Not investment advice. Not a recommendation to buy/sell. Educational only. Do your own research. Full disclaimer
Node 03 / 06 · educational map
Once you stack enough chips, moving tokens becomes a physics problem.
AI Bottleneck Map
03
Copper runs out of steam between racks. Lasers, transceivers, and optical switches become the fabric of a cluster.
Tickers are educational examples of names often cited in public AI-infra discussions. Not investment advice. Not a recommendation to buy/sell. Educational only. Do your own research.
A large training job is a networking problem wearing a model-architecture costume. Gradient updates and expert routing only work if accelerators can talk to each other fast enough, with enough energy efficiency to stay on.
Inside a rack, copper still carries a lot of short-reach traffic. Between racks and rows, optical transceivers, lasers, fiber, and increasingly co-packaged optics show up in public roadmaps as the way to keep scaling.
Optics has its own supply chain: indium phosphide and silicon photonics, DSP silicon, packaging, and the switch chips that sit in the spine of a data hall. That chain does not magically expand just because GPU allocations improve.
When people say “the network is the new wall,” they are usually pointing at this node — the light, not the model card.
Example public names often cited in AI infrastructure discussions — not a buy list, not a rating.
NYSE
COHR
Coherent
Often cited for lasers, transceivers, and optical components that show up in AI-cluster networking discussions.
NASDAQ
LITE
Lumentum
Often cited alongside other photonics names when people map datacom optics for large clusters.
NASDAQ
AVGO
Broadcom
Often cited for switching silicon and optical-related parts of the AI networking stack — not as a price target.
Example public names often cited in AI infrastructure discussions — not a buy list, not a rating.
Free weekly map update
A short educational note when the map changes. No paid course. No WhatsApp blast. Email only.