AI training clusters and hyperscale workloads are hitting a physical wall at the I/O interface. Panel real estate has become a bottleneck, limiting how many lanes can exit a 1RU chassis, and air-cooled architectures cannot dissipate the heat generated by 450W+ modules. To support the new low-latency fabrics required for AI, designs need to combine higher lane density and power delivery with liquid cooling at the module level.
XPO Multi-Source Agreement (MSA)
As a founding member of the XPO MSA, Molex is collaborating with industry leaders to push the limits of performance, density, power and thermal management in cloud and hyperscale data centers. XPO is a new standard built to address the bottlenecks of the highest data throughput systems.
Key technical advantages of today’s XPO interface include:
- Massive Throughput: Supports 64 lanes of 224G PAM-4 per module, with a roadmap to 448G
- Direct Liquid Cooling: Features a blind-mate liquid plug/receptacle for direct-to-module cooling supporting multiple ASICs in the XPO module
- Enhanced Power Delivery: Uses a 40 to 60V input range to support up to 480W per module
- Mechanical Density: Thinner shell and reduced vertical pitch enabling two rows of modules in a single 1RU space, doubling front-panel density
- Simplified Integration: Separate housings for high-speed signals and power, along with independent I2C/I3C channels
A Future-Proof Ecosystem
The XPO MSA ensures that the shift to next-generation speeds is backed by a multi-vendor, interoperable ecosystem that reduces supply chain risk, provides a predictable design path for system engineers and ensures interoperability between manufacturers.
This collaborative framework gives hyperscalers the confidence to deploy next-generation liquid-cooled architectures at scale, knowing they can rely on a common, high-performance form factor.
See It Yourself
Molex will demonstrate live data transfer using the next-generation XPO interconnect in both the Molex and Arista booths at OFC 2026.
Share