Well, I think from the technology side, the performance and capacity you can get in a personal computer (especially a laptop) is absolutely incredible.
It's just component suppl and that supply is being eaten up and re-diverted to data centers. Prices and availability will be in poor shape. Though I am wondering if GPU compute and memory start to diverge enough that AI companies begin using such specialized chips they stop threatening consumer devices. Maybe that's just wishful thinking.
I think you have it backwards. Personal computing was a huge market driver in the 80s and 900 and 2000s.
In the 2010s this became less so with the ramp up of cloud computing, mobile computing, and death of Moore’s law. Now personal computing is a footnote that generally takes the left overs from mobile or server and will continue to get squeezed due to lack of meaningful market demand.
Prices must come down not because AIs switch to accelerators - they still need huge amounts of ram for inference* AND training - but because if RAM isn’t a pricing cartel then supply will increase.
* Technically there’s at least one company I know of burning models into ASICs but you still need the RAM to store the weights. SRAM is too power and heat heavy but RAM will only get a reprieve if Cerebras pans out and given OpenAI is the company that partnered with them and then cornered the DRAM market it suggests there’s challenges scaling that approach.
Personal computing was essentially dead when companies figured that renting hardware and software and charging monthly subscriptions was a lot more profitable.
It's not dead. I refuse to rent hardware and software. I host all of my stuff at home on my own hardware, and encourage those around me to do the same. I have converted countless e-waste laptops to Linux and will continue to do so. Personal computing is only dead if you accept that outcome.
It's just component suppl and that supply is being eaten up and re-diverted to data centers. Prices and availability will be in poor shape. Though I am wondering if GPU compute and memory start to diverge enough that AI companies begin using such specialized chips they stop threatening consumer devices. Maybe that's just wishful thinking.