AI and fiber optics
I enjoyed this blog post by Om Malik about how AI models will become invisible plumbing. Nice history lesson with fiber-optic cables too:
A single strand of glass can carry only so much data at one wavelength β think of it as one lane of a highway. Wavelength Division Multiplexing, or WDM, was the insight that you could send multiple signals down that same strand simultaneously, each on a different color of light, the way a prism splits white light into its spectrum. Each color carries its own independent stream of data. One fiber becomes many.
I agree with his conclusion about AI, although not exactly with his point on open-weight models. Om writes:
Commoditization is already underway. Open-weight models are compressing the advantage that closed frontier models once held. The cost of inference has fallen so fast that capability is no longer a defensible edge.
The difference between “good enough” and “great” is still pretty noticeable. In my current work with local AI models, I’m using Gemma 4, a 26-billion parameter model. It needs something like a 24 GB Mac to run. The frontier models are orders of magnitude larger, likely with hundreds of billions or trillions of parameters.
While you could have an open-weight model that big, how would you run it? At that scale, it is cheaper to pay OpenAI and Anthropic for tokens. Let them manage Nvidia racks with multiple TBs of memory.
For the more important theme of Om’s post, yes, AI will become the fabric of many products. I hope that eventually we’ll even get to the point where we don’t need to clutter up our UI with “AI” labels everywhere. A feature that wasn’t possible before will just exist. (For now, though, AI is so divisive that it’s useful to denote it clearly, so it can be embraced or avoided.)
