Ask Paul: June 5 ⭐

Happy Friday! This was a particularly eventful week, especially for those interested in the future of Windows. Which, come to think of it, reminds me of a tagline from long ago. In any event, let’s jump right in: As is so often the case, many of you seem to be interested in and wondering about the same things as me this week. And there are still so many questions. Starting with …

💻 There are still far too many N1X questions

jrzoomer asks:

Paul, does the NVIDIA N1X hardware architecture feel like a classic high-performance gaming chip that Nvidia is simply rebranding with AI marketing to fit the current moment? Leaked roadmaps for the N1X/RTX Spark date back to at least 2024, back then local inference was touted as the next big thing but that never panned out–no one is running local AI workloads on laptops, everything is run on cloud. Given that the N1X is essentially a Blackwell RTX 5070 GPU (same exact 6144 CUDA cores) and a 20-core Arm CPU fused together (a weak one at that, as early benchmarks show the N1X CPU performing similarly to Apple’s M3 Max from 2023), its real strength seems to just be traditional gaming performance. So are these essentially gaming laptop SoCs?

Hm. I think the major pushback on this is the Arm architecture at the heart of the chips. Games are almost universally non-native on Arm, so they run under emulation, and that will always incur a hit. On Snapdragon X-based PCs, this experience can be good or even excellent, but key AAA titles like those in the Call of Duty series don’t run at all, let alone slowly. Nvidia showed off some select games playing on the new hardware at Computex, and they of course looked great. And while it’s reasonable to expect this experience to be better on these systems than on Snapdragon X because of the GPU, it’s never going to be perfect.

Unless, of course, game makers begin targeting Arm as well as x86. And the hope here, no matter which architecture you prefer, is that Nvidia’s influence triggers that important next step. Today, any mainstream premium laptop with a current-generation AMD or Intel processor can play most games very well, and higher-end SoCs (Panther Lake, for example, in the upper tiers) are terrific. So gaming will be fine on Nvidia-based laptops in the short term, I’m sure; there’s a lot of work in Windows to help with that too. But we need the games to be native. And until that happens, these N1X-based laptops won’t be top-tier experiences.

I feel like the real point of this hardware is to move the needle on local AI, and while the Copilot+ PC thing hasn’t resonated with just about anyone, this is a rare example of Microsoft being ahead of the curve. Cloud-based AI continues to improve at a dramatic rate, but so too does local AI. And while it’s unlikely those capability lines will ever cross on a graph, it’s always been obvious that they will be “good enough” for most mainstream usage cases soon, and there’s always the hybrid model where those who are online can switch, hopefully seamlessly, between local and cloud AI if needed.

The problems are familiar, of course. This is a classic chicken-and-the-egg scenario on the one hand. But there’s also no killer app, no one big thing that anyone can point to as the reason why everyone wants this. I think that’s going to change, as a local-only AI chatbot/agent orchestrator that’s as powerful as what we use in the cloud is not that far away. But in the meantime, we just have individual features in select productivity and creator apps, for the most part, that work better (faster, more efficiently, however you want to measure that) when they can use an NPU (or GPU). But this is transparent to the user; it’s not like a sparkly notification pops up each time to tell you that you just saved half a millisecond or 1 watt of electricity, or whatever. It just happens, and no one notices. And then everyone complains about an NPU that sits dormant in their PC doing nothing.

At Build, Microsoft revealed that it is bringing some local Windows AI APIs that currently work only against an NPU on Copilot+ PCs to normal PCs, where they will work against a CPU or GPU depending on the API. This means that the Copilot+ PC platform/brand will have to change, for starters, something we’ve basically been asking for since we first heard about this thing. It’s insane to require an NPU when so many customers, especially developers, have powerful dedicated GPUs that are far more powerful (and, yes, much less efficient) than the NPUs in Copilot+ PCs. I suspect that Nvidia’s entry here is either what triggers that change or is what Microsoft was waiting for.

For the all compatibility complaints we still get about Windows 11 on Arm–almost always from people who don’t even own Snapdragon X hardware–the only legitimate, mainstream issue or qualifier is gaming. If gaming on a PC really matters to you and you only have one PC, then Windows 11 on Arm isn’t going to cut it. Perhaps Nvidia changes that. I do hope so. The sooner the world kicks x86 to the curb, the better it will be for everyone.

🪟 Scouting for a future that has to arrive someday

Will asks:

With the hinted at upcoming “super app”, Microsoft releasing its own models (and publicly distancing itself from openAI), and all of NVIDIA Spark news, do you think we will see a major shakeup with Windows and Copilot branding later this year and could NVIDIA be pushing this? It seems like a new major Windows release, aka Windows 12, would be one way to reset the branding going forward. The MS Dev box video did show a slight UI change with a new silver Windows start logo.

We’ve been talking and speculating about Windows 12 for years now, and I’m a little nervous about even predicting when that will happen, assuming it does, or what changes have to occur for that to happen. But tied to the above question, Copilot+ PC has to change this year, at the very least. And Windows 12, or whatever it’s called, is definitely happening in the same time frame. (He says more confidently than is warranted.) There are all kinds of hints to suggest all that. And though I keep waiting for clarity from Microsoft, for now we can just piece together what we know.

Timing-wise, Microsoft typically released a major new Windows (11) version each fall, usually in October.

Microsoft has said that this year’s annual release will indeed be Windows 11 version 26H2.

Microsoft in early 2026 released a Windows 11 version 26H1 that is “scoped” specifically for new Snapdragon X2-based laptops. It has said that users with these PCs will not get 26H2 as an upgrade later this year but will instead have some other, still unspecified upgrade path.

Its reluctance to provide a name (or time frame) for that upgrade path is, I think, telling. And I think the timing of the new Nvidia hardware, which, like Snapdragon X2, is based on Arm, is not coincidental. These things will all arrive in the fall, I bet. And they will either run Windows 12, if Microsoft goes with that brand, or some different Windows 11 version whose name is TBD. (It’s not difficult to imagine Nvidia laptops shipping with an early version and Snapdragon X2 getting in 2027 after it matures a bit, similar to the Copilot+ PC situation on Arm and then x86.)

One of many questions here is whether this new thing, Windows 12 perhaps, replaces Copilot+ PC or acts as a sort of upgrade to that spec. On the low-end, Microsoft is working to reduce the memory and resource usage in Windows 11, which still has a base configuration requirement of 4 GB of RAM and 64 GB of storage. But Microsoft had previously raised the hardware requirements with Windows 11 to include a TPM 2.0 chip and whatever else, and then it dramatically raised the system requirements for Copilot+ PC to 16 GB of RAM, 512 GB of storage, a 40+ TOPS NPU, Windows Hello ESS, and more. So perhaps that (with or without some changes) becomes the Windows 12 baseline. Or Copilot+ PC 2.0. Or Windows 11 Copilot+ Edition, God help us. Whatever they call it.

It’s confusing, but mostly because we’re getting bits of information in isolated ways and have to infer or guess what’s happening. At some point, Microsoft will come clean on its plans. And with these Nvidia announcements, we now at least have a timeline. Plus, there will be leaks: Someone will see a Windows version number or name somewhere on some prototype or early version of whatever laptop at some point between now and then.

Microsoft almost renamed Windows 11 version 23H2 to Windows 12, the idea at the time being that AI capabilities would be the dividing line. But it was smart not to do that: Copilot has been a disaster in Windows 11 since the beginning, and its other AI features are either ignored or even feared by most. Copilot+ PC was in some ways a strong effort to push awareness of local AI, though the real advantages of those laptops, especially those based on Snapdragon X, are tied to performance, reliability, efficiency, and battery life. And the whole industry has tried to get agents off the ground for two years, with this technology only now starting to come together. So you could easily make the case that AI–local AI, agentic AI, and whatever else–is still the dividing line between Windows 11 and Windows 12. It’s just taken a few years to come together.

To be clear, this is all speculation at worst and educated guessing at best, but you have to work with the information you have. And then adapt when you learn more. I am eager to learn more. But there is something else coming in late 2026, something that is not Windows 11 version 26H2. And whatever it’s called, however it is rolled out with new Nvidia PCs and new and existing Snapdragon X2-based PCs, it will be different in some ways than “normal” Windows 11. Calling that Windows 12 seems to make sense, but that doesn’t make it real.

I wonder, too, about the current craze in low-end PCs that don’t meet common sense low-end requirements for Windows 11, let alone a Copilot+ PC. A MacBook Neo with 8 GB of RAM only makes sense for users who spend most of their screen time on other devices. But Windows 11 is currently less efficient in hardware usage than macOS, and it’s not clear if the current “pain points” work will completely close that gap. What is clear is that we’re about to be flooded with (relatively) low-cost, underpowered PCs that will work like a Neo in that they’ll be fine for those who spend most of their screen time elsewhere, as secondary devices. But anyone who works all day on a PC will need more than that.

Tied to all that, the component crisis will end someday. Many will remember that the last time gasoline surpassed $4 a gallon in the U.S., car buyers demanded smaller, more efficient cars, and automakers delivered exactly that. But prices fell again, and then Americans went right back to buying bigger vehicles, especially SUVs. Obviously, the current situation is even worse for that market, but we’ve seen amazing advances in electric cars and hybrids since then, so no one is shifting strategies again. I do wonder if there’s a parallel in the PC market in which Arm chipsets play the role of electric and hybrid vehicles. In the same way that most are better off with a Snapdragon X/X2 laptop than anything x86-based; perhaps the Snapdragon C can play that role in the low-end market and not turn this into another generation of netbooks we will all regret, and Nvidia can turn Arm into an ideal replacement for gaming PCs and portable workstations. We’ll see.

As for this super app, when I first heard about Microsoft Scout right before Build, I figured that had to be it: This is Microsoft’s Gemini Spark, a personal AI agent that takes us from Copilot to Autopilot, in Microsoft’s words. But it’s not: During the Build keynote, Satya Nadella said that a Copilot super app would launch this summer before even discussing Scout. Standing in front of a screen that displayed the app’s three rumored components–Chat, Cowork, and Code–he said that it would bring “coding to all knowledge work,” which seems like a curious need. (I also think it’s notable that he had absolutely nothing to demo or preview during Build.) And I suspect “this summer” will be a very early preview only, but I also wonder if the goal is to include this as a core part of what I keep calling Windows 12. The clock is ticking. (The super app will also include Autopilots, with Scout being the first.)

I have nothing but questions, basically.

📝 A golden age for software development

iantrem asks:

We seem to have had so many false dawns with Windows development since, well since they tried moving from Win32, how confident are you that the changes outlined with WinUI are going to make this change finally happen?

Not at all confident, with the one caveat I mentioned in Developers, Developers, Developers! ⭐: In a prerecorded session that last less than 30 minutes, the legendary Chris Anderson did more to overcome my PTSD with this technology than I ever thought possible. There appears to be adult supervision here, and where previous efforts to fix the Windows App SDK/WinUI fell flat because there was no one in a position of power to make it actually happen, this feels different. Perhaps I’m deluding myself. But I think you can compare this with the pain points stuff in Windows 11. Something positive is happening.

Will this move the needle out in the world? That’s a tough one.

WPF was good enough to thrive with .NET developers long after Microsoft gave up on it. There was that modernization effort two years ago and then nothing since. Is that enough for most legacy apps, or is there some incentive to upgrade to Windows App SDK? We’re going to find out.

But maybe that doesn’t matter. It’s pretty clear, because Anderson said it explicitly, that one of the goals of the current work is to make Windows App SDK/WinUI better suited for this new generation of agentic AI coding. And if you think about the vibe-coding experiment I just did with the Windows Development Skills, I used whatever AI to generate a modern Windows 11 app based on the Windows App SDK and WinUI, and I didn’t need to know how to code to make that happen. Developers and enthusiasts like me can debate languages and frameworks and pick whatever it is we prefer, but with AI app generation, it doesn’t matter what we think. Microsoft is making what it’s just calling WinUI a first-class technology for these agentic coding scenarios. If you care about this kind of thing, you can at least rest easy that this is native code, and the most modern way Microsoft has to make apps.

You’ve also, quite rightly, claimed that nobody should just develop for Windows. Do you think the REACT-style declarative development will be used to replace MAUI and finally give MS the cross-platform development system it’s been trying to build for years?

Anderson is clear in that video session that Microsoft has “no intention of building a new framework.” But you could argue that this claim is for Windows desktop specifically, whereas MAUI is cross-platform, though heavily mobile focused. In Developers, Developers, Developers! ⭐, I also mentioned Microsoft UI Reactor, an internal project to bring React-style declarative development to WinUI via C# (as opposed to using XAML), and most of that Anderson session is him showing this off to create desktop apps. But assuming this works and is accepted by developers, there’s no reason it couldn’t come to MAUI. I do like the idea of using C# to create cross-platform apps. (I also like the idea of using C#/XAML, as is the case with MAUI today, but the heavy mobile tilt there is the bigger issue.)

I do wish MAUI were more competitive with Flutter and whatever else. I’ve never really had the sense that it was ever going to get there, though. Like so much else in this space, it hit whatever milestone from a capabilities perspective and has just inched along ever since.

I’ve not found a specific talk where it’s mentioned, but a Visual Studio blog post this week suggests that !the VS/.NET teams may finally have a way for web developers to migrate webforms sites (from the early 2000s) to asp.net core, is the new declarative language the way they’ll do this?

Interesting. I always figured this would somehow fall under Blazor somehow, but the fact that there are still WebForms-based sites out there is incredible, given that this dates back to the same era as WinForms (i.e. 2001-ish). This speaks to the same types of modernization concerns that desktop developers may have coming from WPF (or WinForms) to WinUI. And it makes sense that agentic AI would be the way forward on the web, just as it is on desktop. But I don’t know anything about this. Perhaps as I continue working my way through Build sessions videos, I will come across it.

Speaking of which.

Finally, your developer background will be interested in the “Scott and Mark learn to…” talk that discuses the need for businesses to make sure that junior developers still pick up the coding skills that AI is trying to write for them as they’ll be the senior Devs one day. It’s an interesting (and obviously often humorous) follow up to Scott’s recent TED talk.

Definitely. That’s on my to-watch list, of course. I’m a huge fan of both of these guys and listen to their podcast routinely. As I always do post-Build (and other shows, like Google I/O), I queue up a bunch of session videos to watch, and there are always gems in there. I will spend time on that this afternoon and over the weekend.

🧑‍💻 Is it time to upgrade?

Alex Strickland asks:

Last year, I bought a Copilot+ PC with a Snapdragon X Plus chipset. After reading about the release of the new Snapdragon chipset, I’m wondering whether an upgrade is worth considering or if the Snapdragon X Plus will remain sufficient for the foreseeable future. I mainly use this laptop to tinker with AI models, do some Python coding, and play a few runs of Vampire Survivors. I’m not someone who needs the latest and greatest hardware, but I am curious about the lifespan and long-term usability of the Snapdragon X Plus.

Based on my experiences with three laptops, each running a different tier of Snapdragon X2 chip–X2 Plus, X2 Elite, and X2 Elite Extreme–I am confident stating that there’s no reason for a generation-over-generation upgrade, at least from a processor/performance perspective. In day-to-day use, these three laptops work nearly identically to all the first generation Snapdragon X laptops I have, including the OmniBook 5 with its base-level Snapdragon X that remains a favorite.

Two years into the Snapdragon X era, these laptops are still terrific, and they are still more reliable and more efficient and have better battery life than any x86 laptop. And I’ve yet to have some experience when I thought something had slowed down or wasn’t living up to more recent PCs in whatever ways. Nothing lasts forever, but I suspect this year’s low-level improvements to Windows 11 will only extend the lifetime of Snapdragon X/X Plus hardware.

The only caveat in there, I guess, is RAM, and the component crisis makes an upgrade to go from, say, 16 GB of RAM to 32 less enticing. So those things combined suggest that sticking with what you have isn’t just smart, but is completely viable and will remain so for the foreseeable future.

🔮 There can be only one (AI subscription)

Alex Strickland asks:

I also have one more question, if you don’t mind. With all the AI subscriptions available today, which one would you recommend? I currently use the free versions of Gemini and ChatGPT, and I have a Microsoft 365 Personal subscription. I rarely use Copilot except for help with grammar and sentence structure.

This is an ever-changing thing given how new AI models leapfrog each other, especially in certain use cases that may or may not be important to you. And it’s reasonable to believe that all the major AIs–Anthropic Claude, Google Gemini, Microsoft Copilot, and OpenAI ChatGPT–will be roughly identical capability-wise, as each seems to create all the same features and make similar leaps forward on similar schedules.

There are some wrenches out there in the form of outliers like Perplexity, DeepSeek, and whatever else. And also in the evolution of local AI, noted above, that will quickly become good enough for so many use cases. And then there’s the whole Little AI thing, tied to the above as much as it is to anti-Big Tech sentiment that is certainly understandable.

But leaving the caveats aside, the fact that the big AIs are roughly identical or will be, that actually helps matters quite a bit. You might argue, for example, that since you’re paying for Microsoft 365 Personal anyway, and that gives you whatever Office app access and OneDrive storage usage, that using Copilot makes a lot of sense. Those less beholden to the Microsoft stuff who want lots of online storage should look at the Google One/Gemini AI plans, as they offer 5 TB of Google Drive storage and Gemini is excellent.

I once wrote that I Will Not Pay for AI, the thought there being that I would pay for apps or services but AI is more a set of features than a product. I’ve since caveated that from a developer perspective, because even a $200 per month cost to a professional developer is an easily-justified expense. But the way AI is evolving, there’s a case to be made that, in the future, paying for AI capabilities that can replace apps or services can make sense. That suggests that standalone AIs like ChatGPT and Claude could justify their cost, depending on your needs.

(For example, let’s say that instead of paying for Microsoft 365 because you need Word, you use whatever free app to write and then use an AI to export that writing to Word document formats others will use. This is all very use case specific, but if the goal of AI is to save you time and money, these types of things start to make sense.)

Right now, and ignoring any Big Tech/privacy concerns, I feel that a Google One/Gemini AI plan makes the most sense in some ways because of all that online storage, the ubiquity of Google’s personal and work-related apps and services, and the general strength of Gemini across the board. And Apple is adopting it for Apple Intelligence, though we don’t yet know exactly what that will look like. But that’s today. Tomorrow, next week, next month, or whenever, things will change as they always do with AI.

Disclosure: Copilot was used to clean these questions.

LOL. Nice.

🤩 It’s astonishing that anything can excite me at this point, and yet …

helix2301 asks:

You and Brad talked a lot about build this week, and Windows Central had a lot of news about new Surface Laptop Ultra and Dell XPS 13. Microsoft wants more WinUI 3 apps across Windows 11, and Terminal is getting AI now. Plus, all coverage on your site which as always was great. Build my favorite show of the year. What were you most excited and impressed to see come out of Build personally?

Historically, Build–and, before that, the PDC–has always been my favorite Microsoft or industry conference. Of course, as the emphasis shifted from Windows (and, briefly, Phone) to the cloud, whatever excitement I had was usually beaten out of me to some degree.

But you hint at part of the reason these shows were always so good, and that’s true of this year as well. When Windows was the driving force behind whatever innovations were occurring within Microsoft or throughout the industry, it wasn’t just about software development or apps and then services; it was about hardware devices and new form factors and how technology would keep evolving and getting better. I have such clear memories of certain things, for example, like seeing a Microsoft demo of a Longhorn version of the Calculator app that could be stretched to any size and would rescale as it did so, and how incredible that was compared to what had been, to that date, a world of static bitmap graphics.

Build 2026 was mostly about AI agents, of course. But there was so much news tied to Windows, and across the board, with new hardware being a key component. I can’t recall the last time that happened. Two years ago, it was exciting to me and probably 17 other people that Microsoft was bringing WPF back from the dead. But the things that are happening today are much broader and will be much more interesting to far more people. We’re never going back to a past where it was all Windows, but this has been a fun reminder that it can still matter.

I was delighted that the Build 2026 keynote started with Windows. That the first person on stage not named Satya Nadella was Kayla Cinnamon, who oversees Terminal; I’ve spent a lot of time with command lines this past year, and I am a big fan of her, her work, and her YouTube channel, as it’s full of excellent tips for Terminal users. And I love that there are new hardware platforms aimed specifically at advancing Windows PCs. If you think about it, the vast majority of that 2.5 hour keynote involved Windows to some degree. It’s kind of crazy.

But the highlight for me was Stevie Bathiche’s segment on Project Solara during the keynote. This wasn’t just interesting; it was a fascinating reminder of the past when Microsoft didn’t just dominate computing but would routinely deliver peeks at a future it envisioned. Stevie is a genius and a truly incredible person, and he can communicate technical topics effortlessly. It’s a real skill. But it’s not just Stevie. This is a truly fascinating possible way forward for computing, which, again, is not what Microsoft has communicated in years. This seems to actually make sense, while also being very future-leaning.

Whether it amounts to anything remains to be seen. And I still have many Build session videos to watch. But sitting here now, that was the apex. The best part of a surprisingly interesting Build conference that defied expectations across the board.