Discussion about this post

User's avatar
Pascal Calarco's avatar

An interesting piece. I think there are likely other forces at work that contribute to the inflationary spiral we're seeing in the price of GPUs. Before AI, GPUs were known to be particularly effective at Bitcoin mining around the time of the pandemic, which caused a run on these components. It is absolutely true that Taiwanese semiconductor firms are using their monopoly to focus on producing high-value enterprise-grade GPU for the cabal of AI firms in the market, which drives relative scarcity and high prices amongst consumer models. NVDIA's software stack has been significantly better than its competitors, allowing that company to drive up their asking prices for enterprise and consumer lines. The US war in Iran has offline 25% of global helium production which is a key component in semiconductors. The open source, open weights local LLM models are still very new. Google released Gemma under the very permissive Apache license, which seems counterintuitive if you want to limit choice to only frontier models.The Chinese models are meant to disrupt Western dominance in this space, and frankly, I'd say they are welcome in diversifying local LLM options. I think we're probably just at the start of seeing more open source models, which broadens choice: use a frontier model for heavy lifting inference and lower, lightweight local models for more routine agentic work. The continued development of more open toolsets and models will continue to democratize approaches to AI for the foreseeable future, from my perspective.

Wiggle's avatar

I look forward to seeing how this progresses! Very interesting read!

1 more comment...

No posts

Ready for more?