products are mainly divided into two categories: one is the standalone charge pump module, such as the MYCXXXX series (like MYC0409-NA); the other is the POL (Point-of-Load) regulator module that integrates charge pump technology, such as the MYTXX series (like MYTNA1R84RELA2RA). They work together in AI servers, jointly forming a complete and efficient power supply chain from the 48V bus to the core chips.
 
The MYC0409-NA is a high-efficiency 48V to 12V standalone charge pump module from Murata's UltraCP™ series. It uses proprietary lossless charge pump technology and is specifically designed to efficiently convert 48V bus voltage to 12V intermediate bus voltage. Its input voltage is 20V - 60V (supporting 48V bus architecture), output voltage is Vin/4 (e.g., 12V output at 48V input), maximum output current is 6A, maximum output power is 72W, peak efficiency is up to 96.5%, and physical dimensions are 11.5 x 9.5 x 2.0 mm. It can operate over a temperature range of -40℃ to +105℃.
 
The MYC0409-NA features ultra-high efficiency and low heat generation. Its lossless charge pump technology reduces converter losses by approximately half, thereby lowering product temperature by 50°C or more. Through high integration, its package size is much smaller than traditional brick-style IBC modules, helping to implement distributed power architectures on compact AI server boards. It also supports parallel operation, allowing flexible system construction from low power to high current by simply increasing the number of modules. In addition, the MYC0409-NA has comprehensive protection features, integrating enable control, Power Good indication, frequency synchronization, overcurrent protection, overtemperature protection, undervoltage lockout, and other functions.
 
In AI servers, the MYC0409-NA can directly replace larger traditional IBC modules on AI accelerator cards or motherboards, converting 48V voltage to 12V to power downstream loads such as GPUs, CPUs, and memory. Multiple MYC0409 modules can also be distributed close to the loads (such as GPUs) to form a distributed power architecture, which reduces wiring losses by about 60%, thereby improving overall system efficiency. Moreover, by paralleling multiple MYC0409 modules, a high-current, high-efficiency 12V power array can be built to meet the demands of high-power accelerator cards in AI servers.
 
On the other hand, the MYTNA1R84RELA2RA is a POL regulator with an integrated charge pump. It is a two-stage DC-DC converter from Murata's UltraBK™ series, cleverly combining charge pump technology with a subsequent buck regulator to directly provide ultra-low voltage and high current from the 12V bus to AI core chips. The MYTNA1R84RELA2RA has an input voltage of 6V - 14.4V (suitable for 12V bus), an output voltage of 0.7V - 1.8V (adjustable), a maximum output current of 4A, an output power of 7.2W, and can operate over a temperature range of -40°C to +105°C.
 
The MYTNA1R84RELA2RA also features a highly integrated "Ultra Block" internally, integrating all passive components including inductors within an extremely thin package. It adopts a two-stage high-efficiency architecture, using a two-phase charge pump as the first stage, coupled with a second-stage regulator, achieving ultra-high conversion efficiency. Its ultra-thin design, with a typical height of only 2.1mm, is ideal for placement in areas with strict height constraints, such as on the back side of AI accelerator cards. It also has comprehensive protection mechanisms, integrating output overcurrent protection, short-circuit protection, and overtemperature protection to ensure system safety under abnormal conditions such as load faults.
 
In AI servers, the MYTNA1R84RELA2RA can power core chips. As a last-mile POL regulator, it is placed directly near core chips such as GPUs, CPUs, and AI ASICs, efficiently converting the required 12V voltage to ultra-low core voltages below 1V. It also solves high-density layout challenges. With its ultra-compact size and ultra-low profile, it can be easily integrated into the increasingly compact layouts of AI accelerator cards, supporting high-density computing.

Today, as the AI computing power race intensifies, power supply capability has become one of the key bottlenecks restricting system performance. Murata does not offer just a single component, but rather a proven, collaboratively optimized, and full-link complete power solution. From the cabinet-level MWOCES Power Shelves, to the onboard MPQ800 IBC modules, to the MYC series charge pumps and POL converters placed right next to the chips, Murata integrates high efficiency, high power density, and high reliability through its outstanding engineering capabilities. This not only helps data center operators significantly reduce power loss and cooling costs, but also unleashes the maximum computing potential of AI servers within limited physical space. For developers and enterprises seeking to build stable, efficient, and scalable AI infrastructure, Murata's complete power solution is undoubtedly a choice worthy of in-depth evaluation and trust.