We are in the midst of a significant inflection point in the growth of optical networks. According to lan Redpath, Research Dir., SP Networks at OMDIA, “Al is catalyzing a generational optical-network build cycle. The largest cloud service providers are constructing gigawatt Al training centers, and CSPs are responding with connectivity at Pbps scale.”
Generative AI is driving foundational demand for bandwidth, and that demand will only grow as applications evolve, requiring ever-faster connectivity between GPU resources and enterprise applications.
Here’s how bandwidth requirements scale as AI capabilities progress:
- Words to words: Tools like ChatGPT summarize documents or draft emails. MS Office Copilot handles simple responses. This requires relatively low bandwidth.
- Pixels to words: AI analyzes images in real time, delivering intelligent insights and inferences.
- Words to video: Remember the Hollywood strikes over AI job displacement? We’re seeing why. Creators can now type commands into software and generate full videos, complete with realistic, never-before-seen actors. This process demands massive GPU power and bandwidth.
- Words to proteins: AI can now generate proteins that didn’t previously exist, potentially unlocking cures for diseases. Imagine early cancer detection from a blood test, then using AI to design a treatment optimized for that individual’s unique needs. This demands a significant amount of bandwidth.
As AI progresses up this chain, it’s also evolving from a passive tool to an active agent, not just assisting humans, but autonomously analyzing, creating, and taking action based on real-time data. The more agentic AI becomes, the more it depends on high-performance networks that can support real-time, machine-speed decisions across distributed systems.
Perspective provided by:
Kevin J. Sheehan, CTO of the Americas, VP of Sales Engineering and S3, Ciena
Richard A. Heptinstall, Director, Business Development, Ciena