Looking for Some Clarity on Copilot+ PC (Premium)

Copilot+ PC. Is it a brand? A specification? A secret agreement with Qualcomm designed to disenfranchise Intel? A little clarity would be nice.

To fully understand what’s happening here, we need to revisit the past. And don’t I just wish that I had Microsoft Recall to help me with that? But since that’s not yet possible, I’ve had to do this the old-fashioned way, using my own (admittedly flawed) memory, Thurrott.com search, and Google Search. And here’s the (relevant) history as I understand it.

Microsoft released the Surface Pro X in November 2019. Powered by a Qualcomm Snapdragon 8cx-based Microsoft SQ1 processor, Surface Pro X was billed as an “always-connected PC,” which, as you may recall was a big part of the Windows 10 on Arm push. The focus back then was on connectivity, not AI, and always-connected PCs came with a smartphone-like eSIM for, well, always-on connectivity.

So why start with Surface Pro X and always-connected PCs? Because of the parallels with today: Then, as now, Intel, Microsoft’s biggest silicon partner by far, was upset by the software giant’s dalliances with Qualcomm. And so it demanded to be included in the always-connected initiative, just as it had previously demanded to be included in the SoC discussion (with Atom) when Microsoft announced its first Arm push with Windows RT years earlier. History just keeps repeating itself.

Anyway, Surface Pro X was revved with a second-generation version with an SQ2 processor (8cx Gen 2) in October 2020. And then that was replaced by Surface Pro 9, which came in Intel and Arm/SQ3 (8cx Gen 3) processor versions, in October 2022. Few customers purchased any of these PCs because of ongoing performance and compatibility issues. The focus remained on connectivity: Indeed, the Arm version of Surface Pro 9 was marketed as “Surface Pro 9 with 5G.”

Microsoft fixed the compatibility problems for the most part, but Qualcomm’s processors continued to lag behind in performance. Except in one area, which seemed unimportant at the time. Behind the scenes, the transition to hardware accelerated AI was quietly underway: Unlike their predecessors, the SQ2/8cx Gen 2 included an integrated NPU (Neural Processing Unit) in addition to their Kryo CPU and Adreno GPU. The SQ2/8cx Gen 2 could achieve 9 TOPS (trillion of operations per second)of hardware accelerated AI performance, and the SQ3/8cx Gen 3 landed at 29 TOPS. Not that we ever discussed such things back then.

What we did discuss, haltingly, was what we could do with this curious new hardware feature. And among the benefits it marketed for Surface Pro 9 with 5G was compatibility with something called Windows Studio Effects, a set of webcam and microphone features that required an NPU and provided things like Portrait Blur, Automatic Framing, Eye Contact, and Voice Focus. Surface Pro 9 with 5G was, for a time, the only PC compatible with Windows Studio Effects.

Windows Studio Effects deserves a short discussion. It’s an important milestone in this history. Not that we understood so at the time.

Indeed, we mostly ignored this set of features when it was first announced, specifically because it was limited to just one PC. And there’s a parallel to that, too: Like the Copilot+ PC features Microsoft recently announced, the availability of hardware accelerated PCs was, and still is, so limited that getting excited about them–in ways good or bad–is a bit premature. But what I remember most about this time was the Stevie Bathiche demonstration of Eye Contact during the launch, our first weird experience with the creepiness of AI in Windows.

Not surprisingly, Stevie telegraphed the future.

“The NPU allows the offload from the CPU and GPU,” he explained in October 2022 with an image of Task Manager onscreen. “I have all Windows Studio Effects features turned on, and you can note that nothing is touching the CPU and the GPU. I even have cloud model, captioning, running at the same time … Local offload and now cloud offload, all on one machine. That allows us to run really complex algorithms [only against the NPU].”

That was almost two years ago. Since then, OpenAI ChatGPT happened. Microsoft’s AI push happened, and among the many inroads there, we saw the introduction of several AI-based features in Windows 11. Curiously, none of them ran on an NPU even if you had one: They are all software based, and Windows Studio Effects remains, to this day, the one AI-accelerated Windows 11 feature that requires an NPU.

Qualcomm finally getting its act together happened too. In October 2023, it announced the Snapdragon X family of Arm-based PC processors with dramatically improved performance across its CPU, GPU, and NPU, the latter of which is rated at 45 TOPS. And Qualcomm went after the MacBook Air from the get-go, comparing the Snapdragon X Elite against the then-current Apple Silicon M2-based MacBook Air and even the M2 Max in the MacBook Pro. (The M2, incidentally, is capable of 15.8 TOPS off its NPU; the M3 in the new MacBook Air is capable of 18 TOPS.)

Of course, Qualcomm also took aim at the traditional x86 PC market dominated by Intel. At its launch event that October, Qualcomm said that it, a relative newcomer and also-ran in this market, was redefining the PC experience around AI and generative AI.

“There is a new sheriff in town,” Qualcomm CEO Cristiano Amon said.

“We will literally have lots of applications that will have local models, hybrid models, and I think that is the future,” Microsoft CEO Satya Nadella said in a prerecorded segment that debuted at Qualcomm’s event. “There is a new generation of AI PCs that are getting created.”

AI PCs. Not AI as a feature. Not AI accelerated in some cases because of an NPU. But AI as the focus, the point, of a new kind of PC. AI PCs.

And this is where Intel enters the story again. Once again unexcited by the love Microsoft was showing Qualcomm and Arm, and by its exclusion in this AI PC discussion, Intel did what it always does: It formalized its own AI PC specification and demanded that Microsoft join it down this foolhardy path.

In December 2023, about six weeks after the Snapdragon X reveal, Intel announced its first-generation Core Ultra (“Meteor Lake”) chips for PCs. This couldn’t have been more confusing: The first Core Ultra chips were only for mobile PCs, and not desktops, upending Intel’s normal announcement schedule. They featured an NPU with AI hardware acceleration, but Intel never revealed a TOPS score for that NPU, so Qualcomm provided that data later by showing that it was less than one-third the speed of its NPU. Clearly, Intel had rushed out this product so it could have something in market to take on all the positive PR that Qualcomm was getting. And it was so lackluster, it could ship immediately, which was indeed an advantage over Qualcomm, albeit it a temporary one.

But the confusion goes deeper than all that. With the Core Ultra, Intel was “ushering in the age of the AI PC.” Its partners–Acer, ASUS, Dell, Dynabook, Gigabyte, Google Chromebook, HP, Lenovo, LG, Microsoft Surface, MSI, and Samsung–were launching “the world’s first AI PCs,” some that very day.

“The AI PC represents a new generation of personal computers to meet this demand,” Intel said. With dedicated AI acceleration capability spread across the central processing unit (CPU), graphics processing unit (GPU) and neural processing unit (NPU) architectures, Intel Core Ultra is the most AI-capable and power-efficient client processor in Intel’s history.”

On a side note, AMD introduced AI accelerated mobile PC chips with NPUs before Intel, but it never used the term AI PC. It then quickly adopted the term: By January, its NPU-based chipsets were “leading the AI PC era.”

Intel’s partners went on to deliver another round of Meteor Lake-powered AI PCs at CES in January 2024, except for Microsoft, which waited until March. Its Meteor Lake-powered Surface Pro 10 and Surface Laptop 6 were AI PCs and they targeted businesses only.

But it is perhaps notable that Microsoft only used the term “AI PC” once in its announcement of those devices. The issue is two-fold: Meteor Lake NPUs are capable of running Windows Studio Effects but not much else, putting them on par with 2022’s Surface Pro 9 with 5G. And Microsoft was again plotting with Qualcomm to promote an Arm-powered future; its coming Copilot+ PC specification would wildly exceed the capabilities of Meteor Lake-based PCs.

And that’s the problem. Microsoft announced the Surface Pro 10 and Laptop 6 in March, and then announced the Qualcomm-based Surface Pro 11 and Laptop 7 just two months later. That these newer devices don’t totally replace their predecessors is a nuanced point, but it didn’t look good. For Intel, it wasn’t good: Microsoft has obsoleted its AI PC designation just six months after it was first announced, assuming all we focus on here is NPU performance.

And to be clear, that is all that Microsoft is focused on. After spending decades trying to convince Intel to make more efficient PC chips that could help the platform better compete with mobile devices, it finally found an out: Qualcomm had finally delivered a no-compromises PC platform that was perfectly timed for the AI wave and the push to on-device/local and hybrid AI.

Microsoft had also partnered with Intel on the AI PC designation, yes, but it was always unclear what it meant. In February, they explained that it wasn’t all that formal, you just needed a Core Ultra (with NPU) chip, a Copilot key on the keyboard, and Copilot running in Windows 11. But as noted, AMD was using the term by then, and it didn’t seem like a formal specification like Intel Evo. Perhaps Intel should have been more suspicious of this vagueness.

In any event, Microsoft launched the Copilot+ PC specification in May at an event that heavily favored the Qualcomm Snapdragon X chips that it was using in its latest PCs. But that event also downplayed AMD, Intel, Nvidia, and Qualcomm, not to mention the several PC makers who had shown up to support the effort. Microsoft’s PC were “the ultimate Copilot+ PCs,” and so they got top billing. No one was happy.

Especially Intel, which got all pissy on Qualcomm’s big day, once again pulling a familiar tactic from its play book and pre-announcing a next-generation Core Ultra that would match the Copilot+ PC specs. More on that in a moment.

Among its many sins, Copilot+ PC is confusing to customers too. As with Windows Studio Effects, it’s a set of hardware-accelerated AI features–in this case, Recall, Cocreator, and more–that requires customers to buy a very specific kind of new Windows PC. This time, at least, there are choices, with roughly a dozen PC models at launch and not just one. But all these PCs are based on unproven Snapdragon X processors. When–and how–would these capabilities come to other PCs?

On this, Microsoft was vague. Is still vague. I’m getting a strong “no comment” vibe here, in fact.

At the Copilot+ PC announcement, Microsoft did note that it would have “more [PCs] in this category from Intel and AMD later this year,” but that was the only time the two companies were mentioned. And since then, the Computex trade show occurred in Taiwan, with AMD, Intel, and Nvidia all announcing something about Copilot+ PC.

This, too, is worth a discussion. Because we all have questions.

As a specification for a specific type of PC, Copilot+ PC applies only to new PCs that meet this specification, as was the case with Media Center and Tablet PC back in the day. That is, you couldn’t buy a normal PC and add Media Center capabilities. You had to buy a new PC.

As of this writing, there are no Copilot+ PCs in the market. But the first Copilot+ PCs arrive soon, on June 18, and all of them are based on Qualcomm Snapdragon X chips. None feature AMD, Intel, or Nvidia silicon of any kind.

So what did AMD, Intel, and Nvidia really say about Copilot+ PC in their respective announcements? Is there some clarity there?

Let’s take a look.

AMD is offering Ryzen AI 300 series processors for mobile and Ryzen 9000 series processors for desktop that utilize an NPU with 50 TOPS of hardware accelerated AI performance. PCs based on these chips will ship “later this year,” with the desktop variants arriving first, in July. But things get fuzzy when it comes to Copilot+ PC.

These chips are “Copilot+ ready,” AMD says.

“The Ryzen 9000 series … provides leadership AI and compute performance for ultrathin and premium Copilot+ PC,” AMD’s Jack Huynh is quoted as saying.

But here’s the money quote.

“We are excited to partner with AMD to deliver these new Ryzen AI-powered Copilot+ PCs,” Microsoft CEO Satya Nadella said.

So that alone answers one question: There will be Copilot+ PCs with AMD (and thus Intel and Nvidia) chips inside. But it doesn’t answer questions around the timing–does Qualcomm have some kind of exclusivity window?–or whether these capabilities will ever come to existing PCs in the market.

So let’s move on to Intel.

Intel, like AMD, is addressing the Snapdragon threat with a new generation PC chip, the second generation Core Ultra (“Lunar Lake”), which includes an NPU with 40+ TOPS of accelerated AI performance (and other interesting advances). This chip “provides the capabilities necessary for Copilot+ experiences coming to market,” Intel says, vaguely.

But here again, we get a Microsoft quote, this time from Windows head Pavan Davuluri.

“We are excited to see Lunar Lake come to market with a 40+ TOPS NPU which will deliver Microsoft’s Copilot+ experiences at scale when available,” Davuluri said.

This seems to address the timing issue. Intel says that Lunar Lake will arrive in Q3–meaning, July through September–and in time for the holiday selling season. If PCs based on these chips will “deliver Microsoft’s Copilot+ experiences when available,” then there’s no wait, and no exclusivity window.

Except for one thing. “When available” doesn’t mean “when Lunar Lake is available.” It means “when Copilot+ experiences are available for Lunar Lake-based PCs.

Take a second to let that sink in.

Here’s why we know this. Elsewhere in that same announcement, Intel adds writes, “Lunar Lake processors will extend unmatched global AI PC scale and receive free updates to Copilot+ when available.” Indeed, this mentioned twice. “Lunar Lake will get the Copilot+ experiences, like Recall, via an update when available.”

Like I said, confusing. Especially since Microsoft has explicitly said that Copilot+ would not be delivered via a software update to existing PCs. But here, again, Microsoft is using deliberately confusing language. It’s kind of shameful, frankly.

But Nvidia in some ways

“Newly announced RTX AI PC laptops from ASUS and MSI feature up to GeForce RTX 4070 GPUs and power-efficient systems-on-a-chip with Windows 11 AI PC capabilities,” Nvidia writes. “These Windows 11 AI PCs will receive a free update to Copilot+ PC experiences when available.”

There’s that language again, confirming that new PCs with modern NPU-powered AMD, Intel, and Nvidia hardware will arrive in the market later this year and then, at some time in the future, receive an update that will give them the Copilot+ PC experiences.

When will that be?

My guess is that it will occur with the full release of Windows 11 version 24H2 in October. And that if Microsoft and its partners can’t meet that date, that Copilot+ PC capabilities will ship to compatible PCs in a post-24H2 monthly update, in keeping with the haphazard way that Microsoft updates Windows 11 now. And that this helps explain the split release of 24H2, with Snapdragon-based PCs getting the update this month (in unfinished form) and the rest of the world getting it in the usual October time frame.

But what about existing PCs, PCs that predate the chips that AMD, Intel, and Nvidia just announced? Of the three, only Nvidia would qualify, as it makes powerful GPUs that in many ways surpass the performance of those 40+ TOPS NPUs. But as I wrote in Orchestration (Premium), all the AI-powered capabilities unique to Copilot+ PC require that NPU, and none will work against a GPU. So the question here is two-fold: Do previous-generation Nvidia GPUs include compatible NPUs and, if so, will Copilot+ capabilities ever come to those PCs?

My guess is no. But that’s only a guess, based on my understanding of the go-to market strategy here. Microsoft and its partners are using Copilot+ PC capabilities to lure customers into buying new PCs, which is completely understandable. At some point, PCs will evolve to the point where the (current?) Copilot+ PC specification seems quaint–i.e. 40+ TOPS NPUs will be common if not universal–and at that point, I expect these features to become part of Windows 11. This is what happened with Media Center and Tablet PC, of course.

The caveat here is that Nvidia’s announcement specifically mentions the Windows Copilot Runtime and that is “collaborating with Microsoft to add GPU acceleration for local PC small language models (SLMs)” like those used by Copilot+ PCs. Accelerating SLMs generally would benefit any PC, I guess, but it specifically says Windows Copilot Runtime, and that is exclusive to Copilot+ PCs. No one would expect developers to get new PCs to support this work, and so it’s possible that PCs with powerful current-generation Nvidia GPUs will be able to upgrade to Copilot+ PC capabilities.

Possible, but not definitely. And this work won’t happen any time soon: The API Microsoft and Nvidia are creating won’t even ship in preview until “later this year.”

And as for the history I laid out at the start of this article, Nvidia now claims that it launched the era of AI PCs in 2018 with the release of RTX Tensor Core GPUs and NVIDIA DLSS.

I guess there are all kinds of ways to look at this.