As the power density of North American AI data center racks soars from 10 kilowatts to 30 kilowatts and beyond, air cooling has hit the ceiling of physics. Liquid cooling, whether direct-to-chip cold plates or full immersion solutions, is transitioning from the experimental plots of a few innovators to a mainstream deployment option for the industry. Yet, in discussions about this cooling revolution, one collectively overlooked link has remained lurking beneath the surface: Are the optical modules plugged into switch ports truly prepared for liquid environments?
On the surface, this question might not seem worth fussing over. Optical modules are sealed metal devices inserted into switch cages. If the switch itself is an air-cooled version, the optical modules continue to enjoy air cooling; if the switch is retrofitted with cold-plate liquid cooling, the modules seemingly only contact the airflow path on the switch panel, with no direct liquid contact. But the real challenge arises in full immersion liquid cooling scenarios. When an entire switch is submerged in a dielectric coolant liquid, every component of the optical module—its housing, optical interface, internal fill gases, and even the fiber connector—comes into direct material contact with the liquid. This contact does not cause immediate failure, but a slow, difficult-to-diagnose process of material degradation.
The first issue to face is material compatibility. Insulating coolant liquids are typically synthetic hydrocarbons or fluorinated fluids with good chemical inertness, but prolonged immersion can still cause swelling effects in certain polymers. The epoxy resins used to secure optical components inside the module, as well as the sealing rings and dust cap materials of the external fiber connector, if not selected based on compatibility testing with the specific medium, will undergo minute dimensional changes after thousands of hours of immersion. A few percentage points of swelling in a sealing ring might seem negligible, but for the micron-level end-face positioning accuracy of single-mode fiber connectors, it is enough to introduce additional insertion loss fluctuations. From the outset of its development, HaloWill's AquaCore series made material compatibility its primary design constraint. We have established joint testing protocols with major North American coolant suppliers, subjecting every polymer, elastomer, and adhesive in the module with potential liquid contact to long-term immersion accelerated aging tests. Based on this, we reconstructed the bill of materials to ensure stable material performance in any recommended cooling medium.
The second, more insidious problem occurs at the optical interface. The optical port of a traditional module operates in air, with an air gap formed between the fiber end-face and the adapter inside the module, where the refractive index difference is definite and stable. But when the module is immersed in liquid, the liquid can slowly penetrate the tiny gaps of the fiber adapter, changing the refractive index of the medium between the end-faces and thus affecting the return loss. For DDM functions that rely on reflection for monitoring, liquid ingress can cause a drift in the baseline value of optical power monitoring, triggering false alarms. HaloWill's AquaCore module adds an optical window sealing structure with a special lyophobic treatment at the front end of the fiber adapter. This structure does not entirely prevent liquid from entering every crevice of the adapter, but it ensures that along the actual path of the optical signal, the medium between the end-faces remains a controlled microenvironment, undisturbed by external liquid.
The redesign of the thermal management path is equally indispensable. Although immersion liquid cooling has a higher overall heat transfer efficiency than air, the heat generated inside the optical module still needs to be conducted through the housing into the liquid. Traditional modules rely on air convection over the metal housing surface to carry away heat, whereas in liquid, the heat transfer mechanism becomes natural or forced convection between the liquid and the housing. HaloWill's AquaCore series features a housing geometry optimized through fluid dynamics. While maintaining standard QSFP-DD form factor compatibility, it incorporates a surface texture that promotes micro-convection between the liquid and the ports. Concurrently, we have adopted more direct multi-point thermal conduction paths inside the housing to distribute the heat generated by the laser and DSP more evenly to the housing surface, preventing localized hot spots that could cause film boiling or thermal decomposition of the liquid.
For North American AI cluster operators, the value of the AquaCore series lies in eliminating the last "non-liquid-cooling-compatible" worry in an all-liquid-cooled solution. In the past, architects planning immersion liquid cooling racks often had to compromise on switch selection or reserve a localized air-cooled zone to accommodate the optical module connection layer. Such hybrid cooling schemes added complexity to the cooling infrastructure and diminished the inherent high-density advantage of liquid cooling. But when the optical module itself possesses the capability to operate stably in liquid over the long term, the cooling architecture of the entire rack can achieve complete unification. HaloWill is willing to provide every North American customer evaluating liquid cooling solutions with the media compatibility data package and long-term immersion test reports for AquaCore modules, because we believe that in a world surrounded by liquid, the survivability of an optical module is the physical boundary of the confidence you deliver to your customers.
