Statistician (but not a Census statistician) here with a few answers:
Black people have a higher than average non-response rate, even controlling for income, possibly due to a cultural mistrust of government institutions (for good reason).
Recent immigrants may not be able to read the language the census was mailed to them in, and may also be hesitant to respond due to fears of deportation- the point of the census isn’t to hunt down illegals, and people here legally shouldn’t be deported anyway, but the current administration hasn’t done much to reassure people. These effects will apply in differing amounts to Latino and Asian communities.
Finally, the bulk of the census is conducted by mail, and it takes a while to complete. Families without a permanent home might not be counted because they are only temporarily living with another family when the census arrives at that house, or they might be moving between addresses so that they aren’t counted in the response by any household at those addresses. This also disproportionately affects poor people and minorities.
And at least for me, I only received instructions in the mail telling me to do the census online. Not all families have easy access to the internet, and some just can’t be bothered to take the time to respond. This again disproportionately affects poor people and minorities.
To try to counteract these problems, the Census hires people to walk on foot and get responses from low coverage areas. But if you cut down on that effort, you’re going to miss some. One of the results of this could be that the census shows fewer poor people in some cities which will affect government funding decisions. But the tough part for a statistician is showing how much time is needed to walk around and get the coverage that you desire (and how many people you hire to do it). You can make some relatively small changes to your assumptions and get different answers. But cutting the time by a third is a pretty dramatic cut, so without looking at the details, I’m guessing that the coverage has decreased. That’s generally a bad thing since a census is supposed to be a full count of the population and not a sample.
Edit: I’ll also tack on a more nefarious interpretation than just using the Census to cut federal aid money to some cities. The Census is also used to portion electoral college votes for presidential elections. Because of the communities that are affected by this undercoverage, states that normally vote Democrat are more likely to lose votes than states that normally vote Republican. So this could make it easier for a Republican president to be elected in the next couple elections (the votes for the current election will not change however).