Arrow's theorem only applies to ranked systems. That said, cardinal systems are still susceptible to strategic voting (Gibbard's theorem)
When people think about strategic voting coming from a FPTP context, the strategy they're thinking of is putting in a dishonest vote for a non-preferred candidate because that provides a better chance at matching their preferences than an honest vote. This is not what Gibbard's theorem is about. Strategic voting in cardinal systems, for example, tends to look basically like dynamic range compression and at an individual level generally only "tiebreaks" between candidates that were otherwise fairly evenly matched, while the FPTP strategy of bullet voting a non-preferred candidate is just straight up bad for your preferences.
This still somewhat exists. For example, with score voting: preferences A>B>C but you really hate C, and he's leading in polls. It might be reasonable to "dishonestly" score B higher. This minimizes chances that C wins, and while it doesn't harm A it does decrease strength of your preference for A>B. "Dynamic range compression" is good way to put it, or maybe "ballot influence conservation": one ballot has fixed influence, and you can either spend it entirely on A/B>C or A>B/C, or any ratio in between.
Personally, this is a major factor in preferring approval over score vote, which requires someone who intuitively "feels", say, a 0.3/0.6/0.7 about candidates to either vote 0/0.75/1 to maximize their vote spread, or give more influence to people who do. Notice that this is not strategic in Gibbon's sense—it is always better to vote 0/0.75/1 than it is to vote 0.3/0.6/0.7—but it's probably more counterintuitive!
I would go further and be somewhat dismissive even of "Gibbard's theorem", since it only shows that an insincere vote can strategically produce a better outcome for a voter if they can identify what that that vote should be, and if they are one of the voters for whom such a strategy is available.
So it doesn't guarantee (as far as I'm aware) that even a single voter (under every possible cardinal voting system) will necessarily have enough information before or during the election to confidently pick some specific insincere vote that will increase their chances of getting the outcome they want.
Gibbard's theorem is even weaker than that. It only assures the constructibility of vote sets for which the right "final" votes to achieve a particular outcome are distinct. It does not guarantee that there are any insincere votes which perform better than a sincere vote!
Optimal strategic approval voting with a strict ordering is always a choice between sincere votes, in fact (since adding a vote for a strictly more preferred candidate never hurts your preferences, you can always find a sincere vote which performs better than an insincere strategic vote by additionally voting for all candidates more preferred than the least preferred candidate on the strategic vote).
