I agree this specific post lacks any useful detail, and is highly unlikely to be anything more than vaporware.
That said the claim that one could develop a sub-$10 tracker consisting of a GPS receiver, LPWAN tranceiver, and a small ARM for generating signed messages isn't completely absurd -- such a device would have a BOM in the ballpark of $35-40 if cobbled together from retail priced dev modules. It's friggin absurd how cheap hardware has become. Look at the OrangePi Zero as an example. As low as $7 to the end user for a quad core board with WiFi, Ethernet, 512mb ram, and some SPI flash.
I agree with you on the intermediate server issue -- not because a tiny ARM can't directly submit messages, but because most reasonably priced existing LPWAN solutions don't provide internet IP connectivity to devices. If the devices were provided with full IP, there's no reason a blockchain client couldn't run on a cortex m0 -- keep in mind being able to submit messages to a blockchain doesn't mean you have to ever even look at, let alone store or process the chain.
I think it's wise to be sceptical about this project. I'm just less convinced that it's wise to equate blockchains with hype.