Epic Microsystems, a semiconductor company developing power delivery systems for artificial intelligence infrastructure, today announced it has raised $21 million in early-stage funding to accelerate the commercialization of its vertical power delivery technology for AI data centers.

Seligman Ventures led the company’s Series A funding round, which brought the total funding for the company to $26 million to date. Intel Capital, AICONIC Ventures and Cambium Capital also participated in the round, alongside existing seed investors A&E Investments, Assam Ventures and Nepenthe Capital.

Current AI workloads are driving not only the construction of new data centers, but also efforts to raise the density and power capacity of existing facilities.

Leading AI racks today can draw more than 100 kilowatts of power, and next-generation designs are being built with a path toward 1 megawatt per rack. At those densities, traditional low-voltage, in-rack power-delivery approaches start to hit practical limits in efficiency, thermal management, copper requirements and ease of maintenance.

“AI data center architectures are breaking historic power boundaries, and existing power delivery ecosystems simply can’t keep up,” said co-founder and Chief Executive Sabin Eftimie. “Our approach reimagines DC-DC power delivery from the ground up.”

By switching out inductor-style systems for what the Epic said is a hybrid switch capacitor architecture approach, AI datacenters can deliver higher density with improved thermal performance.

The company said its new power-delivery component design takes up less vertical space, which makes it easier to pack more compute and cooling into a dense rack. This can provide greater space for airflow or a cooling layout, because lower-profile power stages leave more room for heatsinks, cold plates, manifolds, tubing or service clearances.

AI technology and infrastructure leader Nvidia Corp. made a similar argument at the rack level when it said moving certain power conversion elements out of the rack frees space for more compute and improves cooling efficiency.

Epic believes that by changing the underlying power delivery and conversion technology used by AI data centers, it will help drive the next-generation of power delivery and conversion. By both addressing today’s increasing demand for compute and opening opportunities for the future roadmap.

“By addressing efficiency, thermals and rack densification simultaneously, we give hyperscalers greater architectural flexibility as rack power scales toward the multi-megawatt era,” said Eftimie.

Image: SiliconANGLE/Microsoft Designer