We have already reported about various developments and trends in the graphics card market in the last few days, among other things that AMD might forgo high-end chips in its next generation and Nvidia might also slow down its GPU development a bit.
The first rumors about Nvidia's upcoming RTX 5000 series graphics card generation now fit in well with the current overall picture in the industry. First, perhaps the most relevant: Nvidia is actually slowing down a bit and leaving its usual two-year release cycle. The new graphics cards will not be available in fall 2024, but in 2025 at the earliest. In return, they will increase the memory interface unusually with the RTX 5090.
After Ada, the next chip generation will be called Blackwell. Some of the specifications of the upcoming models, which are already estimated as very likely by different leaker sources, are also hardening. 
Hardwareluxx has summarized them nicely with some background information.
Nvidia is also supposed to drop a chip class, but not the high-end, but the mid-range. Accordingly, a GB204 version isn't supposed to exist.
The upcoming high-end card, on the other hand, will be powered by the GB202 chip, which will allegedly even connect to the new DDR7 memory via a 512 bit memory interface. According to current knowledge, this would more than double the data transfer rate compared to an RTX 4090 (approx. 2,300 GByte/s).
The entire line-up of the next generation would then look something like this:
RTX 5090 max. 32 GB, 512 bit with approx. 2.300 GByte/s
RTX 5080 max, 24 GB, 320/384 bit with approx. 1.400 - 1.700 GByte/s
RTX 5070 (Ti) max 16 GB, 256 bit with approx. 670 GByte/s
RTX 5060 (Ti) max 12 GB, 160/192 bit, approx. 360-430 GByte/s
Even though these details and specification estimates already seem plausible, they are of course still very speculative to classify.
;)