How do you calculate the value of LLM output in Marx's terms?

I mean either the value produced (v + s) or the value of the output (c + v) + s using the variables:

c = constant capital v = variable capital s = surplus value

If the value of a commodity reflects the quantity of labor it took to make it (including the wear and tear of the, in this case extremely expensive, means of production), how do you calculate the product you get (the response or output) from using an LLM? There is no labor involved in generating the answer, but a lot of labor involved in the infrastructure, from mining all the way to plugging in GPUs in a data center. The "money" that the whole operation is built with is just IOUs from the least reputable people you can imagine, but regardless comes with interest to be paid with money people got from working (not sure if this affects the value).

Is constant capital in the billions (data centers), variable capital zero, and surplus value zero? Is surplus value the depreciation of the data center and GPUs? How do you calculate the value of all the training data like pirated books, the entire StackOverflow, and Reddit threads?

Currently you have to buy tokens and a token is used up by inputting (and the machine outputting) about 3/4 of a word, so 1M tokens is like 750,000 words and costs about $50. Assuming one question to the LLM is made up of 20 words and yields an answer of 230 words, that's 250 words for the whole process, and you can do that whole process 3,000 times with your $50, or 60 times for $1. In other words each question and answer costs $0.0167.

Is that a serious price?