Alibaba Network Proposal OSFP MSA Passes Unanimously, Introducing the First Liquid‑Cooled OSFP Cage Standard

Alibaba Cloud’s infrastructure network team’s split‑type OSFP Cage proposal was unanimously approved by the OSFP MSA committee in April 2025 and incorporated into OSFP MSA Rev 5.2, becoming the industry’s first standard that supports liquid‑cooled OSFP cold plates.

The split‑type Cage offers low cost, easy assembly, easy maintenance, and a decoupled supply‑chain architecture, enabling it to solve the liquid‑cooling challenges of high‑density AI switches whose optical module power consumption has grown more than twenty‑fold, surpassing the cooling capacity of traditional air‑cooling.

MSA (Multi‑Source Agreement) is a global consortium that defines standardized interfaces and specifications for optical modules such as SFP, SFP+, XFP, QSFP, and OSFP. Only proposals that receive unanimous agreement from worldwide users and vendors are adopted and published.

The industry faces a critical problem: optical‑module cold‑plate liquid cooling. As switch bandwidth increases, module power consumption exceeds 1.6 T–3.2 T OSFP thresholds, making air cooling insufficient and demanding liquid‑cooling solutions, especially for AI switches with large radix designs.

Alibaba’s split‑type OSFP Cage decouples the cold‑plate supplier from the cage supplier, allowing a single cold plate to serve multiple underlying modules, reducing cost and simplifying assembly. It supports both fly‑over and non‑fly‑over designs to meet diverse deployment scenarios.

In practice, the 51.2 T liquid‑cooled AI switch “Liquid Tigatron” deployed in 2024 uses the split‑type Cage. Its liquid‑cooling cost is comparable to air‑cooling, while delivering energy‑saving benefits; each deployed unit frees space for an additional GPU, boosting cluster compute capacity.

Liquid cooling offers energy‑saving and emission‑reduction benefits that alleviate power bottlenecks in AI workloads, but it presents greater reliability challenges than air cooling, requiring early large‑scale deployment and industry‑wide ecosystem preparation to fully embrace the forthcoming liquid‑cooling era.