I did not know Thom so I can't speak to all his personal motivations. But I can speak as someone that has read much of his work carefully and I think I have some insights into this.
The primary motivation for his theorem boils down to thinking carefully about the implicit function theorem, and asking when a subset of Euclidean space is the pre-image of a regular value of a smooth function.
These are the problems most students grapple with when learning about smooth manifolds for the first time, and Thom's considerations flowed out of these kinds of naive considerations. Thom just went a little further than most.
The first step is realizing a submanifold of $S^n$ is the pre-image of a regular value of a smooth function
$$f : S^n \to S^m$$
if and only if it has no boundary, is compact and has a trivial tubular neighbourhood. And this is the key insight into the Pontriagin construction.
The next step is asking about general manifolds, and this is where you take the step up to maps to Thom spaces. Specifically the structure of the normal bundle is key in the Pontriagin construction. So if you have a non-trivial normal bundle you need some feature of your map to take that into account. Grassman manifolds are the key object related to maps out of vector bundles. But once you see that Grassman manifolds are the key analogue to the Pontriagin construction that gives you your map defined in a tubular neighbourhood of your submanifold. Coning off as in the Pontriagin construction gives you the Thom space.
From this point of view you can see his theorem as a direct extrapolation from the trivial tubular neighbourhood case. I think for people that like to think categorically it could maybe feel unintuitive since the Thom space is a departure from manifolds.
As has been mentioned, spectra came after the fact, as the idea was germinating around that time. In Thom's paper he largely phrased things in the language of stable homotopy groups.