In June 2012 in front of an audience of journalists, Microsoft's then-CEO Steve Ballmer was in philosophical mode, pondering how important hardware was to Microsoft, a company best known for its software.
"At our foundation, Bill Gates and Paul Allen made a bet - a bet on software," Ballmer said. "At the same time, it was always clear that our unique view of what software could do would require us to push hardware sometimes in ways that even the makers of the hardware themselves had yet to envision. That's the nature of the dynamic between hardware and software."
Ballmer was about to make another bet on software and hardware with a new tablet device called the Surface, running a new operating system.
The Surface tablet was a reaction to the runaway success of first Apple's iPad and the myriad of cheap Android tablets that followed, biting huge chunks out of the PC market along the way.
While Microsoft's hardware partners had been building Windows tablets for years, they had never generated much momentum. Surface was Microsoft's way of proving that a Windows-powered tablet could be cool, and maybe even sexy, at a time when Windows tablets were dull and utilitarian.
The hope was that with Surface, by taking control of the hardware and the software (like arch-rival Apple) Microsoft could come up with a Windows tablet that could take on the iPad and not spook PC makers too - hence Ballmer's emphasis on Microsoft's history with hardware.
"Much like Windows 1.0 needed the mouse to complete the experience we wanted to give Windows 8 its own companion hardware innovation," he said at the unveiling of the Surface. "The Microsoft Surface embodies the notion of hardware and software really pushing each other."
The tablet went on sale in October, and soon analysts were predicting that Microsoft would sell millions of the device - the company was reported to have built somewhere between three and five million in the run up to the holiday season.
Almost immediately, the Surface failed to live up to these expectations: industry watchers said less than one million sold in the first quarter. Despite an initial positive reception from some, reviews were tepid.
And barely a year after Ballmer had unveiled the Surface, Microsoft had to take a shock $900m writedown on unsold Surface stock, with one calculation suggesting the company was sitting on as many as six million unsold tablets. Price cuts and various special offers followed as the company tried to reduce the levels of inventory it carried.
Still, Microsoft stuck with the Windows RT version of Surface, launching a second version in October 2013.
Gradually other PC makers scaled back and then abandoned plans for Windows RT devices until, apart from Microsoft, only Nokia (itself soon to be bought by Microsoft) was left making Windows RT tablets. It launched its first, the Lumia 2520, in late 2013.
And last month Microsoft confirmed that it had ceased production of both the Surface 2 and the Lumia 2520: the strange story of Windows RT is all but over, and it lasted just over two years.
Rather than the hardware and software pushing each other on to greater things as Ballmer had hoped, it was a combination that left users underwhelmed.
What was different about Windows RT was that it was built specifically to run on ARM processors - more typically found in a smartphone - rather than the x86 chips found in most PCs.
Using ARM chips was attractive to Microsoft because they use less power, and so more practical for mobile devices.
But rewriting Windows for ARM was challenging and meant compromises: Windows RT couldn't run standard Windows applications, only those piped in from the new Windows Store.
Developers were cautious about rewriting apps for such a small market so the app ecosystem - essential for any consumer device - failed to gain enough momentum to persuade wavering would-be buyers.
"It's a whole lot harder to create an ecosystem than it is to lay out a vision to create an ecosystem," said Al Gillen an analyst who covers servers and system software at research company IDC, adding one of the problems for Windows RT was the lack of compelling applications.
"Nobody buys any device for the operating system these days. Customers selecting Android phones or iPhones or iPads are doing so because of the applications they hope to run on those devices," he said.
"Microsoft was trying to address low cost and low power with ARM processors, and Windows RT was the OS that was created for that platform. That in and of itself was not enough: you also need an application portfolio."
Branding and positioning was a problem too: consumers didn't really understand what Windows RT was, or why they might want to buy what they perceived to be an underpowered product over a full Windows 8 device.
Sure, Microsoft fans could argue that the iPad didn't have the same functionality as a MacBook, so RT didn't need to do everything Windows 8 did. But rightly or wrongly the iPad and MacBook were perceived as separate, standalone products whereas RT was inevitably seen as the runt of the Windows litter.
"These ARM-based devices led to enormous confusion about what was a Windows tablet and what was not (even though it was called Windows and looked like Windows)," said Gillen.
Another headache was that, thanks to the quirks of Microsoft licensing, the version of Office of on the Surface couldn't be used in business (unless they upgraded their licence).
According to figures from Gartner, around 1.1m Windows RT devices were sold in 2012 and another two million in 2013. But compared to the total tablet sales of 240m this was "it was a drop in the ocean", said Annette Jump, research director at the analyst house.
With a new user interface, new app store and new hardware Microsoft was "trying to change too many things and not clearly communicating the changes," she said. Microsoft didn't have in mind a clear audience "which made it even more confusing".
Perhaps unexpectedly the Surface hardware has proved more successful than the software: Microsoft is now onto the third generation of Surface Pro, the version which runs full Windows 8 and is being mainly aimed at the much smaller ultrabook/MacBook Air market rather than trying to chase the mass iPad and tablet market.
And Windows RT isn't dead yet, at least not quite. Microsoft has said that there will be some sort of upgrade for Windows RT devices to give them some of the functionality which will come with Windows 10 - but has not so far provided much detail. "We will have more to share later," Microsoft told ZDNet.
And the company will continue to support Windows RT devices for some time to come even if it isn't selling them, but hopes of using Windows RT as a way of attacking the tablet market are long gone.
Some of RT's failure to launch is thanks to the onward march of technology: there are piles of small tablets running full Windows 8 on Intel Atom processors now.
Some of it is also down to pricing - Microsoft is offering cheaper Windows licences to makers of small tablets, helping them compete better with low-cost Android skates while still using the full version of Windows.
The arrival of Windows 10 is another factor: Microsoft has just unveiled the first public preview of Windows 10 designed to run on ARM-based Windows Phones and ARM- and Intel-powered small tablets and phablets.
And some of it is also down to a change in fashion: tablet shipments may decline this year as the result of the growth in phablets, while PC two-in-one hybrids (like Surface) are gradually gaining ground. In a strange irony, it may be that Windows tablets are the ones that will generate some of the growth in the tablet market this year - just not ones running Windows RT.