In the late 1990's supercomputers will have computational performance approaching one trillion floating point operations per second. These high performance systems will contain tens to hundreds of giga bytes of internal high speed memory either distributed, centralized or combined to provide the required bandwidth for efficient computation. The impact on the on-line storage and mass storage archive subsystems is estimated from actual user experience with current supercomputers. The on-line portion of the storage subsystem must satisfy the dual constraint of high capacity and very high transfer rates, while the mass storage portion may have greater capacity with a lower transfer rate requirement. The actual 1990's storage subsystem will be composed of both magnetic and optical disk and tape units. Some of the factors critical to design of the total storage subsystem are examined.