The circular financing, the lack of product, the underwhelming results, in concert with the obviously inflated valuations and lack of government involvement on any market level leads one to believe that this entire house of cards is doomed to fail. The only question is when.
The next question is what happens afterwards. They've floated the idea of a government bailout. Which is what I think the pretense of the IPOs was truly for, to bind the fate of the stockmarket to these companies, and to force the government to step in whether the people want it or not.
The terms of that bailout was what I was thinking about. It seems extremely likely that the government would take ownership of some or all of the compute capacity formerly owned or leased by these companies. It would give them access to an extremely powerful resource, put to another use, and it would allow them to sidestep all the "anti-military" and "anti-government" sentiment currently in the workforce of these companies.
The real worry is that they then take all of the data they currently have, all federal surveillance cameras, all prison cameras, all the flock and ring and other private cloud based surveillance systems, and start feeding that into the system. Then combine that with all the data they've been hoarding off the Internet for more than 2 decades in data centers in Utah and run it through inference and training and you've got a powerful citizen prediction model that could be used for anything.
Westworld season 2 and 3 were a joke, but the idea of something like the Rehoboam starts to look more and more real.
This is pure theory. Thoughts?
IMO they aren't designed to fail. They're just not being developed to be good long-term use tools. They want more investor money, and more tax money. When it gets too bad, the C-levels can jump ship, and take a portion of that money with them, converted to other forms of value stores.
Altman seems to have bet on US govt bailouts as part of his business model. The guys running these companies know it's a bubble, and they're betting on having insider knowledge to use, to escape the fallout, just before it all crashes down.
Thing is, that's part of what the general purpose models are built for. Be big, be flashy, and make some headlines, to get more new money rolling in. Being more useful to potential or existing customers, on any cost and performance basis, won't matter until the bubble starts bursting.
The panopticon part will happen, if not actively stopped, regardless. It will need massive amounts of storage, but only a small fraction of the data centers being built, as models are forced to get more efficient, rather than just bigger.