A recent report on AI hardware supply chains reveals a significant shift in focus, moving from a battle over cooling technologies to the complex world of optical testing. This development signals where the real value is being created in the next generation of AI infrastructure.
For some time, the conversation has centered on how to cool increasingly powerful AI chips, like those in NVIDIA's upcoming Vera Rubin platform. The debate was often framed as a competition between two liquid cooling technologies: MCCP (micro-channel cold plate) and MCL (micro-channel lid). However, a key supplier, Hung Ching Precision, has clarified that these technologies will coexist. MCCP, being easier to scale, will likely be mass-produced first for broader use, while the more complex and higher-performing MCL will be reserved for the most demanding, high-end applications. This ends the 'winner-take-all' narrative.
More importantly, the report highlights that the primary profit engine is no longer just the cooling hardware itself. Instead, it's migrating to the testing of Co-Packaged Optics (CPO), a technology that integrates optical and electronic components onto the same package to enable faster, more efficient data transfer. The critical, high-value stage is the final optical-electrical co-test, dubbed 'Insertion-4'. This process is so complex that adding vision inspection systems can increase the average selling price of test equipment by 10–15%.
This market shift didn't happen overnight. It's the result of a clear causal chain. First, NVIDIA's decision to make its Rubin platform entirely liquid-cooled created a massive, guaranteed market for advanced cooling solutions, making both MCCP and MCL viable. Second, and crucially, the technology behind CPO has rapidly matured. Industry groups like the Optical Internetworking Forum (OIF) established critical standards, and successful multi-vendor interoperability demonstrations at major conferences like OFC 2026 proved the ecosystem is ready for mass production.
Ultimately, this means the key to understanding the future of AI hardware investment is no longer about picking the winning cooler. It's about recognizing that the manufacturing and testing of optical interconnects has become the new bottleneck and, therefore, the new center of value creation. The companies that can master the complexity of CPO co-testing are best positioned for growth as we enter this new phase of AI infrastructure.
- Co-Packaged Optics (CPO): A technology that places optical connectivity components very close to electronic chips (like GPUs) within the same package. This reduces power consumption and improves data transfer speeds compared to traditional pluggable optical modules.
- MCCP/MCL: These are both direct-to-chip liquid cooling technologies. MCCP (micro-channel cold plate) is a more established approach, while MCL (micro-channel lid) is a more advanced technique that integrates cooling directly into the chip's lid for even better thermal performance.
- Insertion-4: A term for the final stage in the semiconductor testing process where optical and electrical functionalities of a co-packaged device are tested together before final assembly.