WeWorks
The lock-in warning rather assumes there'll be something worth locking into. Plenty of enterprises are already past the honeymoon and into the hangover phase, where the maths gets uncomfortable. Frontier models cost more than the engineers using them, and because the output is voluminous and inconsistent (adaptive reasoning being a polite term for "different answer every time"), you end up hiring more humans to scrutinise what the machine produced. At which point someone in finance asks the obvious question.
So the quiet trend is reversion: humans doing the work, AI used conservatively at the edges. That's a problem for vendors who've capitalised themselves on the assumption of exponential consumption. Until inference costs fall by an order of magnitude, they're sitting on warehouses of depreciating silicon that nobody can afford to run at the advertised scale.
Which rather explains the sudden enthusiasm for IPOs. It's not a victory lap, it's a liquidity event before the music stops. Forrester worries OpenAI might be AI's BlackBerry. But BlackBerry was killed by a better product. The less flattering scenario is that there's no iPhone coming, just a row of WeWorks with GPUs.