Recently, LightCounting, a market research firm in the optical communication industry, pointed out in a report that DeepSeek's successful case in early 2025 revealed a path to lowering the threshold for using AI models, which will attract more small and medium-sized enterprises and developers to invest in A | application innovation. And all of this cannot be achieved without high-speed interconnected network infrastructure.

LightCounting points out that the global AI competition has ushered in rapid development of intelligent computing data centers, digital transformation of many industries, and a surge in cloud based services, including video based A! Model training and inference. Fiber optic networks and the global telecommunications industry will become an indispensable part of A!'s future, opening up new opportunities for telecommunications service providers and their suppliers.

LightCounting has been continuously predicting the demand for optical connectivity in telecommunications networks and cloud data centers for the past twenty years. Short term (1-2 years) forecasts often appear optimistic due to market fluctuations, but long-term (3-5 years) forecasts tend to be conservative. The demand for bandwidth continues to exceed expectations and may continue in the next five years.

The following figure shows the historical data and forecast of the total annual deployment bandwidth of DWDM connections divided by application. The share of Cloud Data Center Interconnection (DCI) has risen from 30% in 2018 to 55-60% in 2023-2024, and is expected to reach 60-70% in 2025-2029. Cloud vendors have taken the lead in adopting 1.2T/1.6T high-speed ports, and operators are catching up.

Emerging AI applications will drive up the demand for DC1, enterprise network, and telecommunications network bandwidth, supporting the growth of DWDM module sales in the next five years.

The "end edge cloud" collaborative model between AI smartphones and AI PCs will become mainstream: small models handle simple local reasoning (such as search), while large cloud models undertake complex intelligent computing such as video analysis and graphics processing.