Gordon put it quite excellently:
Computers are not Turing machines.
Arrays on "real" typical general purpose machines aren't stored on the "infinitely long" linear storage, but in RAM, Random Access Memory. Technically, (and quite frankly simplifying a lot), you just get any address from RAM by understanding it as a path through the memory addresses. So access to any address takes the same amount of time.
Now, for arrays, you can directly calculate the address of n'th Element by taking the Address of the first element and adding n times the size of a single element.
Remember: Turing machines are a concept of how to prove and understand certain things, and do not reflect reality of how things are actually done. The same goes for complexity calculations: Of course, in reality access to any element in a vector doesn't always take exactly the same time, because assumptions that computer science people need to make to say interesting things about algorithms can't always fully represent every physical machine that can run a given algorithm -- real modern computers have caches and prefetching memory controllers, so that accessing a piece of memory that you "just" visited is much, much, much faster than just getting any memory.