I think it is considered a failure because it doesn't really work
for uber.
If surge is low, it isn't worth a driver heading across town without a fare just for a 20 or 40% bonus in fares (especially if surge is gone by the time he gets there). Low surge doesn't do much to increase supply.
However, if surge is high, customers won't want to pay. At least in my market, taxis are still a perfectly viable option (and can be flagged off any street corner), and last I checked, a surge of 1.7x or more made UberX more expensive than a taxi. I would also be willing to take a bus or ride a bike in some instances (or simply wait for surge to go away). I think I have only paid surge over 2X once, and it was for a short ride where the money just didn't matter much.
High surge is great at reducing demand. This helps clear the market and is overall a market success. Unfortunately, Uber is in the business of selling you a ride. Reducing demand for their own product is not exactly a success.
The only times that it is really a success for uber are times where many drivers expect surge in advance. New Years Eve is a great example of this. In my city, there weren't crazy surges on NYE like there were a few years ago. Drivers knew there was money to be made and they went out. They may have been disappointed that they weren't seeing huge surges (averages were less than 2...probably a lot in the 1.2-1.4 range), but they were out there driving, and customers were surprised by the low fares and took more ubers.
I think the idea is that they could use machine learning to predict the non-obvious times. Get drivers on the road in time to meet the demand...thus increasing supply rather than decreasing demand.