DC Power solution for AI/Data center and future perspectives

DC Power solution for AI/Data center and future perspectives

🎙 Dr. Yi Jen Chan 👥 5K 📅 February 24, 2026 ⏱ 26 min 👁 76 📄 expert opinion 🧭 2026-08-16
Available in: English (current) Français

Keywords

DC powerAI data centerpower densityGaNSiC

Summary

Dr. Yi Jen Chan from Cyntec presents a talk on DC power solutions for AI data centers, focusing on the increasing power demands of AI GPUs and the need for high power density. He outlines the evolution from AC to DC power architectures, highlighting the shift to high-voltage DC (HVDC) with solid-state transformers and modular DC-DC converters. He discusses the role of wide-bandgap semiconductors like GaN and SiC in improving efficiency and power density. He showcases Cyntec’s products, including 400V to 48V converters with 98.5% efficiency and 2.4 kW/in³ power density, and 48V to 12V IBC modules. He also covers the trend towards vertical power delivery and integrated voltage regulators closer to the chips. He predicts explosive growth for GaN due to its ability to operate at high frequencies, while SiC will see steady growth. He concludes with a note on the importance of the AI market and encourages young engineers.

151 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the current state and future directions of DC power delivery for AI data centers. The speaker supports his points with concrete data, such as GPU power increases from 1 kW to over 1 MW per rack, and specific product specifications like 98.5% efficiency and 2.4 kW/in³ power density. He also references industry trends like vertical power delivery and integrated voltage regulators. The argumentation is logical, moving from problem (increasing power demand) to solution (HVDC architecture, modular converters, wide-bandgap devices) to future outlook. However, some claims are speculative, such as the explosive growth of GaN, and lack detailed evidence.

Scientific Rigor, Source Quality, Title Accuracy

The speaker is a recognized expert, and the talk is based on industry data and product developments. He references specific companies like Nvidia, TSMC, and Infineon, and mentions industry trends. However, he does not provide formal citations or references to academic literature. The title accurately reflects the content, focusing on DC power solutions and future perspectives. The talk is a conference presentation, so the rigor is appropriate for that format.

189 words

Title / Content Match

The title accurately reflects the content, which focuses on DC power solutions for AI data centers and future trends.

Quality & Reliability

7/10

The speaker is a domain expert with 40 years of experience, and the content is based on industry data and product developments. However, some claims are forward-looking and not peer-reviewed, and the presentation is a conference talk rather than a formal study.

Key Moments

Cited Sources

  • SEMICON Taiwan 2025 — Conference where the talk was presented
  • Cyntec Co., Ltd — Speaker's company, manufacturer of power modules
  • Nvidia — Mentioned as example of GPU power demand
  • TSMC — Mentioned for integrated voltage regulators and GaN business
  • Infineon — Mentioned for last-stage power conversion products

Concurring Sources

  • SEMICON Taiwan 2025 — Conference where the talk was presented

Contribution & Novelties

The talk provides an industry perspective on DC power solutions for AI data centers, highlighting the shift to HVDC architectures and the importance of power density. It offers specific product examples and performance metrics, and discusses the role of GaN and SiC in future power delivery. The speaker’s insights on vertical power delivery and high-frequency operation are forward-looking.

Pour aller plus loin :

90 words

Radar Profile

The radar profile shows high scores in technical level and information quantity, indicating a detailed and specialized talk. The quality and reliability scores are moderate, reflecting the expert opinion nature and lack of formal citations.

Reliability 7/10