Computational Analysis requires material properties to be assumed to be modeled. Those assumptions assume some uniformity across the materials - uniformity that is very difficult to obtain is large composite structures. Different loads are handled by different parts of the composite structure (fibers handle tensile stress only, fibers must be laid in varying directions to handle tensile loads in various directions, compression loads are handled by the epoxy filler, composite structures don't have traditional elastic/plastic properties, etc).

Large composite structures must have the fibers of the various layers oriented exactly, the epoxies applied uniformly and appropriately, and the pressure and heat applied precisely to obtain a "theoretical" curing - there is no way to measure actual curing and completeness of lamination. The best they can do is try to apply uniform pressure and place thermo-couples at strategic locations and hope that proper temperature patterns at those locations are indicative of proper temperature profiles throughout the part.

On the other hand, thousands of years of working with metals have led to processes that consistently produce materials that are uniform in properties that are easily modeled (compressive strength, tensile strength, ductility, etc).

Nothing is impossible, but this technology is being advanced as far in 20 years as it took metals 1,000 years to advance.