The rapid advancements in machine learning—particularly large language models (LLMs)—are driving an exponential increase in compute demands, with requirements growing by roughly an order of magnitude every 18 months. Traditional gains in silicon scaling are no longer sufficient, prompting a shift from general-purpose computing to accelerated computing and massive parallelism. This evolution amplifies the critical role of interconnect bandwidth, now a primary system bottleneck. Co-packaged optics (CPO) offer a transformative path forward, delivering scalable, high-bandwidth, and energy-efficient interconnects for next-generation workloads. Realizing CPO’s full potential will require advances in 2.5D and 3D integration to achieve new levels of performance and efficiency. In this talk, we will examine the state of current CPO solutions and explore the integration and scaling options that can unlock their full promise.