In the case of metric moderator variables, the product of two variables is used to represent the interaction effect. I try to choose an intuitive example for your case:

“The effect of X1 (categorial independent variable) on Y is negative” – so let’s say X1 is a shot of Gin and Y is the value of the beverage.

“The effect of X2 (moderator) on Y is negative” – so we choose X2 as the amount of tonic water.

“The interaction term of X1 and X2 on Y is positive” – I think you refer to the beta coefficient of the interaction.

So in a more formal way, Y = beta0 + beta1 x Gin + beta2 x Tonic + beta3 x (Gin x Tonic) + error term.

We can easily change this equation to the following form for a better interpretation:

Y = beta0 + (beta1 + beta3 x Tonic) x Gin + beta2 x Tonic + error term.

If the moderator is standardized and increases by one standard deviation, the slope of Y changes from ‘beta1’ to ‘beta1+beta3’. In your case, the beta3 coefficient is positive. Therefore, on can sag that the value of the beverage Y can be increased when mixing Gin with Tonic water.