According to a Goldman Sachs report, optical networking is emerging as the next major growth area in artificial intelligence (AI) infrastructure, as rising AI computing demand drives the need for faster data exchange, lower latency and larger graphics processing unit (GPU) clusters.
As per the report, networking is becoming a critical enabler of computing capability for AI chips, allowing multiple chips to be connected and to communicate seamlessly with one another.
Goldman Sachs estimated that the total addressable market (TAM) across scale-up and scale-out networking configurations could grow nearly nine times to $154 billion by 2028, from around $15 billion in 2026, driven by the rapid expansion of AI infrastructure and increasing computing power per rack.
Scale-up networking, which involves adding more GPUs and computing resources within the same server rack or across connected racks, is expected to account for approximately 69 per cent of the total TAM, equivalent to nearly $106 billion. The report also projected a 13 times larger addressable market for optical modules and optical engines as networking extends from scale-out to scale-up systems.
Co-packaged optics (CPO), an advanced hardware architecture that integrates optical transceivers and electronic processing chips on the same silicon substrate, is expected to play a significant role in future AI networking systems. Goldman Sachs estimated that CPO could represent around $91 billion, or 59 per cent of the total $154 billion market opportunity, assuming a 29 per cent penetration rate in scale-out networking.
The report also identified three broad methods through which AI infrastructure is being expanded. The first is scale-up, where more GPUs are connected within the same equipment or across linked racks to form supernodes with optimised networking speeds. The second is scale-out, which connects multiple pieces of equipment through switching technologies, with modern AI clusters now capable of supporting connections across more than 100,000 GPUs. The third is scale-across, where servers located in different data centres are linked together.
Additionally, Goldman Sachs noted that Nvidia has introduced scale-across networking solutions using its own ethernet switch and network interface controller products.
