"... you would need around 133 million ... chips ... and it would burn about 15.3 megawatts ..."
The important thing to take from this statement is that it ** can ** -- from an engineering and architectural standpoint -- be done.
We already have supercomputers on this planet, like the Fujitsu K:
-- -- Wikipedia: K computer
-- -- -- -- http://en.wikipedia.org/wiki/K_computer
that consume electricity at similar scales (9.8 Mw for the Fujitsu K, 15.3 Mw for a full-blown human brain emulator using these synapse chips), and interconnecting 133 million components, while admittedly a very daunting engineering challenge, is not impossible.
The key factors against building a device like this are cost and time. Current geopolitical sentiment is leaning away from "big science" endeavours, especially those that appear to be wholly theoretical, and thus be of little immediate commercial, military, or consumer benefit.
Which means we're still probably at least a few decades away from building -- and then having to live with -- a Forbin-esque computer.