There will be a new generation that learns about VPNs and torrents. Life is cyclical.

any actually useful real world tasks you need at least 24GB you can dedicate to the LLM alone, and if this is fully GPU vRAM

Agreed. For anyone who doesn’t have a 24GB card, the AMD 7900 XTX is basically the only sanely priced 24GB GPU left in 2026 and can be had for under a grand new, whereas even a used 3090 is $1200 and $2000 new.

Qwen 3.8 27B runs nicely on my XTX. I have to use Claude at work, and Qwen 3.8 in llama.cpp with the OpenCode desktop app leaves nothing to be desired in comparison. 3.8 27B is about on par with Sonnet 5 according to benchmarks, though I even prefer it over Opus because Qwen writes in clear, understandable language. We are at the point now where anyone with a 24GB GPU has little reason to use data center LLMs.

AI won't really do that, because it's a very different workload.

Mining Bitcoin needs very little memory but extremely complex maths calculations. It has a built-in difficulty mechanism that makes the math more difficult if mining goes too fast.

ASICs are unbeatable when it comes to super-fast pure calculation.

AI work is mostly memory-bound. You need huge amounts of RAM (24GB for a somewhat decent model, a few TB for frontier models) and then you perform pretty simple vector math on that huge amount of data in memory. So the main limiting factors are RAM size and bandwidth. The computation power of a GPU doesn't matter that much. Also GPUs contain ASICs for vector math already.

Standalone ASICs have no advantage over GPUs for this workload. The only thing you could do with a specialty chip is remove some unnecessary parts of a GPU to save a bit of cost and energy, thus turning a GPU that can be repurposed or resold if the AI bubble pops into a single-purpose device that has to be discarded if the bubble pops.

Some do, others just charge for everything or charge for something that's released for free elsewhere.. Cults3d and Yeggi are bad for that from recent memory.