The difference is that with GPUs nearly every model has a different layout of components around the actual chip, and those components need to be cooled too, so making a universal cold plate for them can be really challenging. The AIO I bought for my last computer could be used with any CPU, AMD or Intel, released in the past decade. The water block I bought for the GPU I put in my desktop two computers ago could only be used with reference designs of the AMD R9 290 or 290x.
There are models of GPUs with AIOs, EVGA makes their Hydro series, Gigabyte has their waterforce series. They just usually aren't worth the cost since you have to design a new full cover block not just got every generation but also usually for different cards in the same gen too.
Lastly, TDP is not a defined standard and each manufacturer defines it differently. Intel for example defines TDP as the maximum allowable power draw under full load. How that translates to actual heat output depends on architecture and process node. Intel's most recent chips for example all max out at just over 100 watts TDP, but they're known to need 200+ watts of heat dissipation for sustained workloads.
There are models of GPUs with AIOs, EVGA makes their Hydro series, Gigabyte has their waterforce series. They just usually aren't worth the cost since you have to design a new full cover block not just got every generation but also usually for different cards in the same gen too.
Lastly, TDP is not a defined standard and each manufacturer defines it differently. Intel for example defines TDP as the maximum allowable power draw under full load. How that translates to actual heat output depends on architecture and process node. Intel's most recent chips for example all max out at just over 100 watts TDP, but they're known to need 200+ watts of heat dissipation for sustained workloads.