Add on to my comments earlier:

I see the unveiling of Google’s KV cache compression technology (TurboQuant) as a short-term noise and not a long-term threat. Reasons being:

🎯 Scope is narrow – Inference only and not training or HBM

📉 Real-world impact overstated – Gains are relative to older 32-bit models

📈 Jevons Paradox in play – Lower costs → more AI adoption → more total demand

Conclusion , AI-driven memory up-cycle remains intact. Some view the dip as a potential entry point.