I think the main problem is that the failure of a small section is a catastrophic failure of the whole system.

I don't see why you can't compartmentalize it exactly like how I described. A leak in one of the ISS's modules wouldn't take down the whole ISS.

And then what? The train just stops in the middle of nowhere for however many days until the one chamber is fixed?

It returns obviously. Making repair or replacement of damaged sections economical is one of the engineering challenges. This by itself suggests compartmentalizing along the same lines I suggested.

It would probably be ideal if you could factory make each section and ship it on a standard truck, but with the growth of additive manufacturing, other possibilities also open up, like 3D printing the large metal container onsite.

I think you'd have to print it onsite, otherwise you have no gain in the loop being underground as you'd have to dig basically the entire ground up to swap sections.

Having a vacuum chamber large enough for a train to not only fit in but actually move through at high speed before getting to the next section still requires a few hundred meters per section if not more. That in itself is insanely difficult. Connecting multiple such sections while maintaining the vacuum is more difficult still. And still, you'll need thousands of pumping stations all around the middle of nowhere, that typically need constant maintenance.

For reference, the largest existing vacuum chamber is some 30mx35m (100 feet x 120 feet). Each one of the sections you would need here would have to be many times longer than the biggest vacuum chamber ever built to make any sort of sense.

The LHC achieves a hard vacuum over 27 km, and also features superconductors along its length, so clearly it's doable. I'm not trying to trivialize the engineering challenges here, but I've seen that people have tendency to immediately jump from "it's hard" to "it's impossible or infeasible", even though no one's even made an attempt. Some clever engineering could await discovery that makes it all simpler.

Sometimes this sort of skepticism is warranted because our understanding is sufficient (like our understanding of material science needed for a space elevator), but sometimes this is less clear, and I think hyperloop concepts fall into the latter category.

Is it large enough for a train?

The LHC tube is 27km long but ~6cm in diameter. It is true that they are pumping down a more impressive vacuum volume though, but not in the tube, but around each of the many, many superconducting magnets. That's about 9000m³ (compared to the ~150m³ of the actual tube).

Still, the LHC is hardly maintaining that vacuum continuously all year round. And it is one of the most sophisticated engineering and scientific projects attempted in history. Hardly a good idea for a train project.

It was the most sophisticated engineering not because of the long vacuum, but because of the tolerances, the semi-novelty of the superconducting coils at the time, the sensitivity of the equipment needed, and the extreme data collection capabilities needed to capture and store as much information as possible on picosecond timescales. None of these factors are necessarily applicable to a hyperloop.

Even so, once you've built one LHC, subsequent ones would be substantially cheaper because the primary difficulty is building the scientific and engineering knowledge needed while you're building it. Similar factors apply to programming: the first version of your program took awhile to build, but if you suddenly lost all of the code, rebuilding it would take a fraction of the time because you don't have to build your understanding of the domain the second time around.

Maybe all of these factors combined still wouldn't be enough to make hyperloop economical, but that conclusion is not obvious.