The current fastest supercomputer consumes 17.6 MW and 24 MW with cooling and occupies 720 square meters of space. The previous record holder used the same amount of power, but was significantly less powerful. Supercomputers are constantly evolving, and engineers are designing them to be more and more powerful ,while using less power and taking up less space. This is done by improving the processor architecture and taking advantage of new technology.
The current supercomputers are limited by the amount of power they can draw, and the technology they have available. In theory, there is no limit to how fast a supercomputer can be. They are only limited due to environmental issues such as using too much power or requiring too much space.
Cooling is also an issue with the current super computers, The more powerful the supercomputer is, the more heat it outputs, resulting in more energy spent cooling it. Currently, cooling super computers with water has proven to be the most efficient, but a lot of power and radiators are dedicated to that task.
Interconnects have also been a issue with the fast growth of supercomputers. Every processor in the super computer must communicate with each other, as well as send and receive data. The sending and receiving of data is called I/O. With faster processors, they can perform more I/O operations a second, but they can only do as many as they are sent. This is where interconnects come into play, these are responsible for transferring data. Similar to a pipe, you can only move a certain amount of water through a giving pipe. So if a pipe was able to move 400L/s, but 600L/s is trying to move through it, only 400L will go through a second. So the faster processors get, interconnects have to catch up to be able to move that large amounts of data.
Source: Computer World, Intel,