This is Windows 12 (Premium)

This is the year that AI (artificial intelligence) breaks into the mainstream thanks to online tools like ChatGPT, DALL-E, and the like. And if nearly 30 years of covering the personal technology market has taught me anything, it’s that we’re going to be inundated with AI-based services and apps, creating a clear dividing line between the past and the future. Everything is about to change.

Hyperbolic? Not at all: AI-based solutions had been creeping into the public consciousness over the past two years or so, thanks in part to the hybrid work improvements we’ve seen in productivity software like Microsoft Teams and Zoom, and in related hardware advances in webcams, microphones, and presence detection sensors, since the start of the COVID-19 pandemic. But now the Big Bang has occurred, with ChatGPT being the AI shot heard ‘round the world. Google was caught so flat-footed by this revelation that it dusted off its cofounders, reconsidered projects that previously were found to be too creepy (or, knowing Google, too good, which would curtail ad revenue), and has issued a “code red” internally. Now, it will pull out the stops to show the world that it can compete on an even footing with Microsoft and OpenAI.

I have little doubt that Google will be competitive. But the difference between Google and Microsoft, in this specific case, is notable: AI is an existential threat to Google, so any failure on its part could literally kill the company, as could regulatory intervention. But for Microsoft, AI is just a potential growth market. Microsoft is diversified enough to come out of this just fine should Google somehow defeat it.

So how will Microsoft integrate AI into its products and services? The AI Big Bang was so unexpected and so sudden, we’ve had little chance to calmly consider how this change will impact the Microsoft stack. But we have a few details.

We know that Microsoft plans to integrate ChapGPT capabilities into Bing Search, for example, but that’s mostly a curiosity at this point, given how utterly inept and unpopular Bing is today. Bing will need more than ChatGPT to defeat Google Search, and I just don’t see that happening. Consider how Apple Maps, despite years of improvements, has done little to unseat Google Maps on the iPhone. It’s not enough to be very good when the product you’re fighting is the standard. Inertia is a powerful force.

We also know that Microsoft plans to commercialize AI for third parties in the Azure cloud and add ChatGPT-like AI capabilities to “every Microsoft product,” in the words of CEO Satya Nadella. I described that latter bit as AI being “infused” across the Microsoft stack, but at the time I wrote those words, I was mostly thinking in terms of Microsoft 356, or what we still think of as Office. After all, that is the center of Microsoft’s productivity offerings.

But what about Windows?

Windows today is an interesting problem for Microsoft. It’s a legacy desktop platform in an era of mobile and web computing, but it also has a massive user base of over one billion individuals and is responsible for many tens of billions of dollars in revenue each year. And while the PC market is now experiencing what should have been an all-too-predictable downturn in the wake of the pandemic-era buying spree, this market isn’t going anywhere. And not just because of inertia. As it turns out, the PC is still the right tool for certain key productivity tasks. It’s just easier—and better—to perform certain functions on a PC as opposed to a tablet or a phone.

The problem, of course, is that Windows is the only major Microsoft product to not fit into the cloud computing model that the rest of the company rallied around after Ray Ozzie’s two memos, “The Internet Services Disruption” (2005) and “Dawn of a New Day” (2010), which accurately foresaw the rise of cloud computing and the diminishing of the PC market several years before either was obvious to the rest of the world.

Since those memos, Microsoft has tried to adapt Windows to a changing world. Windows 8 was an overreaction to the iPad. Windows 10 returned the focus to the desktop while adapting the software platform to be cross-platform, but only within Microsoft’s own devices. And Windows 11 … I don’t know what the point is here, it’s just pretty with not much else new to offer. Microsoft has also brought Windows to the cloud with Windows 365, but it’s fair to say that this offering won’t replace traditional Windows installs for most and is simply another way for customers to acquire Windows licenses.

But as I wrote in Maybe AI is the Next Wave (Premium), Windows chief Panos Panay appeared onstage with AMD CEO Lisa Su at CES and vaguely touted something that sounds an awful lot like a Windows 12. It was an awkward love fest for AI, which requires a neural processing unit (NPU) on the PC’s chipset to run effectively. Qualcomm got there first in the PC space with its Snapdragon processors for PCs, but no one uses those. Intel is bringing M.2-based AI accelerator modules to its 13th Gen Core desktop chipsets (Raptor Lake), and it will provide native NPU capabilities with its 13th Gen Core mobile chipsets (Meteor Lake) this year. And AMD is bringing NPUs to its chipsets too, starting with the one that Panay and Su were babbling about at CES.

An NPU is to AI and machine learning (ML) what a GPU is to graphics: it dramatically accelerates the process using custom-designed hardware, and it frees the CPU for the more mundane computing tasks that CPUs perform. And if this sounds at all familiar, it’s likely because Apple has been marketing the hell out of its own NPU, which it calls the Neural Engine, for years. Because of course it does.

Apple didn’t invent the NPU, but there is a key innovation to its work: before Apple added the Neural Engine to its chipsets, AI and ML tasks would require roundtrips to some datacenter in the cloud. This meant that the device had to be online, which, granted, it would be most of the time, but it also incurred a speed penalty. Apple’s innovation—which was arguably born out of the fact that it does not have the type of AI resources in the cloud that Google and Microsoft have—was to put that processing power right in the device. This fit in nicely with its privacy marketing—what happens on iPhone stays on iPhone—and it gave its products a nice competitive advantage.

What’s interesting to me about that is that Apple basically proved that this kind of configuration was advantageous, and so other platform makers—again, Google and Microsoft—began working on their own local NPUs too. And Google began marketing how many things it could do right on a Pixel or whatever other device without needing the cloud, just as Apple does. But the combination of local and cloud-based AI processing is inarguably the best configuration of all. And this is where Google—and now Microsoft—can potentially leapfrog Apple.

This brings me back to Windows 12.

In the wake of Satya Nadella’s comments about “AI-ing all the things,” as I called it, I speculated on Windows Weekly that perhaps Microsoft would take the artificial hardware limitations of Windows 11 and formalize them into something more concrete for Windows 12. That is, what if Windows 12 required an NPU? Those billion-plus PCs out in the world would suddenly be obsolete. And if you wanted whatever richness Windows 12 provided, you would need to upgrade the hardware.

This is a dream scenario for Microsoft and its PC maker partners. And while I immediately poo-pooed my own idea, arguing that it was more likely that Windows 12 would work on new and existing PCs but would simply handle AI/ML faster—OK, much faster—on new NPU-equipped PCs. But the more I think about this, the more I’m coming around to the notion that Windows 12 will effectively require an NPU. It just makes sense.

And the reason it makes sense is all over the news: normal people are flooded with stories about new AI products, services, and advances almost every day now. They see what’s possible with ChatGPT, DALL-E, and other solutions. And if Microsoft could somehow market some incredible AI capabilities in Windows 12, those people may suddenly discover that upgrading to a new PC isn’t just a great idea, but an obvious idea.

I’m also wondering about the need to add AI to all of Microsoft’s products and services and am further wondering if that was just a feint. Aside from the obvious—it would be incredibly hard to add AI capabilities up and down the Microsoft stack, let alone do so quickly—I feel compelled to point out that Microsoft tried that before. With .NET. And that initiative didn’t just fail, it divided the company internally in a way that still exists today.

Back in 2000-2001, Microsoft was going to “.NET all the things.” Products like Windows and Office would be renamed, to Windows.NET and Office.NET, and .NET technologies would be infused on the client, on the server, and in what we now call the cloud (back then it was just services). But this idea imploded despite the fact that it was championed by Bill Gates, and it imploded quickly: when Windows XP arrived in October 2001, it didn’t just not have .NET in the name, it didn’t have .NET in the product either. The reason? .NET wasn’t ready and no one could wait around long enough for that to happen.

But there is an easier and perhaps more logical way to add AI to the Microsoft stack. And that is to do it in the one place where Microsoft controls the client and can require an NPU in the same way that it now requires a TPM. And that place is the PC. That place is Windows.

Look, I understand that Microsoft 365/Office exists outside of Windows, but I will make the argument that most work gets done via that platform in Windows, and not on a phone or tablet, or in a web browser. And that the logical place for Microsoft to pivot on AI is not in the cloud—though that will happen—but rather on the client, because it can take advantage of that ideal configuration of local and cloud-based AI processing.

I also understand that this flies in the face of Microsoft’s strategy under Mr. Nadella, and that the days of “Windows only,” “Windows first,” and “Windows best” are long over. But the advent of the AI era will require new thinking. Or, in this case, perhaps classical thinking. And a “Windows best” approach here actually makes sense.

There’s some interesting symmetry there, too. Those who remember the document-centric days of Windows 95 might see a shift to an AI-centric world in which we ask Windows to perform some task, and it uses some combination of Edge, Bing, Word, Excel, and whatever other Microsoft tools to pull together a hybrid document of whatever kind, and the user wouldn’t need to know much if anything about any of those apps or services to make it happen.

Today, Windows is basically just plumbing or, in the case of Windows 11, plumbing with a pretty face. But we see the beginnings of a more sophisticated product in the hybrid work functionality in Windows Studio Effects, which, yes, requires an NPU. This small thing shows how AI could make Windows relevant again. Not just from inertia or for performing legacy tasks. But for getting anything done. It could become the key to how we are productive at work in the future.

Whatever happens is going to happen quickly. We have Build in late May and presumably Ignite in November, but I expect Microsoft, like Google, to work outside those artificial milestones and announce advances again and again throughout the year.

Get excited. I feel that an AI-infused Windows 12 could be the biggest of those announcements.