Your AI Power Testing Partner in the NVIDIA Ecosystem
On May 20, 2025, NVIDIA officially announced the 800 VDC data center power architecture, ushering in a fundamental shift in power delivery. Power and electronics suppliers worldwide are gearing up for what may become the greatest transition in electrical systems since the “war of the currents” in the 19th century. Two key technologies are making this possible: HVDC (high-voltage direct current) and SST (solid-state transformer).
HVDC: The Enabler of the AI Era
Traditional data center architectures have long relied on 50 VDC as the standard operating voltage for IT equipment. However, the explosive power demands of the AI boom have pushed this standard to its limits. Maintaining 50 VDC means busbar currents become unmanageable. Even with engineering solutions like sidecar modules or liquid-cooled busbars, massive heat loss and footprint sacrifices remain unavoidable.
To overcome these limitations, the Diablo 400 project (a collaboration between Microsoft, Meta, and Google) adopted a ±400 VDC or 800 VDC HVDC power distribution architecture, significantly improving transmission efficiency and reducing energy losses.
For AI data centers, power is the new computing bottleneck. The adoption of HVDC allows energy to travel through the shortest path with minimal conversion losses to directly feed the GPUs. This aligns with NVIDIA’s vision for 800 VDC: a fully DC-powered ecosystem, delivering energy from the grid straight to AI compute.
SST: Bridging Power and Intelligence
Solid-state transformers (SSTs) use advanced power semiconductor switching to achieve efficient AC/DC conversion. This allows for directly transforming 13.8 kVAC grid power into 800 VDC to supply AI racks. With SSTs enabling direct HVDC 800V delivery from grid to rack, overall energy efficiency improves by more than 5%. For AI data centers operating at the gigawatt scale, this gain is not just an optimization, but may well be the only viable path forward to control energy and cost. Compared with traditional transformers, SSTs offer higher efficiency, smaller footprint, and reduced copper and iron usage. Their modular design also enhances flexibility and scalability, which positions SSTs as a pivotal element in future AI power architectures.
In NVIDIA’s October 13, 2025 publication, “Building the 800 VDC Ecosystem for Efficient, Scalable AI Factories”, the company highlights SSTs as a high-potential enabler of the 800 VDC infrastructure.
Setting the Standard in AI Power Testing: Chroma 62450D-2000HL
As NVIDIA and global CSPs (cloud service providers) advance their HVDC integration, power testing and validation become mission-critical with zero tolerance for error. In AI workloads, instantaneous power surges demand uncompromised reliability, as any brief drop in power can undermine GPU computation. To address this demand, Chroma developed the 62450D-2000HL, a bidirectional DC power supply and regenerative electronic load purpose-engineered for high-voltage, high-power testing. Thanks to its bidirectional design, the 62450D-2000HL can switch between modes to perform dual roles:
- Source Mode: Provides 800 VDC output, simulating the power source to validate server rack performance.
- Load Mode: Accepts 800 VDC input, emulating server rack loads to verify power rack stability.
With a 1.8MW power capacity, the system fully supports HVDC power rack standards. Its 94% energy regeneration efficiency not only reduces testing energy costs but also advances the vision of sustainable, green power testing.
Enabling the Future of AI Power
In the AI-driven era, electricity is no longer just energy. It’s fast becoming the foundation of intelligence. Through innovative testing technologies and high-power solutions, Chroma ATE provides the most reliable validation platform for HVDC systems and SST equipment, empowering global partners to advance confidently into the 800 VDC future. Together, we power the intelligence of tomorrow.
Discover how Chroma’s 1.8MW Bidirectional Power Solution is powering the next generation of AI HVDC architectures.