It will still be mineable, it just will be in minuscule amount gotten from transaction fees.

Yeah, I think that was their plan, to make it based on the fees, which you can already bump to prioritise your transaction getting picked up from the mempool.

But the idea is with scarcity, the price increases. If it costs $100,000 to mine 1 BTC, you better hope 1 BTC is worth more than $100,000, or nobody will do it. But if they don't do it, the hash difficulty drops, so it costs less to mine. It's a weird system in constant tension.

If someone could invent a "proof of useful work", it would be amazing, and none of the energy would be "wasted', but unfortunately, the work has to be hard to do, but easy to verify.

If you have to fold the protein a second time to prove the first one was right, it's no good with this system. Maybe if it was consensus-based and multiple people had to get the same result, but it would be so easy to just lie about what result you got and do a 51% attack.

I've been working in crypto full time since 2019. 

I used to work at one of the largest bitcoin mining companies in North America (funilly enough, they transitioning solely to AI compute over the last 6 or so months) and now I work at one of the largest proof of stake infrastructure companies in the world. 

I've held through the highs and lows, multiple times.

In my free time, I collect NFTs and spend too much time in those communities. 

Crypto/web3 is pretty much my 24/7.

Yeah, I heard that is happening a lot. There's money to be made atm, but I feel like a lot of the LLM inferencing is going to become local in future, like how thin clients and mainframes were replaced with personal computers.

I tried Gemma4 and Quen3 on a Mac Mini M4 and an RTX 3050 and it wasn't terrible, but it wasn't great either  I regret not getting the 24GB unified memory upgrade, who knew it was going to 10x in price.