by 2030 with the use of optical accelerator's super computers could reach 5 - 10 exa flops and low watt gaming devices might hit over 10 terra flops.
by 2035 (early pentium level IC fabrication may be done at home and 32nm tech might be cheap enough for low cost small batch runs So you could make your own amiga cheap enough) this might be more like 50 exa flop systems and 100 terra flop low watt computing once the accelerators have reached the low end. by this same time a micro sd might have reached 128TB and £2000 1 petabyte drives might be dime a dozen putting the price of exa scale computing at a few 10's of millions.
by 2050 (7nm tech fabricatable at home and 0.8nm for cheap small batches) with further acceleration they may have hit zeta scale super computing with some quantum acceleration perks and sd cards might be 4- 8 PB with £2000 256 PB sugar cube drives dime a dozen, exascale computing would be below £20,000 py and even a low watt system would operate at 10's of peta flops enough for full general intelligence piggybacking off masses of trained data on the zeta scale cloud service. what takes a day on a super computer today would take less than a minute what took years now takes less than a day. a 10 terra flop roll out tablet may only cost £5 because of it only being 2d level computing and very basic for 2050.
At zeta scale efficiency our bots would be intelligent and clear and able to invent, our holographic devices would be crystal clear and even a 4 year old child could work with such level of AI and over a few weeks produce a cinematic level full blown open world game. So movies will be made more by writers, big games again will be made by small teams of people over weeks and all that sort of content will be swamping people a lot more.
By 2050 our natural world optimising and climate change modelling may be very precise and allow for a lot more efficiency in how we work with the world and AI scientific and mathematical analysis may be through the roof allowing for far more functional developable and precision products technologies and services.
By 2065 (0.8 nm home fabrication 0.1 nm 5 layer optical cheap small batch runs) exa scale computing may have got below the £1000 and be more of a low watt option. we may have super computers running at 100's of zeta flops, A 16 exa byte sugar cube ssd might be £2000. even low end AR systems can operate upto near an exaflop..
What takes years on a super computer today now take less that 30 mins and what took days before now takes but a few seconds. So you could train up the nvidia dlss 3.0 dataset in less than an hour on one of these supercomputers, in just a few hours you could train a bot to drive cars near perfect. in a few days of training your bot could predict the next 15 - 20 days of weather near perfectly. with a few years worth of this level of computing you could develop a bot that estimated very accurately the history of the solar system and earth relative to whats left to deduce probably revealing countless revelations.