[SK Hynix Optical Interconnect Memory Roadmap: Local HBM Connected to Larger Shared Memory Pools] Odaily Planet Daily News – Citrini analyst Jukan posted on the X platform, stating that at first glance, SK Hynix's CPO technology roadmap seems to have one axis pointing toward HBM and another toward optical communication. However, the underlying logic is that AI competition is shifting from the performance of individual chips to the data transfer efficiency of the entire system. Over the past few years, HBM has addressed the issue of GPUs being unable to receive data quickly enough. By vertically stacking multiple layers of DRAM and placing them next to the GPU, HBM can provide extremely high bandwidth. However, as more HBM stacks are placed around each GPU, packaging area, interposer edge space, power supply, and cooling capacity are nearing their limits. Meanwhile, AI clusters have expanded to thousands or even tens of thousands of GPUs. Regardless of how fast a single GPU computes, if data cannot be efficiently transferred between GPUs and racks, the entire system will still be constrained by the "bandwidth wall." SK Hynix envisions extending optical interconnects to memory. Low-latency, high-bandwidth local HBM will still be placed next to GPUs, while being connected via optical fibers to a larger shared memory pool. This avoids limiting the total memory capacity to a single GPU package and enables horizontal scaling at the rack level. From an investment perspective, this does not mean HBM will be replaced in the near term. Instead, a new memory hierarchy is more likely to emerge: the most frequently accessed data will still reside in local HBM, while larger-scale, less frequently accessed data will be stored in optical interconnect memory pools, HBF, or SSDs. SK Hynix is repositioning its business from selling standardized memory chips to co-designing HBM, controllers, advanced packaging, and system-level memory architectures with customers. If this roadmap is realized, SK Hynix could enhance customer stickiness, increase product value-add, and improve its ability to secure long-term contracts. At the same time, the company could proactively address the potential impact of future memory disaggregation on the traditional HBM business model. In terms of the supply chain, areas that could benefit in the long term include silicon photonics chips, optical engines, lasers, fiber coupling technologies, and advanced 2.5D and 3D packaging. However, this concept is still in a very early stage and is essentially just a roadmap at this point. Jukan noted that a person he interviewed today, who is involved in TSMC's packaging business, was completely unaware of this plan.