-quote
Cray's Adaptive Supercomputing model is based on the idea that in order to achieve truly superior application performance, as well as address the need for higher user productivity, HPC systems must go a step beyond simply providing access to a variety of processing technologies. Different applications achieve optimal performance on different types of processors, but most high-performance computers typically offer only one type of processor. Adaptive Supercomputing combines multiple processing technologies in a single system and hides this complexity from the user through innovative new software technologies, allowing scientific and engineering problems to be solved more quickly — and programmers and end users to be more productive — by adapting processing to the requirements of each application.
-endquote

This is considered to be a supercomputer because it enables the scientists as well as engineers in academia, government, and industry to apply this in various applications. The Cray supercomputers provide superior performance in various critical applications in the filed of science and engineering.

Source:
http://www.cray.com/About.aspx

Building supercomputers was a dream, an aspiration, and a life's pursuit for Seymour Cray, and his work on the computers that bore his name was the culmination of work he had done for the U.S. Navy, for CDC Data Corporation, and finally for his namesake company.

When Cray left CDC in 1972, after his work on the 6600, 7600, and minimally the 8600, he took much of the supercomputer fire with him.

While Cray's departure from CDC wasn't overly dramatic, his impact on supercomputing was. Cray artfully designed computers so that each part worked to efficiently speed up the whole, and he usually didn't rely on the newest experimental components, preferring instead to tweak existing technologies for maximum performance. For instance, the Cray-1 was the first Cray machine to use integrated circuits, despite their having been on the market for about a decade.

At 160 MFLOPS, the Cray-1 was the fastest machine at the time, and despite what seemed like only a niche market for expensive superfast machines, Cray Research sold more than a hundred of them.

Form and size were always concerns for Cray, as far back as his days developing the CDC 160, which was built into an ordinary desk. There was also a big concern with the heat that could be generated by so many parts being packaged so tightly together, so Cray's designs typically involved unique cooling solutions, whether it be Freon on the Cray-1, or Fluorinert, in which Cray-2's circuit boards were immersed.

Cray Inc is a company that produce supercomputers and have been since 1972. While there are other companies that produce supercomputers, Cray is probably the most well known or recognisable as a supercomputer producer. Intel also make them but are much more known for their processors.

What are supercomputers? Well basically they are super... computers. They have enormous amounts of processing power which allow them to do typically infeasible computing tasks in a much shorter period of time. Today most supercomputers achieve this increased processing power by using multiple processors in parallel, similar to how common dual or quad core processors work, although generally these are completely separate processors. The fastest supercomputer in the world, a Cray XT5 "Jaguar" has a processing speed of over 1.75 petaflops, or 1,750,000,000,000,000 floating point operations per second. To give an idea of scale, a common desktop 3 GHz processor can do 3,000,000,000 operations per second, although these two measures aren't really comparable as floating point operations are more complicated. The Jaguar computer has over 200,000 processing cores which enables this huge speed.

So what kind of businesses need supercomputers? Generally, not very many. If you aren't sure if you need one you probably don't. Things that are generally achieved with supercomputers would be almost impossible any other way. Generally they are used by academic and research institutions for completing complex calculations or completing detailed simulations. The best problems solved on supercomputers are those sometimes called "ridiculously parallel" problems. This means that they can be split into many many sub computations that can all be run at the same time for merging later, unlike most problems which need to be worked through sequentially with the results of a previous process feeding into the next.