> I’m still baffled why larger VRAM cards haven’t come out more competitive. Because that would eat into the market of hyper expensive dedicated AI accelerator

> I’m still baffled why larger VRAM cards haven’t come out more competitive. Because that would eat into the market of hyper expensive dedicated AI accelerator cards. Wasn't it obvious? At TSMC you'll pay 20 per square inch of silicon(number I made up for exemplification), would you rather a) print a GPU on it and sell it to consumers for $800 or b) print an AI accelerator on it and sell it to datacenters for $8000? It's really that simple. What company you know that says no to making more money?