Back in the late 1990's I worked for a company that had a datacenter filled with about 3,000 servers from a PC vendor I won't mention (located in Texas). Over the course of a month or so we had three separate incidents where individual servers caught fire. The server would first completely stop responding - lights were on but you couldn't see anything on a monitor, couldn't connect to it over the network, couldn't even power it off by pressing & holding the power button. In the end the only thing we could do is pull the power, wait a minute, then plug it back in. Each time we did that the video chip on the motherboard would erupt in a geyser of flame.
A few weeks after we reported these incidents to the manufacturer they informed us that they had implemented a recall on a batch of motherboards, and about 700 of the servers in our datacenter were subject to it. It was my task to go into the datacenter and identify those 700 servers. The manufacturer sent a team of workers to the datacenter who set up an assembly line to replace all those motherboards, which took about a week to complete.
At some point well after that little nightmare we learned that the manufacturer had traced the fault back to a vendor in China that had provided the capacitors used in the power regulator circuitry on the motherboards. The capacitors had been apparently manufactured with faulty dielectric material, and after time they started failing in spectacular fashion. They'd apparently build up a huge charge and maintain it until power was cut & restored, at which time it would release a surge of all that stored up power all at once. Something about the design of the motherboard caused that surge to go directly into the video chip.
My completely unscientific guess is that Samsung will likely eventually trace this to something similar - a fault in a component provided by a third party that under certain conditions results in an overload.
