Then and Now, Part 1: RT (Premium)

A lot has changed since Microsoft first announced that it would port Windows to the Arm processor architecture. I’ve referenced some of this history throughout my recent writing about Windows 11 on Arm, Qualcomm’s Snapdragon X chips, and Copilot+ PCs as needed for context and perspective. But the full story is, I think, fascinating, and could become as important to the history of Windows as the shift from MS-DOS, the development of NT, or Windows 95.

That story could be told different ways. But there have been three major milestones in the history of Windows on Arm, the initial reveal in February 2012, the Windows 10 on Arm announcement in late 2016, and the beginning of the Snapdragon X-based Copilot+ PC era in mid-2024. So that makes some sense to me: Three separate stories that together form a cohesive whole. This is part one.

That said, there are also threads that run throughout this story. For example, Intel—the dominant chipmaker in the PC market—has sought to undermine Arm at every turn. Unfortunately, it has done so to date solely by flexing its market power: Each time Microsoft announced a major Arm milestone, Intel was on hand, like some uninvited house guest, to remind the world that it wasn’t going anywhere. This is a major part of the Windows on Arm story as well, which makes sense: Moving to Arm is all about embracing mobility and efficiency and rejecting the past.

Rejecting the past can be difficult.

An architecture for the future

Many readers are familiar with how NT was originally designed for non-Intel processors, and that the team was forced to embrace Intel’s 32-bit x86 architecture because it was ubiquitous with PCs. Likewise, many will point to the various platforms to which NT-based Windows versions were ported over the years.

But less well known, or perhaps less well-remembered, is that Microsoft also adapted Windows—before NT and since—to a series of architectural shifts within the x86 architecture. After initially supporting the 8/16-bit Intel 8088 processor, subsequent Windows versions were tailored for the unique architectures of the 16-bit Intel 80286 and its segmented memory model, the 32-bit 808386 and its flat memory model, and then finally 64-bit x86 (“x64”) designs that continued with a flat memory model but added 64-bit pointers and data types.

The transition to 64 bits began with Windows XP, initially for the Intel/HP Itanium, but later for x86-64, first via Windows XP Professional x64 Edition. That shift, which led to the modern 64-bit Windows 11 we use today, is particularly relevant to this story, and for two reasons.

Even on x86, 64-bit Windows versions run 32-bit code using an emulator called WOW64 (Windows on Windows 64-bit) that, among other things, abstracts the hardware for 32-bit apps and makes the system appear to them like a native 32-bit version of Windows. This design choice was tied to how Microsoft implemented backward compatibility with 32-bit x86 apps on the 64-bit Itanium and its new architecture. It made sense to use the same design with other 64-bit Windows implementations.

Unfortunately for Intel, x64 was created by its rival AMD at a time when Intel was pushing Itanium as the 64-bit future. This was the first major break between Microsoft and the chip giant because Microsoft chose AMD’s x86-64 design for Windows after NT project lead David Cutler, who was driving the 64-bit push at the company in the late 1990s, chose it over rival schemes, seeing it as the obvious way forward. This forced Intel, the inventor of x86, to embrace the x86-64 instruction set as an official extension of its architecture. This did not sit well with Intel, and this embarrassment caused it to more closely scrutinize anything Microsoft did with rival chip architectures so it could react more quickly each time and try to control the narrative.

The uneasy alliance between the two companies survived the failure of Itanium because Intel continued to dominate the PC market. And for the next decade, Windows and Intel were essentially synonymous, carrying the Wintel legacy forward into the 21st century. Intel’s alliance with Apple—Steve Jobs announced that the Mac would transition from the PowerPC architecture to Intel’s x86-64 chips in 2005—just as Microsoft was readying Windows Vista in both 32-bit and 64-bit versions served to cement Intel’s dominance in personal computing. That it came at Microsoft’s expense was, of course, perfect.

And then everything changed.

The rise of Arm in consumer electronics

Thanks to the iPhone and Android, the personal computing market shifted from PCs to smartphones and other mobile devices in the subsequent decade: Those devices are all based on Arm-based architectures, not Intel’s x86-64. Thanks to the success of the iPhone, Apple became the biggest company on earth, and its subsequent hardware platforms—iPad, Apple TV, Apple Watch, and so on—were all based on Arm, despite incessant lobbying from Intel. And in 2020, Apple delivered another embarrassment to Intel when it announced that it would transition the Mac off x64 and to the Arm architecture as well.

Intel wasn’t the only company impacted by the sudden and rapid rise of Arm. Microsoft saw Arm-based mobile devices, which were simpler, more efficient, and more reliable than PCs, as a threat early on. It had also spent several years trying to convince its hardware partners—chipmakers like Intel and AMD, but also PC makers—to embrace mobility and efficiency. For various reasons, this had little effect: As Apple innovated in turn with its sleek MacBook Air, innovative iPhone, and then the iPad, Microsoft sounded the alarm each time. But its partners did little or nothing in each case.

And Microsoft forced the issue.

Microsoft shipped Windows 7 in 2009, two years after Apple launched the first iPhone. Like its predecessor, Windows 7 was available to customers in both 32-bit and 64-bit versions, exclusively for PCs based on the x86/x64 architecture. Improvements in Windows 7 and the passage of time helped Microsoft solve the problems with Vista, and it was an immediate success, going on to sell an average of 20 million copies every month during its first three years of general availability. But Microsoft could see that the shift to Arm-based mobile computing would soon upend the PC market: It was only a matter of time before smartphones would outsell the PC.

Windows, reimagined

Windows 7 provided a solid foundation for 10 years of traditional PC usage, so Steven Sinofsky’s Windows team decided it was the right moment to go big and embrace the technologies that defined the mobile wave that was passing the PC by. They hatched a plan to reimagine Windows, which had settled into a comfortable but dated usage model that Microsoft had first delivered almost 15 years earlier with Windows 95. It was time to shake things up.

This reimagined Windows would adopt the best attributes of the Arm-based mobile devices that Microsoft was so threatened by. But it would also adopt Arm, literally. For the first time since Itanium and the beginning of the 64-bit era, Microsoft would port Windows to a new processor architecture. It would port Windows to Arm.

The resulting product, Windows 8, would ship in both 32-bit and 64-bit versions, exclusively for PCs based on the x86/x64 architecture as before. But there would also be a separate product called Windows RT that brought most of Windows 8 to Arm. Assuming it worked, that is: In the early days of this port, the system architects in Windows weren’t sure if running Windows on the Arm chips of the day was even feasible. With Itanium, especially, it had worked with a more powerful chip than those used in the PCs of that time. Arm chips were more efficient than x86-64 chips, but they were not as powerful.

This was a different kind of port.

In some ways, it was more like the work Microsoft had done bringing Windows to mobile devices—Windows CE, at first, but eventually various Pocket PC and Windows Mobile variants—or embedded platforms. But where those systems were always based on older mainstream Windows internals, stripped-down subsets tailored for the needs of the respective device types, the Arm port of Windows 8 also had to be up-to-date, in lockstep, with its x86-based sibling. It was a new world.

Fortunately, plans for this architectural shift had been laid years earlier when Jim Allchin pushed to componentize Windows so that his team could more easily mix and match capabilities to create specific product SKUs, or editions, and better scale the platform to meet the needs of servers. This work continued under Sinofsky with projects like MinWin that further componentized Windows, removing dependencies, and making further efficiency gains.

Then, the Windows system architects had a breakthrough. As Sinofsky writes in Hardcore Software, they brought a full version of Windows 7 up on an Arm-based Windows Mobile handset. “Actual Windows 7,” Sinofsky writes of his first demo in January 2011. “All of Windows 7. Windows Mobile at the time was built on a subset of the ancient 16-bit and 32-bit Windows code. They were running the modern Windows NT code base on a phone with an Arm chipset. It was nuts … I was speechless … This visit made it clear we were on our way … This was no parlor trick.”

Undermined by Office

That team then spent the next year bringing “the full Windows experience to Arm chips.” But thanks to the unique nature of the Arm architecture, the “full Windows experience” was changing, and those changes would impact the x86/64 versions of Windows 8 too. In the end, Windows 8 and RT would both provide identical user experiences that offered a full-screen mobile interface and the classic Windows desktop side-by-side.

That wasn’t the original plan: The Windows team initially wanted to abandon the Windows desktop and its legacy apps in Windows RT, and provide a pure mobile experience similar to that on Apple’s iPad. But the Office team refused to support the new mobile apps platform, called Metro, in part because its desktop productivity app suite was already best-in-class, and it could easily be adapted to run on Arm. Why create a new version of this complex product from scratch?

Worse, this team was also actively (and secretly) working overtime on a port of Office to the iPad, and it didn’t want to pull people off that potentially lucrative project. This rankled Sinofsky badly—he had led the Office organization for several years before taking on Windows—and it is curiously he was unable to convince Microsoft CEO Steve Ballmer to step in and force Office to fully support Metro. After all, Ballmer wasn’t interested in bringing Office to the iPad, and he did go on to block that release. But without a new Office for Windows RT—the team got a modern version of OneNote, and a few small utilities only—there was no other choice: Windows RT would include the Windows 8 Desktop, so users could run a bundled version of the desktop Office 2013 app suite.

This compromise, combined with other design decisions, would undermine Windows RT and Windows 8, not to mention the Metro apps platform. And so Microsoft quickly killed RT and redesigned Windows 8 in a spate of post-release updates designed to return the system to its normal desktop focus.

When life hands you compromise, you make lemonade

But before any of that could happen, Sinofsky and his Windows team tried to counter Apple’s bifurcated world view with a more unified vision of personal computing. Where Apple presented a world in which the iPhone, iPad, and Mac were distinct products with distinct use cases, Microsoft saw them all as a single platform. A platform called Windows.

This was an interesting approach. And given enough time, it might have worked. The problem was one of timing: The Windows team was working on Metro and the other changes in Windows 8 for PCs and tablets, while a separate team created and then rapidly iterated on Windows Phone, Microsoft’s belated response to the iPhone and Android. Ideally, these efforts would be combined, unified. But they each addressed specific competitive concerns, and had different priorities, and each were on separate tracks initially.

With a late 2012 release date locked in for Windows 8 and RT, there wasn’t enough time to bring Windows Phone into the fold. For the initial release, Windows RT and Windows 8 would provide a Metro platform targeted tablets and traditional PCs. This platform would offer a single user experience that scaled to—was tailored for—each form factor, and a new mobile apps platform backed by a new mobile apps store that would let developers create reliable, safe, efficient, and modern apps that could run on both Arm and x86/x64 PCs and devices. The desktop and Metro worlds would coexist in Windows. And in a future release, the Metro platform would come to Windows Phone as well, completing the unification.

The problem for Windows RT was that the Desktop compromise resulted in a product that looked and worked like a mainstream version of Windows but couldn’t run any traditional Windows apps. RT came with a mostly complete version of Microsoft Office 2013 (minus the battery-hogging Outlook, at first) and whatever desktop utilities users also got with Windows 8. But customers could not download or install any other traditional Windows apps. The Arm chips on which these devices were built weren’t powerful enough to emulate x86/x64 apps, but Microsoft rationalized that Metro mobile apps were the way forward.

They were not. But this isn’t the only reason Windows RT failed. Customers rebelled against the unfamiliar full-screen user interface that was common to Windows 8 and RT because it was so counterintuitive, poorly designed, and a poor fit for traditional PCs. And as Microsoft quickly pushed to reverse those changes after their initial release, a tablet-only version of this system that couldn’t even run mainstream Windows apps just didn’t make sense. Few developers created notable Metro apps of any kind, helping to seal its fate.

Microsoft’s hardware partners also played outsized roles in undermining Windows RT and scuttling any chance that this new Windows on Arm could succeed.

Intel got the ball rolling in December 2010, when Microsoft informed it that it would soon unveil the Windows on Arm efforts that would result in Windows RT. To soften the blow, Microsoft CEO Steve Ballmer and Sinofsky told Intel CEO Paul Otellini that they also planned to promote an Intel system-on-a-chip (SoC) design that vaguely emulated the packaging advantages of the integrated Arm chips.

“Paul, you’re not going to view this favorably,” Ballmer told Otellini.

Intel’s CEO was mostly concerned that Microsoft or its silicon partners—Nvidia, Qualcomm, and Texas Instruments were all jockeying to be the first to deliver an Arm chipset for PCs—would emulate x86 code at the hardware level, a violation of the chipmaker’s intellectual property rights, its jealousy-guarded crown jewels. There, Microsoft had good news: It had ported Windows to Arm, but it would not emulate x86 code in any way.

Otellini then asked about Office, knowing how key this app suite was to the success of the platform. At the time of this discussion, the fate of Office was still up in the air—but it was internally “frustrating” to Sinofsky for the reasons noted above—and so Ballmer and Sinofsky simply told him the truth, they didn’t know yet. This was likely seen as a positive signal by Intel, and since Microsoft would present its Arm port as a shift to “system on a chip” (Soc) hardware designs that would also include the Intel Atom, Otellini seemed mollified.

But Intel had learned its lesson from Microsoft’s adoption of AMD’s x64 extensions. It would go on to undermine Arm at every step.

Technology demonstration

Microsoft unveiled its work bringing Windows to the Arm architecture at CES 2011 in Las Vegas. There were two reals, one soft and aimed at enthusiasts, and one public.

The first was a private press event, during which Sinofsky told us that the Windows port to Arm was done in part to support iPad-like Windows slate devices. But there were bigger plans for the future: The Arm SoC designs it was working with were so tiny—they could fit on a fingernail, he noted—this shift would bring Windows to a far broader range of new form factors.

He also said that Microsoft was working with three Arm-based silicon partners—Nvidia, Qualcomm, and Texas Instruments—as well as AMD and Intel, which would offer x86-based SoCs of their own. And that the resulting systems would be “tiny and power-efficient” and provide all-day battery life while being powerful and compatible with PC software.

The public unveiling was more momentous: It occurred during the CES keynote address, delivered by Microsoft CEO Steve Ballmer, who had replaced his predecessor, Bill Gates, as the headliner two years earlier. He droned on for almost an hour before getting to this news, but he had good reason. It was a stressful time for the software giant, which was ceding ground in the consumer space to Apple and other companies at a fairly rapid clip, despite the incredible success of Windows 7 on PCs. Ballmer did what he could to project leadership in this space. He highlighted Microsoft’s ongoing work with Xbox, its best consumer brand, which was temporarily riding high on the briefly successful Kinect add-on and expanding into music and video services. And the company had just launched Windows Phone 7, which, contrary to its version number, was the first generation of a new smartphone product line.

But Microsoft would always be the Windows company to consumers, and with two years into the “amazing success” of Windows 7, Ballmer said, customers increasingly expected to see Windows on modern new form factors. Whether that was true was open to debate, but this comment clearly spoke to Microsoft’s intent to counter the Apple threat with its most versatile software platform. And after a brief demo of some new Windows 7 PCs from its partners–and an even briefer look at Microsoft’s gigantic Surface, then a table (not tablet, but table) form factor PC–Ballmer finally dropped the bomb.

“The next version of Windows will support systems on a chip architectures from Intel, from AMD, and from Arm,” he started, notably slotting Intel into the pole position, despite its SoC designs being horribly inefficient and underpowered, and not the real focus for Microsoft. “This is about enabling a new class of hardware, and new silicon partners for Windows, to bring the widest possible range of form factors to the market. We’re making the announcement now to allow all our partners to work together and build on this innovation. Windows has always been, and will continue to be, about the breadth of hardware and software applications and innovation.”

The talk was light on details in some ways, but rewatching it today, it’s clear that Ballmer subtly differentiated Windows on x86 and Windows on Arm by noting that the former would “including support for native x86 applications.” Meaning, all Windows applications. And he was correct in pointing out that the shift to SoCs, which was really about the shift to Arm, was important to Microsoft: While Apple created a different software platform each time it came to market with a new device, Microsoft would adapt Windows for these new types of devices.

It was a great message, and it still resonates today: Microsoft planned to bring all the advantages of (Arm-based) mobile devices—battery life, reliability, thin and light form factors, and so on—to the PC, but without compromising on everything that made the PC great, including the “basics”—printing, peripheral compatibility—that wasn’t then available on mobile. It was breadth and depth, he said. No compromises.

Ballmer then brought out Mike Angiulo to demo some of the early work Microsoft had done to bring Windows to Arm. This was, Angiulo said, “a technology preview,” not a full look at the next Windows, which he confirmed would include a “new user interface” and new features. Notably, he showed off an Intel system first, a “low-power Atom SoC,” on a pre-production development board connected to a keyboard, mouse, and display. The chip was a bit bigger than a fingernail, but the motherboard was smaller than most of the chips that Intel was then shipping. It seemed like a big step forward, and Angiulo also noted its compatibility with existing software applications was “notable.”

Angiulo quickly moved on to development systems from its three new silicon partners. The first, interestingly, was Qualcomm: It was called Snapdragon, he said, and he showed that the Windows user interface was identical to that on the Intel system and with identical features, including the command line and other advanced functionality. This wasn’t virtualization, he said. It was full Windows running natively on Arm for the first time.

He then moved on to a Texas Instruments development system, on which he launched Word—running natively on Arm—and printed a document to an Epson printer. Microsoft had to recompile the printer driver to run natively on Arm, Angiulo said, but that work was successful, as evidence when the printer on stage spit out the document he printed. Ballmer held up the page to the audience with a smirk. “I’m a real PC,” the document read, the subtle dig at Apple complete. “Of course, I can print.”

The third Arm development system was from an Nvidia, and it is perhaps an interesting historical point that this is the only Arm-based architecture for PCs that made it to market in the coming wave of Windows 8/RT devices. Nvidia was known for high-performance graphics, Angiulo pointed out, and its Arm-based Tegra platform was particularly well-suited to apps that could take advantage of hardware acceleration, like Excel and video players.

“What you’ve seen here today is Windows, real Windows, running Office, devices, high-performance video, all running on the next generation of SoCs,” he concluded. The entire demo lasted barely 5 minutes, but it had made a lasting impression. If Microsoft could pull this off, the next Windows, running on Arm and x64, might have what it would take to blunt the impact of Apple’s iPhone and iPad, and a resurgent Mac. It was, Ballmer said, “the magic of software” redefining what was possible.

It wasn’t going to work.

Interestingly, I correctly opined the day of this announcement that “if this thing [Windows on Arm] is wildly successful from a technical standpoint, what we’ll get on the other side is simply Windows. It won’t seem particularly revolutionary to actual users.” But that was only true because of the secret compromise that had been made internally at Microsoft: A distinct Windows RT offering with no Windows Desktop that worked on both phones and tablets might have had a chance against Apple. But the system Microsoft eventually shipped was DOA, and it quickly failed in the market and then disappeared.

“I have to ask: If you perform an impressive engineering feat and no one buys the resulting product, is it still an impressive engineering feat?” I wrote. “More to the point, is Microsoft doing something because it can be done, or does it do what its customers need and want? If the answer is the former, please, do keep talking vaguely about engineering and do keep pushing Windows to every technological nook and cranny you can find. But understand, too, that when you wake from this engineering stupor, much of the world will have moved on to other companies and products.”

For his part, Steven Sinofsky understood the challenge. PC sales had cooled heading into late 2010. Android tablets appeared in the wake of the iPad, with Google rapidly iterating on new OS versions to address the needs of this new form factor. PC makers were more interested in cheap netbooks than premium PCs that could take on the Mac. And that was going to turn into a serious problem.

More challenges

With the Arm reveal out of the way, Microsoft got to work. But so did its most important partners. Publicly, Intel undermined Arm early and often, and went so far as to demote Windows in its presentations as an alternative, of sorts, to Linux, in what was clearly retribution for the Arm pivot. It also stepped up its efforts to keep its PC maker customers off Arm—as it did to keep them off AMD—by escalating the financial incentives to stick with its chips.

The lobbying worked. PC makers had long operated on tight budgets in a market with tiny, almost imperceptible margins. They used their sales volume to demand the lowest possible prices from component makers. And when those needs couldn’t be met, they simply moved onto to a cheaper supplier. PC quality and reliability was abysmal as a result, and when the netbook arrived with even lower costs and quality, they seized on it immediately, undermining their own businesses.

Prodded in part by Intel, PC makers saw Arm as another opportunity for cost savings: They could put less expensive Arm-based chips in inexpensive netbook hardware and save even more money. But that wasn’t the plan: Microsoft didn’t want to see Arm relegated to the expensive premium end of the market, but it also wouldn’t allow it to be undercut by the netbook. Worse, for the PC makers, Arm-based systems were integrated, and PC makers couldn’t shop around for different components. So their chances for differentiation—mostly cost differentiation—were minimal.

Privately, Microsoft’s PC makers would not stop complaining. They were not happy in the slightest with Windows on Arm. And there was Intel, making the choice obvious. What could Microsoft offer in return?

Microsoft tried. It worked with its three new silicon partners to try and offer differentiated Arm hardware systems. It created class drivers for common peripheral types to ease the transition costs. It designed and implemented its new Windows 8/RT user interface and Metro mobile apps platform, tailoring each to work as well as possible on traditional PCs with keyboards and mice and with new slate and tablet designs with multitouch and smart pens. And it pushed for quality over quantity, arguing that selling fewer well-made premium PCs with unique designs would be more profitable in both the short and long term than pushing forward with cheap, me-too netbooks.

It didn’t work. And not helping matters, Microsoft announced its own PCs in early 2012, including one built on Arm. That was the final straw, and in the end, the PC makers made their point clear by almost universally ignoring the Arm-based Windows RT. By the time the next version of Windows launched in October 2012, every PC maker was on hand with new Windows 8-based PCs in traditional form factors, most of them running on Intel chips. Only two PC makers each offered a single Windows RT-based slate or tablet PC design each at launch. And only one more model arrived by the end of 2012, not including Microsoft’s PC. The message had been received.

Road to fool’s gold

Microsoft spent the next 18 months explaining its plans for Windows 8 and RT, and it delivered a handful of key pre-release milestone builds to the public starting in late 2011. But Windows RT was unavailable for testing because there was no hardware. Sinofsky penned an 8600-word missive on the importance of Arm in February 2011 timed to the Mobile World Congress trade show. And then Ballmer green lit an ill-advised push for first-party PCs—now called devices thanks to Arm—that would see Microsoft repurpose its Surface brand for a new lineup of tablet PCs. The first of which would be based on Arm and run Windows RT.

The problems were many, but the design compromises forced on Windows 8 and RT by the Office team triggered an incredible pushback from customers and partners. At the first public demo of the new user interface in June 2011, even the dullard co-hosts immediately saw the problem with the system’s “two modes,” Metro and Desktop, correctly noting that the shift between the two was “jarring.” It was a downward spiral from there: “The biggest shift in Windows,” as Microsoft called it, was too big a shift for its customers, who were too stuck in the familiarity of the past to accept a full-screen Start menu on their traditional PCs.

The new apps were even worse. Metro-style apps were designed for touch, not a mouse, and they had large buttons and other UI controls that looked out of place on traditional PCs. They were derided as “Fisher Price”-like in look and quality, and while Windows 8 users could at least Alt-Tab to the blessed sanity of the Desktop and get to work in real apps, Windows RT users were stuck in Metro hell. The problems here were many as well, but the biggest is that developers had long since moved on to native mobile platforms from Apple and Google, and to web apps. Windows apps—new, old, whatever—were no longer in demand, they were legacy, and in maintenance mode. The immature Metro platform didn’t entice anyone to believe otherwise.

The problems snowballed. By the time Windows RT and 8 launched, Sinofsky was already gone—he unexpectedly quit the company in defeat rather than clean up the mess he had orchestrated—and Microsoft shifted Windows to a new “rapid release” model in which it would iterate quickly instead of waiting three years for the next release. It addressed the issues with its compromised platform, reversing the most egregious of them over a series of annual updates that started with Windows 8 and RT 8.1 in 2013. And Microsoft’s big push to compete with its partners in the PC market with Windows RT and Surface RT was an immediate failure: Microsoft took a nearly $1 billion write-down on the product, shipped one revision a year later, and then gave up on Windows RT altogether.

The software giant wouldn’t ship a new Windows on Arm release or another Arm-powered PC for almost a decade. But by that time, everything had changed. There was a new Microsoft, and a new Windows team with new leadership, new ideas, and new challenges, and they would go on to complete the PC, tablet, and phone integration work Sinofsky had planned. This second swing at making a go of Arm wasn’t financially successful, but it wasn’t a complete disaster, either. And there were some major technological advances suggesting a brighter future was possible.

Perhaps Microsoft would figure out a way to make Windows on Arm work after all. Some day.