Goldman Sachs: Mạng quang học sẽ thúc đẩy tăng trưởng hạ tầng AI, TAM tăng 9 lần lên 154 tỷ USD

The Huoxing Finance report states that on April 17, 2026, Goldman Sachs’ Global Technology Stock Research Team released a research report indicating that networking connecting multiple chips unlocks the computing power of individual AI chips, enabling seamless data exchange and low latency. With the expansion of AI infrastructure and increased computing power, various configurations are expected to experience growth, further expanding the TAM by 9 times to $154 billion. The report notes that the dollar content increases by 16 times for horizontal scaling and 45 times for vertical scaling; the TAM for optics expands 13 times from horizontal to vertical scaling; the value market for pluggable optical modules in horizontal scaling grows 10 times from AI server models in the second half of 2025 to those in the second half of 2027. Authors include Allen Chang, Verena Jeng, James Schneider, Mark Delaney, among others. The combined dollar content per compute unit for vertical and horizontal scaling is projected to increase from $3.15 million in GB300 NVL72 to $9.4 billion in Rubin Ultra NVL576, a 29-fold increase. The combined TAM value increases 9 times to $154 billion (primarily by 2028), with 69% ($106 billion) coming from vertical scaling. The analysis covers dollar content and market opportunities for configurations including copper cables, pluggable optical modules, CPO, and backplanes in PCBs.