Ooh! This is a tough question!
In short, yes inferring a connection is possible even if the IP address and computer change.
Let's dig into it a bit. Any number of clues can be used to infer a connection. With your description, we have already ruled out some of the more obvious clues like static IP, session cookies, web side-channel analysis, cached canaries, etc. Since you specifically stated ASN is in-bounds, let's start there.
The ASN for the IP issued by your ISP may change when you are issued a new IP, but that is unlikely for most ISPs. Regardless of if your ASN changes, the organization will not (unless you switch ISPs). The network size described by your ASN is very relevant. If you have a small ISP with a /23 then the pool of potential humans behind the IP is in the hundreds. A larger ISP with multiple /12 has a pool in the millions.
Reverse DNS could be a factor. Some ISPs issue DHCP leases from IP address pools pre-configured with reverse DNS local to a specific neighbourhood.
GeoIP, as others have mentioned, may be off by a couple cities. But it could be accurate to the neighbourhood. It depends on your IP block and which GeoIP database you look it up in. The free/public databases are rarely as good as certain paid services.
Keyboard layout and typing profiles can be identified by the timing of keypresses when entering your username/password or filling out account registration forms. Timing analysis of keystrokes can even guess what is being typed by measuring the delay between key presses. If the analysis is being performed by the site that you are entering data into, then they already know what keys are being pressed and that makes for an even more solid digital fingerprint of typing style.
So far, we have considered some passive information gathering methods. There are active methods too. OS and browser specific exploits can be used to compromise computer B. From there, options spiral in dizzying directions. Enabling a camera or microphone, scanning the internal network for computer A, or scanning for another device that was also known to computer A, just to name a few. To the potential extreme of mapping out component serial numbers and stumbling across sequential serials between computer A and B. Memory modules, hard drives, mother boards and dozens of other chips inside the average computer all have unique identifiers.
In this hypothetical scenario, the breadcrumbs that a hypothetical investigator may determine conclusive are darn near impossible to predict. But the landscape of potential breadcrumbs in a digital world is vast. To be clear, any active information gathering takes far more effort than is financially viable for a small organization en-mass. There is also significant risk involved. So if you aren't carrying around nation-state secrets or the formula for WD-40, then this is just a thought experiment. And chances are, nobody will guess that you are the same person who already registered a free trial on some "cat picture" website.
Certainty is not required to infer a correlation.