TAM - part the third

Well... that didn't take long. Yet more reporting, with data, from NYT.

There is, in the econ biz, an acrnonym TANSTAAFL which stands for There Ain't No Such Thing As A Free Lunch. Origin not known, although some/many/most sources go back to 19^th century taverns - "free lunch" if you spent enough.

The AI, as currently implemented, craze has been operating, if not "free", at least way, way below cost.

This latest report from the NYT explores evolved GPT models, driven by the same request. The models are 3.5, 4.o, and 5.5, the number offered for each is the number of "tokens" to process the request. So, here we go:

3.5 448 at $0.00

4.o 771 at $.01

5.5 6,463 at $.19

The report goes on to explain how each is, to all intensive porpoises, just a black box. There have been some reports that not all systems/models/developers actually know what's going on inside these Black Boxes.

Justin Summerville, an executive at the company, said many businesses "spent the past year yelling at their employees to use A.I., use A.I., use A.I."

But that's come at a cost, he added. "And now the finance organizations are waking up and saying, 'Whoa — this is a big number,'" Mr. Summerville said.

Dontcha just hate the bean counters? Always ruining the party. What was that the Wicked Witch of the West said? "I'm melting!" Just as fast as the AI TAM.

Here's another take, similar but not identical, to address the issue.

Faced with escalating variable costs, enterprises are capping their cloud consumption. The financially viable alternative is shifting to owned server architectures. By deploying open-source AI models on proprietary hardware, organizations bypass restrictive and costly subscriptions entirely.

With a grain of salt, naturally. The report is from a stock analyst with a bone to pick someplace. But, if it happens that there does emerge a significant shift from AI-operator operated models to in-house "open source AI" on owned hardware? More melting for the likes of OpenAI, et al.