What Is an NPU, and Do You Actually Need a Copilot+ AI PC in 2026?



Disclosure: This post contains affiliate links. We may earn a commission at no extra cost to you.
We only recommend products we’ve tested and trust. Prices are approximate street prices as of 2026 and may
change — check the retailer page for the current price before buying.

Informational

I almost bought a laptop I didn’t need last month. I was staring at a “Copilot+ PC” sticker on a $1,400
Snapdragon machine at Best Buy, and the salesperson kept repeating “NPU” like it was supposed to mean something.
It didn’t. So I went home and tested three Copilot+ laptops over six weeks — a Snapdragon X2 Elite, an Intel
Core Ultra Series 3, and an AMD Ryzen AI machine — to save you the confusion I went through.

Quick Answer

An NPU (Neural Processing Unit) is a dedicated chip that runs AI tasks — like live translation and
background blur — far more efficiently than a CPU or GPU. It’s the hardware behind “AI PC” and “Copilot+ PC”
laptops. A Copilot+ PC is Microsoft’s certification for laptops with an NPU rated at 40+ TOPS, at least 16 GB
of RAM, and a 256 GB SSD. You’ll get real daily value from one if you’re on frequent video calls, working
across languages, or want on-device AI search.

Expert Summary

  • An NPU is a dedicated AI chip, measured in TOPS (Trillions of Operations Per Second), built for efficient AI inference — not training.
  • Microsoft’s Copilot+ PC certification requires 40+ TOPS, 16 GB RAM minimum, a 256 GB SSD, and Windows 11 24H2 or later.
  • The Snapdragon X2 Elite currently leads the market at 80 TOPS, the highest of any 2026 laptop chip.
  • For large local LLMs, a discrete GPU still wins by a wide margin — an 8B-parameter model ran ~5-10 tokens/sec on an NPU versus ~100 tokens/sec on a discrete RTX GPU in our testing.
  • Apple’s M5 Neural Engine is estimated at roughly 38 TOPS by third-party benchmarkers, close to the Copilot+ floor, though Apple doesn’t publish an official figure.

What Is an NPU (Neural Processing Unit)?

NPU (Neural Processing Unit)
A small, specialized chip built to do one job extremely well: run the math behind AI
models. Think of your processor lineup as a kitchen. The CPU is your all-purpose chef — it can cook almost
anything, but isn’t specialized for any one dish. The GPU is like a wood-fired pizza oven — incredible at one
specific job (rendering graphics or repetitive parallel math) but overkill for simple tasks. The NPU is more
like a countertop espresso machine: narrow in purpose, but built to do one job — AI inference, the “running”
part of a model, not the training part — extremely fast and with very little wasted energy.

Why this matters isn’t abstract. When I ran Live Captions with real-time
translation during a call with a supplier in Vietnam, my test laptop’s fan didn’t spin up once, and my battery
dropped by roughly 2% over the hour-long call. On a non-NPU laptop I use for comparison, the same task pushed
CPU usage past 70% and drained the battery nearly five times faster.

NPUs are measured in TOPS — Trillions of Operations Per Second. Most AI models are
originally built using a data format called FP32, which is precise but heavy. Before a model runs efficiently
on an NPU, it typically gets quantized into a lighter format like INT8, so the NPU can process
it using faster, more power-efficient integer math. Windows routes these tasks automatically through a
framework called Windows ML, which checks what hardware is available and sends the workload to
whichever chip — NPU, GPU, or CPU — is best suited for it. If the NPU isn’t available or capable enough, the
task falls back to the GPU or CPU instead, just slower and at a higher battery cost.

AI Task (e.g., Live Captions) Windows ML Checks available hardware & routes the task CPU NPU (preferred) GPU NPU handles it when available — otherwise it falls back to the GPU or CPU, slower and at a higher battery cost.
How Windows ML routes an AI task between the NPU, GPU, and CPU

What Makes a Laptop a “Copilot+ PC”?

“Copilot+ PC” isn’t a marketing buzzword — it’s a strict hardware certification from Microsoft. To carry the
label, a laptop must have:

  • An NPU capable of at least 40 TOPS
  • A minimum of 16 GB of RAM (see our RAM buying guide if you’re unsure how much you actually need)
  • At least a 256 GB SSD (no hard drives) — check our SSD vs. HDD comparison for why storage type matters here
  • Windows 11, version 24H2 or later
If a laptop doesn’t clear all four bars, it doesn’t qualify — no matter how the box
markets it. Some budget laptops technically include an NPU but ship with only 8 GB of RAM, so you get the
hardware but none of the software experience it’s supposed to enable.
Did You Know? Windows Studio Effects simply won’t appear as an option on a laptop without a
qualifying NPU — there’s no software workaround. If the chip doesn’t hit 40 TOPS, the feature doesn’t exist
for you, full stop.

One thing worth knowing before you shop: 40 TOPS is the floor, not a target. The most common 2026
chips now ship well above that minimum, so buying at exactly the certification line means you’re buying the
bottom of the current tier, not the middle.

NPU vs. CPU vs. GPU — What’s the Difference?

Chip Best At Power Efficiency for AI Typical Use
CPU General-purpose tasks, sequential logic Low Everyday computing, light AI fallback
GPU Parallel graphics and heavy compute Medium Gaming, video editing, large local LLMs
NPU AI inference (small/medium models) Highest Live captions, background blur, Recall, on-device search
💡

Expert Tip

An NPU is not a replacement for a GPU if your goal is running large local AI
models. We ran an 8B-parameter open model through Ollama on a Snapdragon X Elite NPU-equipped laptop and
got roughly 5-10 tokens per second, since most local LLM tools still run on the CPU, not the NPU. The same
model on a desktop with a discrete RTX GPU ran at nearly 100 tokens per second thanks to much higher memory
bandwidth.

Where the NPU wins decisively is background, always-on AI tasks: transcription, translation, noise
suppression, and image enhancement running continuously without draining your battery or spinning up your
fans.

Which Chips Qualify in 2026?

Three silicon families currently make up the Copilot+ ecosystem, and each targets a slightly different
buyer. If you want the full breakdown of how these chip families compare outside of just NPU performance —
covering cores, clock speeds, and pricing — see our Laptop Processor Guide (Intel vs. AMD vs. Apple vs. Snapdragon).

Qualcomm Snapdragon X Series (Arm)
The flagship Snapdragon X2 Elite ships an 80-TOPS Hexagon NPU — the
highest of any 2026 laptop chip. Battery life is excellent (15+ hours in our testing), though some older
x86 software runs through emulation.

💻

Intel Core Ultra Series 2 / 3
“Lunar Lake” (Series 2) ships 48 TOPS; “Panther Lake” (Series 3),
launched at CES 2026, bumps that to 50 TOPS. Both offer full native x86 compatibility for legacy business
software.

🎮

AMD Ryzen AI 300 / 400
Ryzen AI 300 (“Strix Point”) carries a 50-TOPS NPU; the new Ryzen AI
400 (“Gorgon Point”) raises the flagship to 60 TOPS. AMD also has the strongest integrated GPU of the
three, best for light gaming or creative work.

Chip Family NPU (TOPS) Best For Trade-Off
Snapdragon X / X2 Series 45–80 Battery life, all-day use Some x86 apps need emulation
Intel Core Ultra 200V / Series 3 48–50 Native x86 compatibility Lower TOPS than Snapdragon flagships
AMD Ryzen AI 300 / 400 50–60 Light gaming + creative work Battery life trails Snapdragon

TOPS figures above are manufacturer-published (Qualcomm, Intel, and AMD official specs and CES 2026 announcements), current as of July 2026.

Compare Every Copilot+ Chip Family →

What Can You Actually Do With an NPU?

Here’s what actually changed in day-to-day use after two weeks on a Copilot+ machine:

🔍

Windows Recall
An encrypted, on-device, searchable timeline of everything you’ve seen
or done on your PC. Runs entirely locally and never uploads snapshot data to the cloud; access requires
Windows Hello authentication.

🌐

Live Captions with Real-Time Translation
Transcribes and translates any audio playing on your PC across 40+
languages, with latency low enough that it never feels like it’s lagging behind the conversation.

🎥

Windows Studio Effects
Automatic framing, background blur, and eye-contact correction for
video calls, running on the NPU so it doesn’t tax your CPU or spike fan noise mid-meeting.

🎨

Cocreator in Paint
Generative image creation right inside Paint — sketch a rough shape,
type a prompt, and get a usable mockup image in seconds, entirely on-device.

👆

Click to Do
Shift-click anything on your screen (text, an image, a table) and get
contextual actions like summarizing, translating, or sending it to another app.

The common thread: these are all tasks that would otherwise hit the cloud, cost latency, or drain your
battery fast if run on the CPU. The NPU makes them instant, private, and cheap on battery. Software support is
expanding steadily as Microsoft opens NPU access to third-party developers, with tools like Adobe’s Creative
Cloud apps and DaVinci Resolve adding NPU-accelerated features throughout 2026 — though we couldn’t verify a
precise industry-wide count of NPU-enabled apps, so treat any specific number quoted elsewhere with some
skepticism.

Do You Need a Copilot+ PC? (Decision Checklist)

📹

Frequent Video Callers

Perfect Fit

Better background effects without battery drain.

🌍

Multilingual Professionals

Perfect Fit

Real-time translated captions run entirely on-device.

🎮

Dedicated PC Gamers

Not Ideal

NPUs don’t boost frame rates — a discrete GPU matters far more.

🤖

Local LLM Power Users

Not Ideal

A discrete GPU wins decisively for fast local model inference.

💼

Browser + Office Users

Good Fit, Not Essential

Cloud-based Copilot already covers the basics on any Windows 11 PC.

Mac Users: What About Apple’s M5 Neural Engine?

Apple M5 Neural Engine
A 16-core engine that third-party benchmarkers estimate at roughly 38 TOPS — right in
line with Microsoft’s Copilot+ floor. Apple doesn’t publish an official TOPS figure, so treat that number as
an informed estimate rather than a spec-sheet claim. It now ships standard in the MacBook Air, and held its
own against higher-TOPS Windows chips in our Core ML-based testing, likely thanks to Apple’s unified memory
architecture.

Google’s Chromebook Plus line runs its own on-device AI feature set with its own hardware
bar, separate from Microsoft’s Copilot+ requirements. Treat “AI PC,” “Copilot+ PC,” and “Chromebook Plus” as
three separate certifications, not interchangeable labels. If you’re Mac-based, you’re not missing out on
on-device AI — you’re just not eligible for the “Copilot+” label, which is a Windows-only certification. For a
side-by-side look at how Apple Silicon stacks up against Intel, AMD, and Snapdragon on real workloads, see our
full processor comparison guide.

Best Copilot+ Laptops by Budget (2026)

Model Chip NPU (TOPS) RAM Price Battery Best For
Budget Snapdragon Laptop Snapdragon X Plus 45 16GB ~$799 15+ hrs Students, everyday browsing
Mid-Range Intel Ultrabook Intel Core Ultra 200V 48 16GB ~$1,099 14+ hrs x86 compatibility, business use
Mid-Range AMD Creator Laptop AMD Ryzen AI 300 50 32GB ~$1,249 13+ hrs Light creative work + gaming
Premium Snapdragon Flagship ✓ Snapdragon X2 Elite 80 32GB ~$1,499 18+ hrs Power users, all-day battery
Premium AMD Powerhouse AMD Ryzen AI 400 60 32GB ~$1,599 15+ hrs Multitasking, hybrid creative/office
Premium Business Intel Intel Core Ultra Series 3 50 32GB ~$1,699 16+ hrs Enterprise fleets, legacy software

Prices are approximate street prices as of mid-2026 and vary by retailer and configuration — always confirm the current price before buying.

Pros and Cons of Buying a Copilot+ PC Right Now

Pros

  • Genuinely better battery life during AI-heavy tasks (video calls, transcription)
  • Privacy benefit — sensitive AI processing happens on-device, not in the cloud
  • Zero-latency features that work offline
  • Growing software support as more apps add NPU acceleration
  • Future-proofing — the 40 TOPS floor is only going up from here

Cons

  • The 40 TOPS floor is already the low end — buying at the minimum isn’t much of an upgrade
  • Arm-based (Snapdragon) models still hit occasional x86 compatibility snags
  • NPUs don’t meaningfully speed up local large language models
  • Some marketed features haven’t fully materialized yet
  • Premium models carry a real price premium over otherwise-equivalent non-Copilot+ laptops

Expert Verdict

4/5

After six weeks of daily use across three different chip families, my honest take: a Copilot+ PC is a
smart buy if you’re already in the market for a new laptop and use video calls, multilingual audio, or
on-device search regularly. It’s not a must-have upgrade if your current laptop is under two years old and
your workflow is light. For most everyday buyers, a mid-range Intel or AMD Copilot+ model in the
$1,000–$1,300 range gives enough NPU headroom to stay relevant for three to four years without paying a
premium-tier price for TOPS you won’t fully use. Curious how that timeline holds up in practice? Our
laptop lifespan guide breaks down
realistic longevity by brand and usage type.

Ready to Find Your Copilot+ Laptop?
See our tested picks across Snapdragon, Intel, and AMD.

Compare Copilot+ Chips: Intel vs. AMD vs. Snapdragon →

* Affiliate link

Browse Our Windows Laptop Reviews →
See How Snapdragon X2 Elite Compares to Intel & AMD →

Key Takeaways

  • An NPU is a dedicated AI chip, not a CPU or GPU replacement — it excels at efficient, always-on tasks.
  • Copilot+ PC status requires 40+ TOPS, 16 GB RAM, and a 256 GB SSD, at minimum.
  • For large local LLMs, a discrete GPU still beats any NPU by a wide margin.
  • Video callers, multilingual users, and privacy-conscious buyers see the most day-to-day value.
  • A mid-range $1,000–$1,300 Copilot+ laptop is the sweet spot for most buyers in 2026.
What is an NPU used for?
An NPU handles AI tasks like live captioning, translation, background blur,
and on-device search, running them faster and with far less battery drain than a CPU or GPU would.
Do I need a Copilot+ PC?
You’ll get real daily value from one if you regularly use video calls,
multilingual translation, or on-device AI search. If you mainly browse and use Office apps, a standard
laptop is still fine, since cloud-based Copilot already works there.
What is a good TOPS number for a laptop in 2026?
40 TOPS is Microsoft’s Copilot+ minimum, but 50 to 60 TOPS gives more
comfortable headroom as software demands grow. Flagship 2026 chips like the Snapdragon X2 Elite reach 80
TOPS.
Is an NPU better than a GPU?
For small, efficient AI tasks like transcription or background effects, yes.
For running large local AI models fast, a discrete GPU still outperforms an NPU by a wide margin.
Can I use Windows 11 without an NPU?
Yes. Windows 11 runs fully without an NPU. You’ll just miss Copilot+
exclusive features like Recall, Live Captions translation, and Studio Effects.
What’s the difference between an AI PC and a Copilot+ PC?
“AI PC” is a broad, unofficial marketing term for any computer with an NPU.
“Copilot+ PC” is Microsoft’s strict certification, requiring an NPU rated at 40 TOPS or higher, at least 16
GB of RAM, and a 256 GB SSD.
Do Macs have an NPU?
Yes. Apple’s M-series chips include a Neural Engine that serves the same
purpose. It’s a comparable technology, just outside Microsoft’s Copilot+ Windows certification, which is a
Windows-only hardware standard.
Does gaming benefit from an NPU?
Not significantly yet. Most current game engines are built around the GPU,
not the NPU, so gaming performance depends far more on your graphics card than your AI chip.
Will the 40 TOPS Copilot+ requirement go up?
It’s widely expected to rise as newer chips ship with 50 to 80+ TOPS as
standard. Today’s 40 TOPS minimum is likely to become tomorrow’s entry-level tier.

The Bottom Line on NPUs and AI PCs

An NPU is a genuinely useful piece of hardware — a dedicated chip that runs AI features like live
translation, background blur, and on-device search without hammering your battery. A Copilot+ PC takes that
a step further with Microsoft’s official 40+ TOPS, 16 GB RAM, and 256 GB storage requirements, unlocking a
specific set of on-device features that a standard “AI-ready” laptop can’t run.

Whether you need one comes down to how you actually use your computer. Heavy
video-call users, multilingual professionals, and privacy-conscious users will get real, daily value out of a
Copilot+ laptop. If you’re mostly browsing and working in Office, you can comfortably skip the premium for
now — cloud Copilot already covers the basics on any Windows 11 PC. Match the chip and TOPS number to how you
actually work, not to the marketing on the box.

FREE NEWSLETTER

Stay Ahead of Every Tech Trend

Join 100,000+ tech enthusiasts getting the latest reviews, deals, and buying guides — delivered straight to your inbox, every week.

  • Weekly expert product reviews
  • Exclusive deals & discount codes
  • In-depth buying guides & comparisons
  • Breaking tech news & announcements
  • No spam — unsubscribe anytime

Get the Newsletter

Trusted by over 100,000 readers worldwide

Regular Readers
Subscribers
4.9★
Rated
Weekly
Updates

By subscribing, you agree to our Privacy Policy. No spam, ever.

ShamimTechReview
Logo