ChainCatcher said Goldman Sachs’ Global Technology Equity Research team put out a report on April 17, 2026. The takeaway: AI networks can link multiple chips together and get computing power on par with a single AI chip, while still allowing data to move back and forth with low latency. And as AI infrastructure gets bigger and computing power climbs, different configurations are set to grow, pushing the total addressable market (TAM) up 9 times to $154 billion.
Dollar Content Soars in Horizontal and Vertical Scaling
The report says dollar content in horizontal scaling and vertical scaling jumped 16x and 45x, respectively. Optical TAM grew 13x in the move from horizontal scaling to vertical scaling. And the market for pluggable optical modules in horizontal scaling rose 10x, comparing AI server models from the second half of 2025 with those in the second half of 2027.
Per-Unit Dollar Content Skyrockets
Combined dollar content per compute unit across horizontal and vertical scaling is projected to surge from $315,000 for GB300 NVL72 to $9.4 billion for Rubin Ultra NVL576. That is a 29x jump. The total value TAM climbed 9x to $154 billion, with most of that showing up in 2028, and 69% ($106 billion) tied to vertical scaling.
The analysis looks at dollar content and market openings across configurations such as copper cables, pluggable optical modules, co-packaged optics (CPO), and PCB backplanes. The report was written by Allen Chang, Verena Jeng, James Schneider, and Mark Delaney.