Yes, but the point here is, they don't need our help to build "more efficient" supercomputers. If you design your chips to go wide and slow, you can build an efficient architecture on older process nodes. These are wide vector machines just like Intel is shipping; you can determine this from the claimed 140.8 GFLOPs:
1.1 Billion clocks * 2 FPU vector units * 16 cores * 4 instructions/vector unit = 140.8 GFLOPS. Not as wide as the 512-bit vector units in the latest Knights Landing, but really not that far behind.
Since the cost of the processors is invariably out-shined by the cost of the interconnect, which doesn't benefit as much from process shrinks, it's not such an issue dealing with the extra cost of large die processors. So the question is: if China already has supercomputer technology good enough to compete with the world, why can't we make a fast buck selling them more?