Competition to design AI Chips

 From The Economist, Can Nvidia be dethroned? Meet the startups vying for its crown

Nvidia’s market value is more than $2trn, with year-on-year revenue growth of more than 200%. Here is wny it succeeded:

...GPU's do the computational heavy-lifting needed to train and operate large ai models. Yet, oddly, this is not what they were designed for. The acronym stands for “graphics processing unit”, because such chips were originally designed to process video-game graphics. It turned out that, fortuitously for Nvidia, they could be repurposed for AI workloads.

But now several startups are designing chips specifically designed for AI:

Cerebras’s response is to put 900,000 cores, plus lots of memory, onto a single, enormous chip, t...On-chip connections between cores operate hundreds of times faster than connections between separate GPUs, ... [and] reduces energy consumption by more than half...

Groq's [chips] contain their own memory, [and] act as routers, passing data among the interconnected LPUs (language processing units). Clever routing software ... greatly boosts efficiency, and thus speed: Groq says its LPU can run big LLMs (large language models) ten times faster than existing systems.

MatX, also based in California. GPUs contain features and circuitry that provide flexibility for graphics, but are not needed for LLMs, ...[their] gpu-like chip gets rid of such unnecessary cruft [badly designed, unnecessarily complicated, or unwanted code], boosting performance by doing fewer things better.

Bottom line:  How many economists does it take to change a lightbulb?   None, the market will do it.