Sorry, but I don't understand the answers that have been given.
Even looking at the front-end schematics of some scopes in the same class (well, my Rigol DS1054Z and the DS1052E), I can't see any justification for statements as:
Those spikes are too slow for AC.
and
it should definitely show up as something when AC coupled, just at a much smaller amplitude.
Agree, but why at a much smaller amplitude?
How can a large capacitor in series -that is to say "AC mode"- stop narrow pulses?
The repetition rate is , in fact, low: 500ms as can be seen, but this kind of pulses have lots of energy content at high frequency (as I'm sure mostly everyone here is aware).
They look to be more or less 1-2 pixel wide, at 50pixel and 200ms per division, they are shorter than 8ms, possibly much shorter.
So, what should happen, as in the expectation of the OP (and mine) is that the DC component is filtered out, but pulses go through.
If the pulses are
very wide, you might also see sloping roof and floors, but the high dV/dt should go through without problems.
Just to make sure I was not deluding myself, I quickly built a demo with a Nucleo board and two resistors.
(Thanks to the CubeMX taking care of PWM configuration, I only had to add a single line to start it).
The results are below, and match my expectations.
The sample rate is in my case 5MS/s, with 12MS memory.
A very narrow pulse (220ns, the lower limit for this sample rate to catch something...) go through flawlessly, DC and AC.
As does a longer pulse (1ms), AC and DC.
A 10ms pulse shows the long RC time constant of the 220nF cap in the Rigol with all the rest of the input network.
So, maybe the OP really needs a better scope?