Since the release of ChatGPT in November 2022, companies have invested billions in infrastructure for the training and inference of increasingly complex AI models. Each new generation of GPU offers greater computational power, but requires more electrical power to be delivered. With server rack power expected to exceed 1MW by 2030 in advanced AI "factories", existing data center power architecture simply cannot deliver enough power.
Influenced directly by EV power architecture, Nvidia and tens of partners are together developing solutions for 800VDC (HVDC) data center architecture. Driven by increased power density requirements and enabled by wide bandgap technology, this architecture promises to future-proof AI data centers for years to come.
In this webinar, IDTechEx Technology Analyst Matthew Fall will provide key insights on data center power electronics technologies and players involved in the HVDC data center transition. This webinar will discuss:
- Wide bandgap semiconductors SiC and GaN, and their role in enabling the HVDC transition.
- Incumbent data center architecture, and the challenges posed by AI training and inference.
- 800VDC data centers, and the electric vehicle influences on this power architecture.
- Other challenges AI poses for data centers, including the challenges associated with synchronous training loads