5G Systems and Tech ROI Issues

I worked on wireless telecom systems from about 2001 to 2020. Over those years, system capabilities kept expanding as a side effect of increasing capability of CPUs and other computational hardware. All the stuff I worked on was broadcast based so the cost of the system was linearly amortized over each user, so the more users the better in terms of cost per user.

I was surprised to learn, recently, that the computational requirements of 5G networks substantially increase with each user so lots of expensive equipment is required to handle just a handful of users. For example, the ORAN reference open-source software requires around $100K of servers and GPUs to handle just 15 users! A few years ago, those servers and software would probably only cost $30-50K but the AI/datacenter bubble has distorted the price of tech junk dramatically. Even if costs return to normal eventually, tens of thousands of dollars for 10-15 people is stupidly expensive too.

The reference software is probably not optimized, and generic GPUs are inefficient. Presumably a more engineered system could be significantly more efficient maybe by a factor of 10, but the core problem of increased cost per user doesn't go away, it's just not "as bad".

Note that this pattern follows the "LLM" pattern. For example, now when I search for something on google, their LLM spits out a result. The cost per LLM search result is probably significantly higher than the cost per search in their good-old system which probably has that "broadcast" amortized price structure, since the scanning of the Internet, indexing web sites, etc... was done once per many thousand searches. The cost to encode information in the LLM is high, and that's amortized over all the uses of it, that is, each "search" using google's LLM, however, the cost to run the LLM is also high per search. Presumably, google plans to reduce that cost, then charge people monthly subscriptions for their service, which is probably why their search engine has been getting worse. So the "Internet" will effectively turn into a privatized rent seeking scheme. Good luck with that.

Another interesting side-effect of declining tech ROI is actual productive and useful stuff is extremely cheap. Effectively a garage shop scale operation can build almost anything. There's endless examples of this phenomenon on YouTube where a small team of dudes build airplanes, boats, cars, motorcycles, and on and on. It's not hard to imagine an "engineer centric" world emerges from that, where the corporate and finance overhead and parasitism go away.

It's possible the "hit them where they ain't" strategy could work out really well for localities that foster that approach.