Ask HN: Future of AI Hardware – GPUs vs. Groq and Beyond Transformers

[c2p2]

Hey HN,

I have a few questions about the future landscape of AI hardware and architectures:

- Can GPU architectures eventually rival the speeds of Groq and similar specialized hardware? Is there a foreseeable future where GPU inference with hardware sub-systems (e.g. TPU) and/or algorithmic enhancements can match or surpass these speeds?

- Is there a post-Transformer AI world? Are there any new architectures or enhancements on the horizon? Which of these could continue the GPU dominance?

- Are approaches like Groq easy to optimize for future AI models?

Insights and thoughts appreciated!