Because it's not just a chip, it's everything else too. There were 2 startups in Boston, may be 500 meters apart, doing optical computing. 4 years and there is still no product. "everything else" is a lot. It's hardware and software. Just brilliant idea is not enough.

They are delivering Groq cards, nodes and rack. Not vaporware.

See an article from the press with details here: https://www.theregister.com/2020/09/29/groq_ai_dev_kit/

this is more recent and optimistic https://www.nextplatform.com/2023/11/27/groq-says-it-can-dep...

But I don't see any selling numbers or independent performance conformations.

> independent performance conformations.

Groq has just been added to the LLMPerf Leaderboard:

https://github.com/ray-project/llmperf-leaderboard#output-to...

(Disclosure: I work for Groq)

That's why I provided the demo link, see the tokens/second. They are running LLaMA 70B at about 260T/s without quantization. This is the fastest LLaMA2 model