The answer is "yes", a hologram can be made one photon at a time. The experiment has been done. Of course it is a very slow process. The first such experiment I read about used a photographic emulsion to capture the image over a long period of time. A standard 3D hologram recording setup was used. The laser beam was attenuated until an average of less than one photon could be in the holographic setup at a time, and a photographic emulsion recorded the location of each photon at the hologram recording plane. (I'm searching for the paper and will post a link when I find it.)
Modern photon-counting detectors make it a bit easier; and some single-photon-on-demand emitters have been developed in recent years. Some key words you could use in a Google search are "single photon holography" and "photon counting holography"
However, here is where the weirdness of quantum mechanics comes in: In order for a hologram to be recorded, for each photon in the recording it must be impossible to know or even to figure out which path the photon took (the object path or the reference path). If anything is done to control or determine which path each photon takes, there will be no interference pattern. So, if you flipped a mirror very rapidly to direct single photons down the reference or object path alternately (instead of using a beamsplitter for example), then you could not form a hologram because it would be possible to know which path each recorded photon took by detecting its arrival time.
So the answer to the question is a qualified "yes": as long as there is no way to know which path each photon takes, then yes, a hologram will be formed. The question is actually a very good one, because a complete answer would need to explore some fascinating subjects like what a photon is (not the simple answer that it's a packet of electromagnetic energy), how "EPR-based imaging" works, and even this odd paper.
If the question is modified to allow each photon to "make up its own mind" whether to take the object path, the reference path, or both -- and to disallow anything that lets you know which path it took --, then the question still has some interesting content. What if the reference path is a lot longer than the reference path, so we "know" that any interference that occurs is between "somethings" that were emitted at different times?
Well, it turns out that the coherence of, e.g., a Helium-Neon laser is not just the few inches of coherence length described in most specifications. In fact, the coherence is periodic. For a few inches before and after integer multiples of the length of the laser cavity (on the order of a few feet usually), the coherence returns. So photons emitted now are neatly in step with photons emitted at time intervals determined by the laser's cavity length. If the object and reference beams differ in path length by 3 or 6 or 10 times the cavity length, plus or minus an amount smaller than the traditionally calculated coherence length, then it's still possible to record a hologram. In effect, each photon is spread out over a distance of meters -- and over a corresponding length of time.
So, it can be very difficult, even when a laser beam has been attenuated enough that typical calculations would tell us there's only a single photon in the setup at a time, to be sure that the wavefunctions of hundreds of photons aren't present in the setup simultaneously.