POPA ReconMusk Interview, AI bottleneck memory source
Musk Interview, AI bottleneck memory source 1. Key to what Musk said - Only a small percentage of new fabs are needed over the next five years compared to memory requirements - US does not currently have high-performance memory fabs capable of mass production - Micron's fab is also mentioned as around 2028~30 until mass production 2. What the market needs to see is the AI computer - AI infrastructure requires a combination of [logic + memory + packaging] rather than GPUs - The real bottleneck is HBM, DRAM, NAND, packaging that can plug in GPUs - So $MU, $DRAM, $SNDK, $TSM becomes more impor…
The excerpt is quoted from a public Telegram channel, which owns its text.