I can remember, what 20, no, 30 years ago, when the Connection Machine was going to be the harbinger of a new class of computers. We were all going to have to learn to deal with massively parallel programming.
Politics --- literally. The CM's market niche was high-end research, and most of the sales there were to labs and projects funded by ARPA (or DARPA; the D comes and goes). This led to accusations from competitors that [D]ARPA was steering the labs to the CM, which got a sympathetic ear in Congress --- and all of a sudden, the CM was completely unacceptable to [D]ARPA.
(Accounts of this vary; as an ex-employee and friend of many who remained during this period, the technical staff certainly believed that they were winning those sales with a better product, but those at, say, Kendall Square Research almost certainly believed the opposite, in good faith.)
A number of companies are running into those parallel computing issues again, as we're starting to see more cores added to new CPUs instead of large increases in clock speed like we saw in the late 90's. Plus there are larger distributed computation systems out there like the Amazon and Google clouds.
And then I never heard of it again until years later this (http://longnow.org/essays/richard-feynman-and-connection-mac...) article about Richard Feynman popped up on the internet.
I still don't know why the CM never went anywhere. Inertia, scaling limitations, grabbed by the spooks and now only used by NSA for reading my email?