Empower and Marvell Team Up to Tackle AI Power Challenges at the Silicon Level
Joint effort targets 4kW+ processors with high-efficiency, integrated voltage regulation
Empower Semiconductor and Marvell Technology are joining forces to develop advanced integrated power delivery solutions aimed at the rapidly growing demands of AI and cloud computing. The two companies are collaborating to create pre-validated, chip-level power architectures that address the rising current needs of 4-kilowatt processors and beyond—an increasingly common requirement in data center applications.
The collaboration brings together Empower’s strength in integrated voltage regulators (IVRs) and Marvell’s leadership in custom silicon platforms, including their expanding portfolio of XPUs for AI workloads. Together, they aim to simplify power delivery, improve system performance, and reduce energy loss in high-density computing environments.
At the heart of this effort is Empower’s FinFast™ architecture, which enables vertical power delivery and integration of voltage regulation close to—or even inside—the processor package. Unlike traditional board-level power architectures, this silicon-level approach minimizes parasitic losses and offers precise control over power distribution, a critical factor for next-generation AI workloads.
Tim Phillips, CEO of Empower Semiconductor, says this joint effort underscores the importance of localized, efficient power delivery in the AI era. “With integrated voltage regulation, we’re delivering power right at the point of load—with exceptional density, precision, and efficiency,” he said.
Marvell’s Will Chu, senior vice president and general manager of Custom Cloud Solutions, echoed the significance of the integration. “Placing IVRs under, near, or inside the processor package yields substantial gains in both performance and efficiency,” Chu said. “Working with Empower enables us to provide our custom XPU customers with power solutions that are not only pre-validated but also tightly aligned with their silicon and system needs.”
The move reflects a broader trend in the data center industry: as chip power consumption continues to scale upward, power delivery must follow suit—not just in capability, but in physical proximity. Empower and Marvell are betting that tighter integration is the key to unlocking better performance-per-watt in AI and cloud platforms.
For hyperscalers and infrastructure providers, this could mean simpler power architectures, greater energy efficiency, and improved ROI on increasingly compute-heavy deployments.
