1- User acquisition costs (discounts, marketing, etc - for both sides of the platform). this gets more expensive in the face of competition, and there's some hope that if you can "win" the market then eventually these costs will be reduced sharply.
2- Money as a band-aid for reliability/support issues. Frequently, an Uber driver refuses to take me on a long trip (SF to Mountain View, for example), makes me get out of the car, then marks me as a no-show or that I canceled the ride. Fortunately, my rider score is high enough that I just fill out the form describing what happened, and Uber (usually almost instantly) refunds the fee and sometimes also gives me a make-up $5 credit to apologize. With food delivery options, a non-trivial fraction of the time, the wrong order arrives, and similarly any complaint to support usually results in an immediate refund + credit. Without an easy, scalable way for the support team to audit what really happened on the ground, chances are they are not penalizing the driver / restaurant for these incidents unless there is a clear pattern in the data with a particular provider. So this is just loss for the platform. With very small margins it takes a surprisingly low percentage of such cases to really eat away your unit model.
3- Subsidizing low-utilization markets. Markets with a much higher density of supply/demand will be more efficient and profitable; newer markets will often need subsidies to get over the cold start problem (eg when uber launches a new city, you're gonna need to pay some drivers to sit idling on the road, or else someone opening the app won't see a driver; you might also discount rides even more heavily to get past this phase quicker). Compared to the other two factors, this is probably the best-case scenario for why a company might be bleeding money, and why you sometimes hear people say "Uber is profitable in XYZ cities", but it's hard to say from the outside if this is actually what's happening or if there's something more fundamentally wrong.