JetCool, a subsidiary of Flex, has entered a strategic collaboration with Broadcom to deliver advanced liquid cooling technology for next-generation AI XPUs. The initiative combines JetCool’s direct-to-chip cooling expertise with Flex’s global manufacturing scale and Broadcom’s custom silicon design capabilities. Together, the companies aim to solve one of the most pressing challenges facing hyperscale AI infrastructure: managing escalating thermal loads from increasingly powerful processors.

AI training and inference workloads continue to expand across hyperscale environments, pushing silicon power densities into sustained multi-kilowatt ranges per device. As a result, cooling architecture has become a central determinant of system performance, reliability, and deployment speed. In response, JetCool has engineered a single-phase direct-to-chip liquid cooling solution designed to integrate with Broadcom’s mechanical and thermal reference architecture. The platform enables stable operation of multi-kilowatt ASIC devices operating at heat flux levels reaching 4 W/mm² per chip.

The collaboration reflects a broader shift across AI infrastructure design. Operators increasingly treat thermal systems as core architectural components rather than supporting infrastructure. Consequently, integrating cooling strategy earlier in chip and system development cycles has become essential to maintaining performance at scale.

Engineering AI Platforms for High-Density Deployment

Broadcom and JetCool aligned silicon architecture, advanced packaging, mechanical integration, and thermal engineering early in the development process. This coordinated approach enables AI platforms optimized not only for raw performance but also for repeatable high-volume manufacturing.

“At Broadcom, we design AI systems to lead at hyperscale,” said Ken Kutzler, VP of AI Systems Development for Broadcom’s ASIC Products Division. “Supporting multi-kilowatt ASIC platforms requires a tight coordination across silicon architecture, advanced packaging, power delivery, and thermal engineering. Our partnership with JetCool, combined with Flex’s manufacturing and integration capabilities, provides a clear path from advanced ASIC innovation to high-volume AI XPU deployment.”

By combining JetCool’s direct-to-chip liquid cooling technology with Flex’s production infrastructure, the partnership introduces a thermal platform capable of supporting large-scale AI deployments. Moreover, Flex’s global manufacturing network enables rapid replication of cooling systems required for hyperscale environments.

“Cooling has become a primary design constraint for next-generation AI silicon,” said Bernie Malouin, PhD, VP, Liquid Cooling at Flex. “JetCool’s advanced direct-to-chip cold plate technology, combined with Flex’s global manufacturing scale, enables production-ready thermal solutions engineered for high-density silicon architectures. We’ve developed a strong partnership with the Broadcom team, and together we’re advancing scalable thermal solutions for the next generation of AI infrastructure.”

Building a Production-Ready Thermal Foundation for AI Infrastructure

The collaboration supports a new class of AI systems engineered for sustained multi-kilowatt performance across high-density ASIC platforms. JetCool’s direct-to-chip liquid cooling architecture removes heat at the source through precision cold plates, enabling processors to operate within safe thermal envelopes even under extreme computational loads. At the same time, Flex’s global manufacturing footprint provides the supply chain capacity required for hyperscale rollouts.

Furthermore, the partnership prepares infrastructure for the next wave of AI silicon. Chip power densities continue to rise as accelerators scale in transistor counts and memory bandwidth. Thermal designs that can evolve alongside silicon generations will therefore become critical for long-term infrastructure planning.

The systems supported by the collaboration deliver sustained multi-kilowatt-class performance through advanced direct-to-chip liquid cooling while enabling high-volume manufacturing through Flex’s production ecosystem. They also provide a pathway for supporting future AI silicon generations as power densities climb further in the coming years.

Broadcom maintains an extensive partner ecosystem that supports multiple cooling technologies and deployment models across hyperscale environments. Within that ecosystem, JetCool works closely with Broadcom to advance direct-to-chip thermal architectures designed specifically for high-performance AI platforms.

JetCool delivers a comprehensive liquid cooling stack that spans cold plates, manifolds, and coolant distribution units (CDUs). These systems integrate with Flex’s global manufacturing and integration capabilities, enabling scalable deployment across hyperscale data centers. As AI workloads continue to intensify, collaborations like this increasingly define how next-generation compute infrastructure will balance performance, manufacturability, and thermal resilience.