Empowering the AI Era - Preen Power Supplies for AI Server Testing

Introduction to AI Server

In today’s digital era, AI servers are the engines powering the world’s most advanced technologies. Unlike traditional servers, they are purpose-built to meet the explosive demands of artificial intelligence, equipped with high-performance GPUs, ultra-fast interconnects, and large-scale memory to process massive datasets at unprecedented speed.

From accelerating drug discovery in healthcare, to powering fraud detection in financial services, to enabling predictive maintenance in manufacturing, AI servers aren’t just hardware; they are the engines fueling innovation across industries.

Current VS Future of AI Server

Legacy AI servers were designed around 12 V DC, but this is no longer sufficient for today’s high-power AI workloads. To improve efficiency, modern systems increasingly adopt 48 V to 54 V DC power supplies, as they stay within the safety limit of 60 V while balancing efficiency with manageable current levels. Most AI server today operates on 48V DC chassis. The shift to 54 V DC distributions becomes critical as it reduces current flow, lowering copper losses over long distances, improves overall efficiency for advanced server infrastructures.

In hyperscale deployments, some data centers already use ~400 V DC rack distribution as an intermediate stage, stepping down to 48–54 V DC for individual server chassis.

To meet the rapidly growing power demands of AI servers, NVIDIA is advancing 800 V high-voltage DC (HVDC) architectures that deliver power directly to server racks. This design minimizes conversion stages due to inefficient repeated conversion of AC/DC throughout the power chain, reduces transmission losses, and simplifies cabling that is unsustainable for the future of data center. These reasons are especially important for next-generation platforms such as NVIDIA’s Rubin Ultra AI servers, expected in 2027 and projected to draw up to 600 kW per rack.

Supporting AI Server Testing with Preen Power Supplies

The rapid growth of AI computing and diversity of system specification drive the need for dependable power testing. In this proven on-site application, Preen’s high-performance DC power solutions ensure stable and precise testing for AI servers, ensuring accuracy, reliability, and compliance in high-demand AI server validation.

Key Requirements for AI Server Testing:

1. High Power Output: Delivers stable power to GPU clusters with power range on DC 5?kW-300?kW, up to 1.8?MW in parallel;

2. Programmability: Supports predefined QC parameters (burn-in, aging) to simplify testing;

3. Safety and Compliance: Built-in protection against overcurrent, overvoltage, and temperature for sensitive AI components;

4. Simulate Power Disturbances for Reliable Testing: Generates voltage dips, ramps, and interruptions per IEC 61000-4-29 to verify equipment resilience to power disturbances;

5 Flexible Integration: Supports SCPI, Modbus, and LabVIEW for seamless automation and system integration.

Real-world Use Case: Advanced AI Server Testing

In a real world case, a customer purchased the ADG-L Low Voltage (LV) series model ADG-L-80-510-15, a high-density programmable DC power source delivering 15 kW at 80 V DC in a compact 3U form factor — to perform long-duration reliability tests on AI servers. The objective was to evaluate AI Server performance under hours of unbalanced voltage conditions (54.5V DC ±10%).

As it is built-in programmable feature, customer only needs to preset the test parameter with a single click, power output will run the test autonomously. Combined with on-board remote interface and complimentary Preen Program, allow the test to run remotely and generate report within the software. For customer developing their own testing system, ADG-L support both SCPI and Modbus command for seamless integration.

Compatibility to Future of AI Server Needs

As AI servers move toward higher-voltage specifications, Preen is fully prepared to meet these demands with the ADG+ series. The series supports 30kW to 300kW in single unit, and up to 1.8MW with master slave parallel operation, capable of supporting next-generation AI servers that demand up to 800 V DC and 600 kW per rack. The unit also includes programmable feature such as DC pulse, voltage sag, and slew rate control for various testing requirement.