Shannon's capacity refers to the maximum signaling speed at the physical layer but this article muddles it all up with higher levels of the network abstraction layers.

That said, I think the title is probably true. It's been known for a decade that turbo codes and low-density-parity-check (LDPC) codes can approach the Shannon limit, especially if relatively long decoding latencies are tolerable. It just took some time for turbo and LDPC codes to make into standards and get deployed.