Mar. 26 at 4:04 PM
$LRCX $AMAT Spot on. Google's TurboQuant drop yesterday March 25 is the exact catalyst here, it's a training-free KV-cache compression algo that slashes memory footprint by ~6x it's like down to 3 bits with zero accuracy loss on models like Gemma/Mistral, and delivers up to 8x faster attention-logit compute on H100s. Efficiency breakthroughs like this on AI don't kill semiconductor demand they fuckingg accelerate it.
