Some quibbles:
IBM prevented 32-bit OS/2 for 80386 not from bad planning, but an internal struggle to keep 32-bit PCs from biting into the AS400 mini computer business, and there were internal wars for the board approvals.
This was the end of the line for Gates's frustration with IBM, as OS/2 took resources from projects he and Balmer were convinced would take off. Publicly claiming that upcoming Windows 3.0 would not be "Presentation Manager Lite", MS still death-marched developers to produce the release, while devs allotted to IBM sat on their hands or did code reviews for IBM managers. Win 3.0 Program Manager kicked ass on Presentation Manager, it was definitely not "lite" - and it ditched all the heavy-baggage of IBM SNA requirements.
"OS/2 NT" is a bit misleading. Late in the endgame of the IBM/MS relationship, Gates discovered that Dave Cutler was being cut away from DEC, with a recalibration of Prism and the future of Alpha. Cutler had begun a 64-bit microkernel evolution of his VMS system. OS/2 3.0 was on the boards, still dragging MS resources and tying up IP. Gates hired Cutler to build an alternative, skunk works kernel from his Prism design work, with the hope of porting the Windows System 32 layer with dependencies etc. When the last bitter contract work was delivered for IBM, Cutler and the Windows team ground out the hard work of delivering their kernel, TCP stack, and Windows 3.11 port —Windows NT.
Most of this stuff is well-covered in Carroll's "Big Blues" along with Zachary's "Show Stopper!: The Breakneck Race to Create Windows NT". I had a small part at the NT launch in Moscone Center, working for a ghost-writer on the Sybex NT book that launched at that event