Part 2 – Inside the AI PC: CPU, GPU, and the Rise of the NPU

Key takeaways:

• • Not all AI-capable computers are the same. AI PCs, Copilot+ PCs, and AI workstations serve different roles.
  • As AI shifts to the edge, NPUs on modern PCs enable faster, more secure and more efficient on-device experiences.
  • Matching user needs to the right device class improves performance, productivity, and cost control.

When users were asking, “Why is my PC so slow?” during back-to-back Teams calls, the industry didn’t respond with a slogan. It responded with silicon.

On PCs with a discrete GPU—for example, an NVIDIA RTX—Teams could offload its video workload to the GPU. The impact was dramatic:

• • CPU utilization dropped from ~35% to around 8%, with peaks staying under 24%.
  • User experience improved immediately: smoother video, fewer “resource issue” errors, and better responsiveness overall.

This was a turning point. It proved that offloading specialized workloads to dedicated silicon could transform everyday experience. From there, the architecture of the AI PC started to crystallize.

Why not “just a GPU”?

If GPUs are so powerful, why not build a PC around just a GPU?

Because CPUs and GPUs thrive in a symbiotic partnership, each excelling where the other struggles.

• • The CPU handles complex, branching instructions, orchestrating diverse tasks and maintaining overall system control. It’s flexible and general-purpose.
  • The GPU is optimized for simple, repeating operations executed massively in parallel—perfect for graphics and many AI workloads. It’s rigid but incredibly fast when the problem fits its model.

Think of the CPU as the master chef in a fine-dining restaurant. You can order almost anything: custom dishes, tailored pairings or off-menu specials. The chef handles complexity and nuance.

The GPU, in contrast, is like a bustling school cafeteria. Options are limited—burger or pizza, apple or orange—but it serves hundreds or thousands of meals at high speed. When a request breaks the mold and needs special handling, it gets escalated back to the master chef (the CPU).

Together, they keep the kitchen humming. A modern PC needs both roles to function well.

Enter the NPU: a new specialist

It didn’t take long for CPU manufacturers to push the idea further: What if we integrated a specialized, parallel AI engine directly into the processor package?

That’s how the Neural Processing Unit (NPU) was born.

While NPU development began around 2016, modern NPUs for PCs didn’t go mainstream until 2024. Today’s NPUs are:

• • Specialized microprocessors designed specifically to accelerate AI and machine learning workloads.
  • Optimized for matrix operations and tensor math used in neural networks, rather than general-purpose compute.

By offloading AI tasks—like vision, audio, and language inference—from the CPU, NPUs enable faster, more efficient, and more power-conscious AI features right on the device.

Measuring AI horsepower: what are TOPS?

You’ll often hear NPUs measured in TOPS, or Tera Operations Per Second—trillions of operations per second. “Tera” represents 10¹² (1,000,000,000,000). The higher the TOPS rating, the more operations the NPU can perform each second.

Returning to the cafeteria analogy: A 1 TOPS NPU would be like serving 1 trillion hamburgers per second—a staggering amount of throughput.

For Copilot+ PCs, Microsoft currently requires a minimum of 40 TOPS of NPU performance. That’s the bar for enabling a new class of on-device AI experiences: richer Windows Studio Effects in Teams, faster local copilots, and more responsive creative tools that don’t immediately fall back to the cloud.

Why an NPU instead of a GPU (sometimes)?

If GPUs are so strong, why bother with NPUs at all?

The short answer: power efficiency and always-on AI.

An NPU typically draws just 2–3 watts, delivering up to 80% lower energy consumption compared to running similar AI tasks on a CPU or GPU. That efficiency translates into higher performance per watt and significantly extended battery life—often 20 hours or more in modern systems.

The result is a PC that feels more responsive while running more AI features, more often, without killing your battery.

In Part 3, we’ll explore where GPUs still shine, why AI workstations remain critical, and how combining CPU + NPU + discrete GPU—across systems like XPS for creators, Dell Pro for advanced professionals, and Dell Pro Max for heavy AI users—sets you up for an agentic AI future.