Ok so there's a few things here.

To answer your question: yes, for delta-delta-cT analysis, it is ok if your gene of interest has a lower cT value than the reference gene. This means that your delta cTs will be negative numbers, but the final quantification should come out looking normal (with positive numbers, which are all very close to 1 in the reference group.)

A fold change of 5000 is enormous. Sanity check this number: is this what you expected to see? Compare it with literature values. If the gene is negligibly expressed in one cell type and extremely highly expressed in another, this is reasonable, but in most other contexts it would be unreasonably large.

Delta-delta-cT makes the core assumption that the amplification efficiency of your reference gene primers and your gene of interest primers are the same. Have you checked your primer efficiency? This plus a literature search will give you confidence that the very large fold change you observed is real. I also mention this because it's important to understand why diluting your primers would be a bad idea. It's the concentration of the template, not the primers, that determines the cT value, and up to a certain point you could dilute your primers and see no change in the cT - your reactions would just plateau earlier and at a lower peak fluorescence. If you DID start seeing an increase in cT values after diluting your primers a lot, this would be because you had reduced the efficiency of your PCR, and the amount of target DNA was no longer doubling with every cycle. This would violate the core assumption of delta-delta-cT and make your results meaningless. If this concept is new to you, I think it would be a valuable thing to read about, because loads of people publish garbage qPCR results and understanding the assumptions that underlie the method will protect you from accidentally doing this too :)

Just to add one point to Laura's great answer:

Using a single refence gene is dangerous. Duringthe polarization, the expression of this gene may change, biasing all your results. This may be practically irrelevant in your case with huuuge effects in the selected markers. But anyway - it's current good (better) practice to use a couple of reference genes.

Thanks to everyone for your answers :) I did confirm the specificity and efficiency of the primers before my experiments. I am a little bit surprised, but not too confused by these results. I had to dilute my cytometry Ab 1/6400 for the signal to show on the graphs. CD80 is a key marker of M1 and is supposed to be very highly expressed in those cells; not in M2 and M0 (which are often described as «M2-like» after a short amount of time in culture because they naturally tend towards resting, anti-inflammatory state in the absence of polarizing cytokines). I also did ELISA on another key M1 cytokine and it's expressed about 2000 times more, which I have seen in the literature. So I do have 3 methods to confirm the polarization and they all give similar, albeit impressive results.

I do plan on adding a second housekeeping gene, but since I was testing multiple timepoints and testing the primers efficiency this time, I had no wells left on my plate to do so. However, I did include non-differentiated Thp-1 and GAPDH doesn't change much (all between 17.62 and 18.09, so a difference of less than 0.5 ct) between polarized, differentiated and undifferentiated.

Thanks for explaining why I shouldn't just dilute the primer !

Why you shouldn't just dilute the primer:

the whole qPCR analysis relies on the asusmption that the amplification efficiency is (almost) ideal. That means in every cycle each double-stranded amplicon molecule is doubled, generating the double signal. The Ct value is the number of cycles needed for this signal to reach a threshold. This threshold must be reached before the amplification efficiency is limited or compromised (e.g. by resourse depletion or accumulation of PCR inhibitors like pyrophosphates).

Diluting your primers would not change the amplification efficiency during the early cycles, but it would reduce the numer of cycles that could run under optimum efficiency. So it sholud not have any effect on your results, and if it had, your results would not be useful for quantification.

You would only try to use lower promer concentrations when the amplification of unspecific products (like primer-dimers) is a problem. But you said the primers are specific, so this is not relevant for you.

All answers above are great!

I just want to add that, working with the same assay for so long, we learned that polarization towards M1 or M2 highly depends on what stimuli we use. For M1, we usually use LPS and IFN-γ, and a characteristic increase in CD40 and CD80 occurs (compared to M0). For M2, we use IL-4 and IL-13, which results in high expression of CD206 and CD163. This first being done by Flow cytometry. For gene expression, I found CXCL-10 is a very good marker for M1 and TGM2 for the M2 phenotype.