NVIDIA's lack of an AI PC - Good for Investors?

Current open source LLMs need between 320 GB (for Databricks DBRX 16-bit) down to 100GB (new JAMBA LLM) either VRAM or shared memory between CPU, GPU and NPU (AI accelerator).

Current consumer NVIDIA GPU top out at 24GB (4090), and 40GB A100 are financially out of scope for customer implementing their LLM, trained on their company data on-site.

Now a new market segment shows a response: Intel announced a new AI PC in the last days, therefore we analyze technical …

16-bit a100 accelerator ai accelerator ai pc consumer cpu current customer databricks dbrx good gpu investors jamba llm llms memory npu nvidia nvidia gpu open source open source llms