A New AI Era Continues, For Better or Worse (Premium)

The day that Microsoft introduced the world to its new Copilot+ PC branding, I took a press bus from my hotel in Seattle to the firm’s Redmond campus and was immediately confronted by a large sign with a Copilot logo reading, “A new AI era begins.” This messaging was repeated on banner flags attached to every light pole on the walk from the bus to the temporary event space the firm had set up on the edge of a soccer field in the recently modernized campus. And a wall with the same “A new AI era begins” tagline stood between me–and the other press, bloggers, and YouTubers who entered this tent-like structure–and the seats in front of the stage at which the software giant would finally discuss its plans that day.

It was May 20, 2024. And I’ve been looking for clarity ever since.

Copilot+ PC was born in confusion, and that confusion has reined ever since. Microsoft didn’t help–its initial guileless introduction of the controversial Recall feature that’s since been delayed for so long we still don’t have it today was the initial warning sign–and so we did our best. In his initial post about this branding, Laurent compared it to Intel’s Ultrabook brand, a specification–since overtaken by the far less obvious Evo brand–noting that it promoted a new category of PC that would deliver unique AI experiences. That’s accurate, but I’ve since expanded on that notion by comparing it to the multimedia PCs of the 1990s and the Media Center and Tablet PCs of the early 2000s because I feel that this designation is temporary. That is, we won’t need the Copilot+ PC branding, or the vaguer but better AI PC branding, in the years ahead because all PCs will meet its requirements and be capable of delivering the same capabilities.

As Microsoft explained at the time, the first wave of Copilot+ PCs were based on Qualcomm’s new Snapdragon X chips, which brought Windows 11 on Arm to the mainstream with the terrific efficiency, battery life, performance, and compatibility that customers had demanded for years but never received from this platform. But Copilot+ PC was never going to be restricted to Snapdragon, and the primary calling card of this branding–the “new AI era,” if you will–was always more about the on-device AI capabilities that were afforded by having a powerful NPU (neural processing unit) built into the PC’s SoC (system on a chip).

AI is inherently confusing. It arrived suddenly, like a freight train barreling unexpected into a station, a suddenly potent force that, like Windows on Arm, we’d been hearing about for years but to no avail. This is easy to forget, but Microsoft exploded on the scene in February 2023 with its cloud-based and OpenAI ChatGPT 4-powered Bing AI capabilities, setting off an “AI era” that is sharply different from what it is promoting now. In the chaotic year that followed, it rebranded its AI, wisely, to Copilot, and announced and then delivered a sweeping set of AI offerings and functionality across all its major product lines. But all of it was based in the cloud, which requires massive infrastructure investments and is extremely expensive to operate. The firm started plowing money into Azure to bolster its first-mover advantages and leverage the strength of its cloud focus from the previous decade. But it’s been costly, even for a company as rich as Microsoft. In the most recent quarter alone, it reported spending $19 billion on these efforts and promised to continue similarly for the next four quarters too.

The “new AI era” that Microsoft is now promoting is not that AI era. Well, it is, technically, as cloud-hosted AI isn’t going anywhere and will always be required for certain tasks. But part of Microsoft’s long-term plan to reduce the cost of AI is to move as much of the processing as possible down to the edge, the PCs that we all use. In other words, Microsoft is operating like SETI–or even like Bitcoin miners–by offloading the brute processing work that AI requires to us, its customers. But it understands that these customers, perhaps weary of the customer-antagonist behaviors they see in Windows in recent years, aren’t just going to line up to help out the poor software giant. And so it worked with its silicon partners to establish a baseline standard for the capabilities a modern PC would require to participate in this scheme. And from a marketing perspective, it started working on Windows 11 features that would be exclusive to these modern PCs. Even though that distinction is largely artificial.

In other words, the “new AI era”–which will lead to what I think of as a “hybrid AI era”–requires PCs with an NPU. But not just any NPU. Smartphones, Macs, and even previous generation Windows 11 on Arm-based PCs all shipped with basic, underpowered NPUs. But these are not powerful enough for the hardware-accelerated on-device AI tasks that Microsoft envisioned. And so the Copilot+ PC specification was born: To qualify for this branding and to enable the unique Windows 11 features Microsoft planned, PCs required an NPU that delivered 40 TOPS (trillion operations per second) or more of performance. Plus, 16 GB of RAM and 256 GB of speedy solid-state storage at a time when a mainstream Windows 11 PC requires just 4 GB of RAM and 64 GB of storage of any kind.

This decision would theoretically help Microsoft’s PC maker partners, as the resulting devices would be expensive, premium models with higher margins. And with PC sales stagnant after a years-long post-COVID drought, perhaps Copilot+ PC would be strong enough to jumpstart the market more broadly. Sadly, that hasn’t panned out, at least not yet. In addition to being a terrible brand, the messaging on Copilot+ PC has only been confusing. Microsoft came to market with only a single compelling on-device AI feature, Recall, and that was delayed until a 2025 broad release at the earliest. The first Copilot+ PCs were all Snapdragon models, and though this platform has finally delivered on the promises, suspicions remain, and power user influencers aren’t helping matters by constantly harping on compatibility issues that won’t impact most people. Microsoft never clearly articulated whether future AMD- and Intel-based PCs would be branded as Copilot+ PCs and, if so, when. And when it finally did offer some information about the timing, it was vague: You can get that new PC now, but you won’t have the Copilot+ PC features–most of which stink, anyways–until November. In preview or stable? We don’t know.

As we ride into the final quarter of 2024, we face further uncertainties. Industry analysts have said that the AI PC/Copilot+ PC push will not help PC sales, and they now expect 2024 to end with a thud, with perhaps very low single digit PC sales growth year-over-year at best. Qualcomm and Microsoft paid PC makers to support Snapdragon X with multiple PCs at launch, and they complied, with a reported 20+ models across the top five PC makers. But now that AMD and Intel are entering the game, especially Intel, the payola king, they’re shifting effortless to those processors, with each supporting the new AMD and Intel silicon with dozens or even hundreds of new PCs. Intel has also been notably punishing in its marketing, calling out Qualcomm for its remaining compatibility issues and spooking corporate decision makers even more.

Today, we are on the cusp of Windows 11 version 24H2, which will deliver these new Copilot+ PC features–including, eventually, Recall–to any compatible and capable PC. Having used 24H2 nearly exclusively since May, I can tell you that this experience is not particularly unique or rewarding. And that the on-device differentiation Microsoft hoped would inspire new PCs sales is no reason to buy a new PC of any kind. Where the Snapdragon X-based PCs excel is in efficiency, battery life, and performance, at least within the confines of the Ultrabook-like form factors it’s now being used. But AMD and, especially, Intel will kill it in overall performance and with compatibility. And the only question there–again, this era has been marked mostly by uncertainty–is where they land on efficiency and battery life.

I have used the term “in the ballpark” a lot this year.

The Snapdragon X-based Copilot+ PCs do not exceed the efficiency and battery life capabilities of the MacBook Air they so clearly target, but they are in the ballpark. They deliver instant-on performance each day but fall back to a sleep mode after a few days and boot up more slowly, and I’m seeing about 10 hours of battery life on average. But the MacBook Air delivers instant-on every single day over very long periods of time, and it gets 15 hours of battery life. Snapdragon X isn’t “better,” but it’s in the ballpark.

It’s early days on the new AMD and Intel Copilot+ PCs. But my gut feeling, based on speaking with several industry insiders while at IFA earlier this month, is that they will be in the ballpark for efficiency and battery life when compared to Snapdragon X. So maybe in the 6 to 8 hour range, we’ll see. I can tell you that I got up this morning, opened the lid on the AMD Zen 5-based OmniBook Ultra I just received and … it did not come on instantly. It’s just one day. But it’s worrying.

But here’s the thing. Every product is a trade-off of some kind. And the promise that Intel is making explicitly, the same promise that AMD in its unique “we’re just happy to be here” way perhaps makes a bit more implicitly, is that the performance and compatibility gains will outweigh whatever other issues … assuming the efficiency and battery life is, again, in the ballpark. I think they’re going to land there nicely, and my early game-playing experiences on the OmniBook are quite gratifying. This hardware is impressive.

Returning to that “new AI era” bit, I was reviewing Microsoft’s initial Copilot+ announcement from May and came across a line near the end I had forgotten about. We know that this era will be marked by various phases, and that one way to delineate them is to discuss cloud-based AI, on-device AI, and then hybrid AI as three high-level milestones. But if we look at just Copilot+ PC, which focuses exclusively on NPU-based on-device AI, there’s some nuance there as well. And not just the two phases we know about, Snapdragon X and x64. There’s a third phase coming: “In the future, we expect to see devices with this [new AMD and Intel] silicon paired with powerful graphics cards like NVIDIA GeForce RTX and AMD Radeon, bringing Copilot+ PC experiences to reach even broader audiences like advanced gamers and creators,” Microsoft noted at the time.

Interesting. One of the big complaints about on-device AI this year, specifically Copilot+ PC, is that gamers and creators already have powerful CPUs and GPUs that could handle these tasks right now, albeit less efficiently. Why on earth would Microsoft prevent that? I buy the arguments about NPUs and efficiency, as perhaps best explained by Microsoft’s Stevie Batiche at Build 2024 this past May. But those who spent thousands on expensive PCs with powerful GPUs already have hardware that vastly exceeds the 40+ TOPs Copilot+ PC requirement. Surely, these capabilities will be opened to them too?

That line from May suggests it’s the plan. I’ve long argued that a big part of the problem is a lack of sophistication on Microsoft’s part, that it’s spastic and chaotic push to spread AI everywhere means that it hasn’t had the time it needs to make Windows, which could orchestrate on-device AI tasks to the most efficient/best processors types available in any given PC, catch up. This orchestration should include the intelligent offloading of certain tasks to the cloud as well.

This is painful in some ways, but unless I’m missing something, that’s exactly what Apple is doing with Apple Intelligence. This company is routinely criticized in some circles for moving deliberately if not slowly, of being behind, when its competitors are racing to out-AI each other. But all Apple is doing is behaving responsibly. That doesn’t mean that everything it does is correct. Just that its approach is more defensible, more logical. Where Microsoft is behaving as if it will lose if it doesn’t behave so quickly it’s erratic, Apple seems to understand that when AI works, it’s boring. It’s the same features, just better. It’s not random new features sprayed across an operating system like machine gun fire, not caring what it hits or what the outcome is.

The new AI era.

I don’t know. It seems like the old AI era, or at least the same old, same old. I believe that AI is a fundamental leap forward for computing and thus for mankind as well. But I also know we need adults steering the ship. To date, Microsoft had done nothing but sow confusion by moving too quickly and explaining itself poorly. You will find reasons to love particular products and services, and you should be confused that their best attributes aren’t what Microsoft is marketing.

I keep coming back to the same points.

Broadly, AI is like MSG in that it makes existing things better. But no one pays for MSG, it’s just in there, making the food we do pay for taste better. Similarly, I will not pay for AI, just as I would not have paid for spelling or grammar checking in the 1990s. AI isn’t a feature. It’s a feature enhancer.

We can get more specific.

The best parts of Windows 11 version 24H2 have nothing to do with AI. Copilot+ PC currently offers no good reasons to buy a compatible PC, though Recall will change that.

The reason you buy a Snapdragon X-based PC is that it nails the basics without drama while delivering superior efficiency and battery life.

And the reason you may buy a new AMD- or Intel-based PC is that they nail the basics of efficiency and battery life while delivering superior performance (in particular for graphics/gaming) and compatibility. AI is not part of these conversations, and yet it’s almost all Microsoft talks about. This is more important to it than it is to its customers.

Apple? That company just shipped its last-ever platform upgrades without major AI innovations, and we may one day look back on that as the end of its own era. For now, reviewers are forced to evaluate the new systems, and the new hardware on which they run, on their own merits. And you know what? They’re pretty damned good. Sprinkling a little AI on them, as Apple will do in the months ahead, will make them a little better. But this won’t trigger a surge in sales, just as AI has failed to do so in the PC space.

And if you love the chaotic insanity of AI and wish to embrace it fully, Google is there for you: Its latest Pixel devices and software platforms are to mobile what Copilot+ PC is to the PC, but even more dramatically. At least some of those features are compelling.

AI is not a bust. But it is a bubble. And the sooner we get past this mania, the better. There’s enough uncertainty in life as-is. And to date, that’s an area in which AI, and the companies pushing it maniacally, have not helped in the slightest.