Copilot+ PC Launch: What Microsoft Announced in 2024

Copilot Plus PC Launch Record

Microsoft introduced Copilot+ PCs on May 20, 2024 as a Windows 11 hardware category with an NPU capable of more than 40 trillion operations per second, at least 16 GB of memory, at least 256 GB of storage, and access to new Windows AI experiences. It was not one computer, one chip, one assistant, or a rebranded ChatGPT product.

flowchart TD
    A["Copilot+ PC category"] --> B["Specific device and processor"]
    B --> C["CPU, GPU, NPU, memory, storage, and battery"]
    C --> D["Windows build and feature availability"]
    D --> E["Application, runtime, and model"]
    E --> F["Local, cloud, or mixed execution path"]
    F --> G["Measured user workload"]

What Microsoft announced

Microsoft's launch record framed Copilot+ PCs as a new class of Windows 11 systems. The initial devices used Qualcomm Snapdragon X Series processors from Microsoft Surface and partners including Acer, ASUS, Dell, HP, Lenovo, and Samsung.

The launch category required an NPU above 40 TOPS, at least 16 GB of RAM, and at least 256 GB of storage. Microsoft connected that hardware to Windows experiences including Recall, Cocreator, Live Captions, and Windows Studio Effects.

Microsoft later expanded the category to other silicon. Its current Copilot+ PC developer guide lists supported Qualcomm, AMD, and Intel families and notes that feature availability varies.

The category therefore describes a threshold and Windows ecosystem. It does not make every device equivalent.

What an NPU adds

A neural processing unit is a processor specialized for machine-learning operations. It can run supported workloads with different performance and power characteristics from the CPU or GPU.

That can matter for sustained local features such as image processing, background effects, transcription, or model inference. The benefit depends on software support, runtime, model, numeric precision, memory movement, power policy, and the rest of the system.

TOPS is a peak operation-throughput measure under a stated numeric format. It is not application latency, tokens per second, output quality, battery life, or compatibility.

Two NPUs with similar TOPS can perform differently on a real model. A faster accelerator can still sit idle if the application uses the GPU, CPU, or cloud.

The product layers

Copilot+ PC described the hardware category. Windows supplied operating-system features. Device makers supplied the specific computers. Chip vendors supplied processors and runtimes. Application developers decided whether and how to use the NPU.

Microsoft Copilot was an assistant across several product surfaces. ChatGPT was OpenAI's product. GPT-4o was an OpenAI model.

Those layers can connect, but they should not be collapsed. A Windows feature may run locally without being Microsoft Copilot. A Microsoft assistant request may use a cloud model. A Copilot+ PC can meet the NPU threshold without supporting every Windows feature in every market on launch day.

The evaluation should name each layer.

The 2024 performance comparison

Microsoft compared selected Snapdragon X Elite systems with selected Apple M3 MacBook Air configurations and made performance and battery claims under disclosed conditions.

The preserved E018 outline simplified the comparison into "58 percent faster." That phrasing removes the workload, device, operating system, software, power state, and test-owner boundary.

Microsoft's current claims and disclosures page shows how specific the evidence needs to be. Its later claims identify the test type, device set, date, brightness, connectivity, software, and other settings.

A launch comparison can describe the marketing and disclosed test. It cannot rank every current Copilot+ PC against every MacBook or Windows computer.

Recall was one launch feature, not the category

Recall drew the most privacy and security attention because it stored searchable snapshots of screen activity.

The feature changed materially after the initial announcement. Its chronology belongs in [[Microsoft Recall Product and Privacy Record]].

A buyer should not treat Recall as required to benefit from an NPU, and should not treat disabling Recall as disabling every Copilot+ PC feature. The category and the feature are different.

Current requirements need device-level checking

Before buying or deploying a computer, verify the exact model, processor, memory, storage, display, ports, wireless hardware, serviceability, operating-system edition, build, language, region, account, management state, and feature support.

Some experiences arrive through updates. Some depend on the processor family. Some require a Microsoft account, specific Windows components, or additional storage. Managed-device policy can remove or constrain features.

The current device page and release notes control current availability. A 2024 category page cannot.

Benchmark the workload, not the badge

MLCommons' current MLPerf Client benchmark evaluates selected local LLM workloads across client systems and requires accuracy thresholds for supported acceleration paths. It reports defined tasks, models, runtimes, repeated runs, and known issues.

That is stronger evidence than a bare TOPS number. It remains one benchmark suite.

The useful test uses the buyer's actual application and representative input. It records the execution path, accepted quality, latency, throughput where relevant, active task time, review, energy or battery, heat, noise, offline behavior, and failure recovery.

[[How to Evaluate an AI PC Claim]] provides the complete method.

Historical significance

Copilot+ PC made an NPU threshold part of a mainstream Windows hardware category. It also tied hardware marketing more closely to operating-system AI experiences.

That made product comparison harder, not easier. A reader now has to separate the specification, benchmark, operating-system feature, model, application, data path, and workload.

The launch record answers what Microsoft announced. It does not answer which current computer a person should buy.

This product record was developed with AI assistance from the preserved E018 outline and the linked Microsoft, MLCommons, and Venture Step records. Dalton Anderson remains the author. Microsoft sources describe Microsoft claims and product behavior. Named device comparisons require independent hardware, benchmark, accessibility, security, procurement, product, source, and founder review. Publication is not authorized.