The frame will upload to a server, for further processing. I understand the Arduino doesn't have enough memory t keep loads of screen. However in 25 (rows) *80 (columns) (=2000 bytes), should be possible, i only need the boot-up screens, not the OS loading logo.
Since i need to plug it into multiple PC's i want to connect it on the back (where the VGA is), powered with USB. The whole point is avoiding to connect a screen, so a camera is not a option.
I think you need to re-consider your project and approach.
Just look looking at your math above, you are way off on memory needs.
80x25 is a text resolution. You don't have access to that.
You have access to the graphic signals (Not really even the pixels - but kind of close) coming out of the VGA port.
Even 80x25 in the old "CGA" that mode would be a pixel resolution of 640x200 (80x25 with 8x8 characters)
I'd be doubtful that any PC uses that mode anymore since the text quality wasn't very good.
VGA mode is 640 by 480.
So even if you are looking at monochrome characters in VGA mode, that is 307200 pixels.
To hold that many monochrome pixels would take 38,400 bytes to grab and hold the video frame.
Then you could process the frame pixel data to try to convert the pixels back to text.
Also consider that there are many different resolutions and formats that could potentially
be used by the BIOS and the subsequent BootProcess.
To me the task seems very difficult if not impossible even when using some serious hardware
since there seems to be no confines on the PC hardware and it is allowed to change
and you are wanting to capture not only a single image but multiple "boot screens"
from multiple PCs.
While it might be theoretically possible to monitor the video signals in real time to pick out
pixel data and process that pixel data back to text in real time, it would take way more CPU horsepower
than is available in an Arduino and still take some amount of memory to hold enough pixels for
multiple scan lines to allow the image processing to determine characters from the pixel data.