This would not be fire frequency because, if it is point data, it would either be an ignition event or an area (polygon) generalized into a point location. Thus it would aptly be ignition frequency or a misrepresentation of a fire process. Inherently, fires occur across space and as such are associated with area and not discrete point location(s). Sorry about the semantics, but this is actually important terminology. If you data represents actual fire perimeters that have been generalized into point locations then it is quite invalid for any type of spatial analysis because there are variable areas associated with each point.
As to your question, please realize the the results could vary wildly given different raster cell resolutions. However, you can use the Point to Raster tool with "count" as the option for the cell_assignment argument. This will give you the number of points intersecting each cell and thus, frequency.
I imagine that you will receive advice to interpolate this data. This would not be appropriate because you are not interpolating a process that varies with space. Even if you had data that represented actual fire perimeters. In fire, there is a very strong anisotropic process (directional effect), that represents a specific spatial process associated with each given fire. Because of this approaches like Kriging are less than optimal and require some degree of savvy to fit the appropriate model and would have to be done for each fire before deriving frequency. I am not even sure that you could fit a cohesive spatial interpolation model to something like fire frequency because it likely does not follow an explicit spatial process. If you find that it does, publish it fast!