What Happened to Microsoft's Copilot+ PCs?

Copilot+ PCs weren't killed off in a press release. Microsoft and its hardware partners just stopped using the name. Two years of weak sales, enterprise indifference, and a flagship feature (Recall) that scared security researchers pushed the whole industry to quietly retreat from NPU-gated "AI PC" branding, even as the underlying chips kept shipping.

What was a Copilot+ PC supposed to be?

Microsoft launched the category in May 2024 with a hard spec bar: a neural processing unit rated at 40 or more TOPS, 16GB of RAM, and 256GB of storage.¹² The first machines ran exclusively on Qualcomm's Snapdragon X chips, with Intel and AMD versions arriving later that year.¹²

The pitch was exclusivity. Only Copilot+ hardware would get Recall (a searchable timeline of everything on your screen), Cocreator, Click to Do, and enhanced Windows Studio Effects. Satya Nadella framed it as a new era of "joy and wonder" for the PC. The subtext was clear: buy the new chip or miss the new Windows.

How many Copilot+ PCs actually sold?

The sales numbers tell a flatter story than the launch keynote did.

• 2024 market share. Windows PCs meeting the 40+ TOPS bar made up just 0.5% of the entire 2024 PC market.³ IDC put total Copilot+/generative-AI PC unit sales for the year at roughly 3 million units.³
• Early 2025 crawl. By Q1 2025, that share had inched up to 1.9% of the PC market and 2.3% of Windows PCs specifically.³
• Laptop shipments. Mercury Research found Copilot+ compatible laptops were under 10% of total laptop shipments in Q3 2024, even as any-NPU "AI PC" chips already accounted for about 36% of mobile CPU sales.⁴
• Europe, a year in. Context's distributor data showed Copilot+ PCs were only 9% of AI-capable PCs shipped in Europe in Q2 2025, up from 4% in Q1 and 2% in Q4 2024.² Growth, technically, but off a tiny base while broader AI-PC penetration sat around two in five PCs sold.²
• Price cuts didn't move the needle. OEMs cut Copilot+ prices roughly 10% in Europe in Q4 2024. The Copilot+ share of AI-capable laptop sales still only crept from 3% to 5% that quarter, even as overall AI-laptop adoption jumped from 22% to 32%.⁵

The category Microsoft built its entire marketing push around never cracked low single digits of the PC market for most of its two-year run.

Why did Copilot+ PCs fail to sell?

A few things compounded at once.

• Recall was a security problem before it was a feature. Researchers found Recall stored unencrypted, easily extracted screenshots of user activity, including passwords and financial data. Microsoft delayed the feature multiple times, and it shipped without Recall at launch. Years later, it's still labeled "preview."³¹
• The exclusive features weren't that exclusive in value. PCWorld and The Register both found that Recall, Click to Do, Cocreator, and Windows Studio Effects offered little that cloud-based tools already running on any PC couldn't approximate.³ Paying an NPU premium for a marginal upgrade is a hard sell.
• Chip rollout lagged. Qualcomm had Copilot+ hardware first. Intel and AMD versions trailed by months, which meant most buyers shopping in 2024 simply couldn't get a Copilot+ machine in their preferred architecture even if they wanted one.¹²
• The premium was real, the payoff wasn't. IDC's Jitesh Ubrani summed it up bluntly: Copilot+ "created a higher-priced PC with new AI capabilities," but it "hasn't increased the market size in terms of units."⁶ Buyers paid more and got a label, not a reason to upgrade.

Enterprises shrugged even harder

Consumer indifference was one thing. Enterprise indifference was worse, because enterprise refresh cycles are where PC makers actually make money on premium hardware.

Canalys found that by March 2025, 73% of channel partners were already familiar with Copilot+ PCs, so awareness wasn't the problem. But a month later, only 33% of those same partners said AI capability of any kind was an important purchase factor. That gap between awareness and purchase intent is the whole story: IT buyers knew the label existed and mostly didn't care. Deployments that did happen were narrow pilots, not fleet-wide rollouts.

The Register documented the same pattern in the shipment data: Copilot+ PCs were still only 9% of AI-capable PCs moving through European distribution in mid-2025, nearly a year after wide availability.² By late 2025, analysts across IDC, Tirias Research, and Technalysis Research were openly saying the Copilot+ hype needed to end, because it had produced confusion in the channel rather than adoption on desks.⁶

How did Microsoft quietly retire the Copilot+ brand?

Microsoft didn't announce a retreat. It just stopped centering the brand.

In October 2025, the company repositioned its messaging around making "every Windows 11 PC an AI PC," leaning on cloud-powered features like Copilot Vision and "Hey Copilot" that run on any Windows 11 machine, Copilot+ certified or not. The NPU requirement, once the whole point, quietly stopped being the gate.

By Build 2026, the shift was unmistakable. PCMag reported that Nadella opened his keynote by telling developers they now had "the full scope of GPUs" available to them for AI development, dismissing the Copilot+ PC framing entirely.⁷ The takeaway from that keynote, as PCMag put it, was that "the agentic future of Windows won't require a Copilot+ PC."⁷

The hardware itself didn't disappear. New Surface Laptop Ultra models and RTX Spark-powered Surface devices technically clear the Copilot+ spec bar. Microsoft just stopped calling them that.⁸ Surface CVP Brett Ostrum confirmed it directly: "these are not called Copilot+ PCs. They do meet all the requirements of our previous bar."⁸ Qualcomm's SVP of compute, Kedar Kondap, said the industry is now offering the same experiences "probably without just using [Copilot+ PC] terminology now."⁸ Nvidia, notably, never attached the label to RTX Spark at all.⁸

What replaced the Copilot+ pitch?

The pivot shows up in what Microsoft is actually pushing now: cloud-first AI features (Copilot Vision, "Hey Copilot") that don't care what chip is under the keyboard, plus Windows ML support spreading across GPUs and CPUs generally rather than one NPU tier. Smaller models capable of running without a dedicated NPU further undercut the original rationale for exclusivity. Once a model runs fine on a GPU or even a CPU, the 40-TOPS gate stops being a technical necessity and starts looking like a marketing device. That's the same math driving the local-versus-cloud cost case playing out across AI hardware generally: buyers increasingly ask what a machine can run, not what tier it's certified for.

Microsoft still claimed, back during the 2024 holiday season, that Copilot+ systems hit 15% of premium PC sales. It hasn't published a comparable figure since. Meanwhile, Omdia projects broader, non-Copilot+-branded AI PCs will reach 55% of shipments in 2026 and 75% by 2029. The category Microsoft named is shrinking into irrelevance while the underlying trend, some kind of local AI hardware in most new PCs, keeps growing anyway.

The bigger lesson: hardware exclusivity versus owning capable machines

The Copilot+ story is a case study in a mistake that shows up all over software and hardware markets: gating a premium tier behind an artificial requirement instead of just making the thing good enough that people want it regardless of what's under the hood.

Buyers and IT departments didn't reject the idea of AI on their PCs. They rejected paying more for a locked feature set tied to one chip family, especially when the flagship exclusive (Recall) turned into a security story before it turned into a selling point. What won instead was the boring, general-purpose path: capable machines that run whatever software, cloud or local, the user actually wants. That's the same logic behind the broader software ownership renaissance already showing up in how teams choose their tools: people would rather own a flexible, general-purpose stack than rent access to a narrow, gated one.

It's also why the practical advice for anyone actually running models locally has never been "buy the certified box." It's understanding what hardware you actually need for the workload in front of you, whether that's a 27B-class model on a consumer GPU or a lighter model that runs fine on a laptop CPU. NPU certification was always a proxy for capability, and a weak one. Teams building their own AI tooling, the kind of work this blog covers when it looks at how tools like Remy help people run and own their AI stack instead of renting a vendor's version of it, don't ask whether a machine is Copilot+ certified. They ask what it can actually run.

What this means for IT buyers now

If you're refreshing a fleet or buying a new machine, the Copilot+ label is no longer a meaningful signal, because Microsoft itself has stopped treating it as one.⁷⁸ A few things matter more:

• Buy for RAM, CPU, and GPU capability, not a sticker. Whatever AI workload you actually run, cloud-based Copilot, a local model, or nothing at all, will run on capable general-purpose hardware. The NPU tier isn't the deciding factor anymore.
• Don't assume "AI PC" means what it did in 2024. Omdia's 55% shipment projection for 2026 covers any-NPU AI PCs broadly, not the narrow Copilot+ tier that stalled out in the low single digits for most of two years.³²
• Treat exclusive software features skeptically. Recall's rocky rollout is the clearest example of why locking a headline feature to specific hardware, and shipping it before it's secure, backfires with both consumers and IT departments.³¹
• Watch what OEMs actually call their new machines, not what spec sheet they meet. Surface, Qualcomm, and Nvidia have all signaled the branding is fading even on hardware that technically qualifies.⁸

The Copilot+ PC brand isn't dead in the sense of being pulled from shelves. It's dead in the sense that the company that invented it stopped believing it was the story. The AI features people actually use are showing up on every PC, gated hardware or not, and that's the outcome the whole exclusivity bet was supposed to prevent.