NVIDIA Awards Innoscience for AI Datacenters Power Innovation

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

NVIDIA Recognizes Innoscience with GaN Supplier Award as 800 VDC Rack Power Architecture Emerges as Critical AI Infrastructure Enabler

At its annual GTC conference in San Jose (March 15–19), NVIDIA spotlighted not only advances in AI software and large language models, but also the hardware ecosystem underpinning next-generation AI computing. Among the highlights was the presentation of supplier awards to key semiconductor and system partners enabling the rapid evolution of AI datacenter infrastructure. Innoscience, a gallium nitride (GaN) power semiconductor manufacturer, was recognized for its contribution to 800 VDC rack power architecture, which is a critical design shift aimed at addressing the escalating power demands of AI workloads.

800V DC Rack Architecture: A Structural Shift in Datacenter Power

The transition to 800 VDC rack-level power represents a significant departure from legacy datacenter designs. By increasing the input voltage to 800V—analogous to the shift seen in electric vehicle battery systems—datacenter operators can dramatically improve power delivery efficiency.

Key benefits include:

• Higher power throughput: GPU board-level power delivery increases more than fourfold, moving from traditional 12V systems to approximately 54V.
• Reduced current and conduction losses: Higher voltage reduces current by a factor of four, significantly lowering resistive (I²R) losses in copper interconnects.
• Improved energy efficiency: Overall datacenter system efficiency can reach 70%, compared to typical 55% in legacy datacenters, translating to substantially lower thermal losses and cooling requirements.

As AI compute density increases, such architectural changes are becoming essential to maintain both performance scaling and energy efficiency.

Role of Wide Bandgap Semiconductors: GaN and SiC

The viability of 800 VDC rack architecture is closely tied to the adoption of wide bandgap (WBG) semiconductors, particularly GaN, but also silicon carbide (SiC). Both technologies have matured over the past decade and have been adoped in mass production.

• GaN advantages:
  GaN devices offer superior switching efficiency at high frequencies, enabling smaller passive components and higher power density. This is especially critical for board-level power modules embedded directly within GPU assemblies.
• Form factor constraints:
  The 54V conversion stage, located near or on the GPU board, requires extremely compact and efficient power delivery. Currently, GaN is the only technology capable of meeting these density requirements at scale.
• System-level efficiency gains:
  Beyond reducing conduction losses, GaN significantly lowers switching losses within power conversion stages, further improving overall system efficiency.

While SiC remains competitive in high-voltage stages, GaN is expected to dominate the low-voltage, high-density segment critical to GPU-level power delivery.

Expanding Market Opportunity for GaN in AI Infrastructure

The rapid increase in AI compute density—driven by exponential growth in transistor counts and data throughput—has made power delivery a critical focus for the solution.

Industry estimates suggest:

• Per-rack GaN value: Over US$170,000 in GaN content for a megawatt-scale AI compute cabinet
• Datacenter-scale opportunity: Approximately US$180 million in GaN demand for a gigawatt-scale AI facility

As GPU architectures evolve to support multi-agent AI workloads and multi-terabit data processing, power efficiency and density will become even more critical, reinforcing GaN’s strategic importance.

Example of GaN content onboard a single GPU power supply Demo Board

Competitive Positioning: Innoscience and the Global GaN Landscape

Innoscience’s recognition by NVIDIA reflects both its manufacturing scale and its positioning within the global GaN supply chain.

Key factors include:

• Manufacturing scale: The company operates the industry’s largest 8-inch GaN fabrication facilities, with planned capacity approaching 70,000 wafers per month across its Suzhou and Zhuhai sites.
• Shipment volume: Cumulative shipments have exceeded 2 billion GaN devices, supporting applications across consumer electronics, industrial power supplies, solar inverters, and EV onboard chargers.

Innoscience: Scale Matters ~ Capacity and Reliability

Innoscience’s recognition is less about a single design win and more about positioning. The company has been aggressive in building manufacturing scale, particularly with its 8-inch GaN fabs in Suzhou and Zhuhai. At full capacity, output is expected to approach 70,000 wafers per month — a level that no GaN competitor matches as of today.

Innoscience also has more than 2 billion GaN devices in production in consumer power supplies, industrial applications, solar inverters, and EV onboard chargers. Innoscience’s statistical database for high reliability puts it at the forefront as a GaN supplier to high reliability segments such as AI datacenter.

800 VDC Outlook

What stood out at NVIDIA’s GTC Conference this year is that power is no longer a secondary consideration in AI infrastructure — it has become a primary focus. The move to 800 VDC rack architecture reflects that reality, and it is pulling technologies like GaN into the spotlight much faster than many expected. NVIDIA’s supplier recognition, in that sense, is as much a signal to the ecosystem as it is an award: future AI scaling will depend just as much on power semiconductor innovation as it does on compute.