Horsepower Does Not Matter. Performance Matters.
I am not so impressed by the cat and mouse game played by OpenAI, Anthropic, Google, xAI, Microsoft and Nvidia as it relates to the latest and greatest chipset, LLM, and general compute capability.
Inference, speed and accuracy is the name of the game, not pure compute / horsepower, especially as cost becomes an increasing factor.
Companies that are creative in deploying small language models toward focused purposes will do extremely well in the Gen AI space. That is to say they will drive a meaningful ROIC from their Gen AI project, assuming the use case is valid.
Companies that are creative in deploying chip arrays to drive optimal inference without having to shell out for the latest and greatest chip set will do extremely well in the Gen AI space (Groq comes to mind as a company that can make such outcomes possible), again assuming that the use case is valid.
Speed, accuracy and efficiency matter. Performance matters. Who spent the most money on what chip set does not matter. It makes for a nice headline, but does not mean anything if all of that horsepower does not drive ROIC at scale.