A weekly roundup of the latest developments in on-device AI hardware
Your laptop is about to get a lot smarter—and it won't need the cloud to do it. Over the past 18 months, every major chipmaker has shipped or announced a dedicated Neural Processing Unit (NPU) inside their laptop processors. Apple started the trend with the Neural Engine in its M-series chips. Intel followed with Core Ultra. Qualcomm turned heads with Snapdragon X Elite. And AMD is now all-in.
An NPU is a specialized processor designed to accelerate machine learning tasks—things like real-time background blur, AI noise cancellation, and on-device language translation. Unlike your CPU or GPU, which are built for general-purpose computing, an NPU is optimized specifically for the matrix math that powers neural networks.
This roundup covers what NPUs actually do, why they're becoming a must-have laptop feature, the latest chip announcements, and what the performance numbers really mean. By the end, you'll know exactly what to look for when shopping for your next AI-ready laptop.
A Neural Processing Unit is a chip designed to execute neural network operations—specifically inference, the process of running a trained AI model to make predictions or decisions. When you ask an AI to blur your background on a video call, an NPU handles that workload locally, on your device, rather than sending your video feed to a cloud server.
CPUs are generalists. They handle a few tasks at high speed, juggling everything from your operating system to your spreadsheet. GPUs are parallel processors built for graphics and complex math—they're great at training AI models but consume significant power. NPUs sit in a middle ground: they're highly specialized for the specific mathematical operations neural networks use, making them faster and dramatically more energy-efficient than either CPUs or GPUs for AI inference.
At the core of every neural network is matrix multiplication—multiplying large arrays of numbers to simulate how biological neurons fire. NPUs are built with thousands of small processing elements arranged to perform these multiplications in parallel. This architecture allows an NPU to process billions of operations per second while drawing a fraction of the power a GPU would need for the same workload.
On-device AI matters for two reasons: speed and privacy. When processing happens locally, there's no network latency—responses are instant. And your data never leaves your machine, which means your video feed, your photos, and your voice stay private.
Key Takeaway: NPUs are purpose-built for AI inference, executing neural network math far more efficiently than CPUs or GPUs, enabling fast, private, on-device AI features.
Cloud-based AI has limits. Sending data to a server introduces latency, requires a reliable internet connection, and raises privacy concerns. NPUs flip that model. Your laptop becomes the AI engine, processing everything locally.
Here's the practical argument: running AI workloads on a CPU or GPU drains battery fast. A modern GPU can draw 100+ watts under load. An NPU performs similar tasks at 5–15 watts. For laptop users, that's the difference between making it through a workday and hunting for an outlet by noon.
Windows 11's Studio Effects, Apple's on-device Siri improvements, and real-time translation features all rely on NPU acceleration. These aren't niche add-ons—they're becoming core OS features. Microsoft's Copilot+ standard, which requires an NPU with at least 40 TOPS, signals that future Windows AI features will assume NPU hardware exists.
When your NPU handles background blur, noise cancellation, or photo enhancement, your CPU and GPU are free to handle your actual work. The result is better multitasking and a more responsive system overall.
Key Takeaway: NPUs extend battery life, improve privacy, and free your CPU and GPU for other tasks—making them essential for modern laptops.
Apple shipped its first Neural Engine in the A11 Bionic chip (2017) for iPhones. The M-series chips brought it to Macs. The M3's Neural Engine delivers up to 18 TOPS, powering features like Live Text, subject isolation in Photos, and AI-powered editing in Final Cut Pro. Apple doesn't chase raw TOPS numbers the way Qualcomm does, but its integrated approach means the Neural Engine works seamlessly with macOS.
Intel's Core Ultra processors, released in late 2023, were the first mainstream Windows laptop chips with a dedicated NPU. The NPU delivers up to 11 TOPS—modest by today's standards, but enough for tasks like background blur and noise suppression. Intel's next-gen Lunar Lake chips are expected to triple that number.
Qualcomm's Snapdragon X Elite, announced in October 2023, shook up the laptop market. Its NPU delivers up to 45 TOPS, clearing Microsoft's Copilot+ threshold and outperforming Intel and Apple in raw NPU specs. The first Snapdragon X Elite laptops shipped in mid-2024, and early reviews have been strong.
AMD's Ryzen 7040 series introduced the XDNA NPU, capable of up to 10 TOPS. The newer Ryzen AI 300 series ("Strix Point") pushes that to 50 TOPS, making it a serious contender for Copilot+ certification. AMD's strength lies in combining competitive NPU performance with its established CPU and GPU expertise.
In May 2024, Microsoft announced Copilot+ PCs, a new category of Windows laptops that must meet a 40 TOPS NPU requirement. This threshold exists to ensure consistent performance for Windows AI features like Recall, Live Captions, and Studio Effects. It's a significant move—Microsoft is effectively setting the hardware baseline for AI-capable Windows laptops.
Key Takeaway: The NPU landscape shifted dramatically in 2023–2024. Qualcomm leads in raw TOPS, AMD is close behind, and Microsoft's 40 TOPS Copilot+ standard is now the benchmark to beat.
Laptops with NPUs run Windows Studio Effects locally, providing real-time background blur, automatic framing, and "eye contact" correction that makes it look like you're looking at the camera. These effects process at 30+ frames per second without impacting system performance. On a Copilot+ PC, the NPU handles this while your CPU runs your presentation software.
NPU-based noise cancellation filters out background sounds—keyboard typing, a barking dog, construction noise—in real time. Unlike standard noise suppression, AI-based systems learn to distinguish your voice from ambient noise, producing cleaner audio even in chaotic environments.
Apple's M-series Neural Engine accelerates Adobe Photoshop's Neural Filters, which can enhance facial features, colorize black-and-white photos, or apply complex style transfers. Final Cut Pro uses the NPU for the Scene Removal Mask, which isolates subjects in video without manual rotoscoping. On Windows, the Microsoft Photos app uses NPUs for Magic Eraser and Super Resolution.
Windows 11's Live Captions feature translates audio from any app—videos, podcasts, voice calls—into English subtitles in real time. On an NPU-equipped laptop, this runs locally, with low latency, even for lengthy sessions.
Voice assistants like Windows Speech Recognition and Apple's Siri use NPUs for on-device speech processing, reducing the delay between speaking and response. Predictive text and autocorrect also benefit, with NPUs powering more context-aware suggestions without sending your keystrokes to the cloud.
Key Takeaway: NPUs power practical, everyday features—video calls, noise cancellation, photo editing, and translation—that work better, faster, and more privately than cloud-based alternatives.
TOPS stands for Tera Operations Per Second—one trillion operations per second. It's the standard metric for NPU performance, measuring how many math operations (typically integer operations) the chip can execute each second. Higher TOPS generally means faster AI processing.
| Chip | NPU Performance |
|---|---|
| Apple M3 | 18 TOPS |
| Intel Core Ultra (Meteor Lake) | 11 TOPS |
| Intel Lunar Lake (upcoming) | ~40 TOPS |
| AMD Ryzen AI 300 | 50 TOPS |
| Qualcomm Snapdragon X Elite | 45 TOPS |
For basic features like background blur and noise cancellation, 10–15 TOPS is sufficient. For Microsoft's Copilot+ features—including the controversial Recall tool—you need 40+ TOPS. If you're buying a laptop today and want access to the latest Windows AI features, 40 TOPS is the number to target.
Chipmakers aren't slowing down. Intel's Lunar Lake targets 40+ TOPS. AMD's next-gen Strix Point hits 50. Qualcomm's follow-ups will likely exceed 60. Within two years, 40 TOPS will be the baseline, and premium laptops will push past 100.
Key Takeaway: TOPS is the key NPU metric. For future-proofing, look for 40+ TOPS—the Copilot+ threshold—but understand that even 10–15 TOPS enables useful features today.
Yes, both handle parallel processing. But GPUs are general-purpose parallel processors—they're great at graphics, scientific computing, and AI training. NPUs are specialized for AI inference. An NPU can't replace your GPU for gaming, and a GPU is overkill for background blur.
If you use video conferencing, want real-time translation, or care about battery life during AI-heavy tasks, yes. If you're on a tight budget and your current laptop works fine, you can wait a generation. But NPUs are becoming standard—by 2026, most laptops will have one.
No. NPUs are accelerators, not replacements. Your CPU still runs your operating system and applications. Your GPU still renders graphics. The NPU handles AI tasks alongside them, making the whole system more efficient.
Far from it. The gap between Intel's 11 TOPS Meteor Lake NPU and Qualcomm's 45 TOPS Snapdragon X Elite is enormous. Beyond raw TOPS, real-world performance depends on software optimization, driver quality, and how well the NPU integrates with the rest of the chip.
Key Takeaway: NPUs complement CPUs and GPUs—they don't replace them. And not all NPUs are created equal; 40+ TOPS is the threshold for next-gen Windows AI features.
The NPU market is projected to grow from $5.6 billion in 2023 to over $70 billion by 2032—a compound annual growth rate above 32%. Every major chipmaker has NPU development teams, and laptop OEMs are making AI capabilities a selling point across their lineups.
The hardware is ahead of the software right now. That's changing quickly. Microsoft's Copilot+ features are just the start. Expect to see more AI-powered tools in creative software, productivity suites, and even web browsers. As developers learn to leverage NPUs, the range of features will expand significantly.
Within two to three years, an NPU will be as standard in a laptop as a webcam or Wi-Fi. Intel, AMD, and Qualcomm are all integrating NPUs into their mainstream chips. Apple's Neural Engine is already in every Mac. There won't be a "non-NPU" option in the premium segment.
NPUs have limits. They're not designed for training large AI models—that still requires GPUs or cloud data centers. Software optimization varies by platform, and some features are locked to specific chips or operating systems. And as with any new technology, early adopters may find that some promised features arrive later than expected.
Key Takeaway: NPUs are becoming standard hardware. The market is growing fast, software support is expanding, and within a few years, you won't have a choice—every laptop will have one.
NPUs represent the biggest shift in laptop hardware since solid-state drives replaced spinning disks. They enable a new class of AI features that run locally, privately, and efficiently—without draining your battery or depending on a cloud connection.
The current landscape is clear: Qualcomm and AMD lead in raw NPU performance, Apple offers the most polished integration, and Intel is playing catch-up but closing the gap. Microsoft's 40 TOPS Copilot+ threshold has become the benchmark to beat, and it's the number you should look for when shopping.
If you're buying a laptop today, prioritize models with an NPU that meets the Copilot+ standard. You'll get access to the latest Windows AI features, better battery life on AI workloads, and a system that's ready for the software of the next few years. The AI-powered laptop era isn't coming—it's already here.
Ready to future-proof your next laptop? Look for models with an NPU that meets the Copilot+ standard (40+ TOPS) to unlock the full potential of on-device AI. Check out our latest laptop reviews and buying guides to find the perfect AI-powered companion.
A CPU handles general-purpose computing tasks with a few high-speed cores. A GPU handles parallel processing for graphics and complex math with thousands of smaller cores. An NPU is specialized for neural network inference, performing matrix math extremely efficiently. Each is optimized for different types of work.
NPUs enable on-device AI features like real-time background blur, AI noise cancellation, and on-device translation. They handle these tasks faster and with far less power than a CPU or GPU, which means better battery life and a more responsive system.
Not directly. An NPU won't speed up web browsing or spreadsheet work. But by offloading AI workloads, it frees your CPU and GPU, which can improve multitasking and overall system responsiveness.
No. While both handle parallel processing, a GPU is a general-purpose parallel processor. An NPU is specialized for AI inference. An NPU cannot replace a GPU for gaming or graphics rendering.
For basic AI features, 10–15 TOPS is enough. For Microsoft's Copilot+ features, you need 40+ TOPS. If you want your laptop to handle future AI software, 40+ TOPS is the sweet spot.
Not really. NPUs aren't designed for graphics rendering. They're for AI inference. Gaming performance still depends on your GPU.
No. Apple has included a Neural Engine (its NPU) in every Mac since the M1 chip. Qualcomm and AMD also make NPUs for Chromebooks and other devices.
The opposite. NPUs enable on-device processing, which means your data—video feeds, voice recordings, photos—stays on your laptop. You're not sending it to a cloud server.
Not always. Developers need to write code that specifically uses the NPU. That said, operating system features and major applications are increasingly NPU-aware. Microsoft's Copilot+ features require an NPU, and apps like Adobe Photoshop and Final Cut Pro already use NPUs when available.
No. NPUs deliver real, measurable benefits for AI workloads—faster processing, lower power consumption, and better privacy. That said, marketing departments do inflate the importance of TOPS numbers. Focus on real-world features and battery life, not just specs.