Here is an old Sound on Sound article:
The article is 15 years old, but the basic problem has not changed at all: every D/A (and A/D) converter that you'll need for turning a digital signal to an analog signal (or vice versa) uses a clock called word clock that makes the converter "tick" at a steady rate. The steadiness of this clock pulse (i.e. low audio clock jitter) is very important for sound quality in digital audio systems.
If you have multiple audio interfaces, you'll have multiple clocks timing their D/A (and A/D) converters. Each clock can either be synchronized to an external master work clock, or it can run independently at its own speed. If you have multiple unsynchronized clocks, even if they all produce a nominal 44100 Hz or 48000 Hz rate, they will run at slightly different speeds, meaning that at some point in time one of the clocks has ticked one more time than the other clock. And then another tick. And another. And at some point, let's say, audio interface A has output 1000 more samples than audio interface B. But all of the samples are coming from the same source DAW program (Cubase, Ableton, etc.) - what to do?
Let's consider all the software running on the DAW computer as a single "program", even though it consists of multiple parts, including the operating system, device drivers, and application software. Conceptually, the software can resort to one or more of the following methods to manage the problematic situation:
- Tell interface A when it asks for more samples: "I don't have any more samples for you. You'll have to make up stuff, or output silence, or repeat whatever the last buffer was, or something. Sorry!" Interface A's clock will keep ticking and its D/A converter will keep producing an analog signal, regardless of whether it was given data by the computer or not. --> You'll get artifacts in the sound.
- Tell interface B: "Why are you so slow? I have a whole new buffer of samples ready for you, but you haven't even consumed the previous one yet!? This other interface has been able to consume everything I've given to it, but not you. I'm going to throw the data away!" --> You'll get artifacts in the sound.
- Create a virtual "audio clock speed clutch" which gradually slows down or speeds up audio, essentially pitch-shifting and/or time-stretching audio so that on average, all the different audio interfaces stay roughly within the same time window. --> You'll get artifacts in the sound.
These problems can be avoided by having a master word clock in the system. For example so that audio interface A works as the master and has a word clock output, and interface B is a slave with a word clock input. But not every audio interface has these facilities.
If the word clocks are not synchronized, there will be artifacts. What kind of artifacts, how much and are they a problem - you'll have to try and see. Thanks to @ojs, I found that USB audio devices are supposed to somehow synchronize their word clocks to the 1 ms (1000 Hz) bus clock rate. Maybe this works nicely and without artifacts, with the devices that you have.
After the Sound on Sound article was written, a new type of audio device has emerged, the USB microphone, which means that each microphone has a built-in audio interface, A/D conversion and a _word clock. Not getting total word clock sync chaos relies on USB bus clock sync.
I'm not sure how Apple's aggregate audio device and various other similar systems work, but there has to be some kind of a compromise with audio quality/artifacts and latency. There's a "drift correction" feature in the Apple thingy, which tries to keep the average speed in control. Probably just fine for casual music consumption and many other uses. But for some uses such a system might not be OK. Using an aggregate device for mixing, if there are no word clock input/output facilities, might be on the NOT OK side. You'll have to test if it's good enough for your purposes.
Whether this works well enough for your needs or not, depends on your needs, and on the exact combination of hardware and software and their settings. It might be doable with some combination of products and settings, for some uses.
Anyway, an application program such as Cubase or Ableton, wants to see just one output audio interface, and that's why a virtual "aggregate device" is needed. On the Mac, the operating system's aggregate device is reported to work well for multiple USB devices (see here for an example), but on Windows, something like ASIO4ALL is needed to create such an aggregate device https://www.sweetwater.com/sweetcare/articles/aggregate-audio-devices-and-drivers-for-pc-and-mac/#ASIO4ALL-for-Windows