The AI boom is running into a copper problem (www.businessinsider.com)

Silicon Valley's AI infrastructure faces a significant challenge: traditional copper wiring is struggling to keep up with the increasing demands of AI systems, especially as they scale. Copper connections have limitations in distance and energy efficiency, leading to overheating and reduced performance as more GPUs are interconnected. Lightmatter, an innovative startup, is shifting the focus to photonics—using light signals for data transmission. This approach not only allows data to move faster and over greater distances but also reduces energy consumption, enabling AI models to be trained more efficiently. Lightmatter's recent advancements, including their entry into Nvidia's NVLink Fusion ecosystem, highlight the urgency for the AI community to adopt photonics. CEO Nick Harris emphasizes that this shift could effectively triple performance for AI workloads, drastically improving training times for complex models like Claude. The introduction of technologies such as BiDi, which reduces the cabling needed for data centers by allowing bidirectional communication over a single cable, could simplify the design and maintenance of AI infrastructures while cutting costs. This evolution in data transmission technology is no longer just a novelty; it’s becoming a necessity for maintaining competitive advantage in the rapidly evolving AI landscape.