AI Without the Internet: Why Everyone Suddenly Talks About It

Whenever we hear “AI”, we picture the internet. Open ChatGPT, type a question, and the answer comes back from a server somewhere. Fair enough — that’s how it has always worked.

But something is changing. Chips in laptops and phones now include a dedicated part called the NPU. Simply put: a small brain designed to run AI. That means some AI work can happen right on your device — no internet, no data sent anywhere. It’s called on-device AI.

Why does this matter? Not for one reason, but four. And this isn’t my invention — it’s what the people building it say.

First: cost. This is the biggest reason, according to Qualcomm. Chris Patrick, one of Qualcomm’s executives in charge of phone chips, once put it like this: if you only use a chatbot once a week, why bother running it on the device? Just use the cloud. But once AI becomes part of daily life — translating, transcribing, an always-on assistant — the math changes. Cloud compute bills balloon. Gemini’s token consumption alone grew sevenfold in a year, across a billion users. Every AI task that runs on the device’s chip is a task nobody has to pay the cloud for.

Second: privacy. Microsoft executives say on-device processing can keep sensitive information on the device — never sent, never stored on someone else’s server. Data that never leaves your laptop can’t leak along the way. And data regulations are only getting stricter everywhere.

Third: speed, and working offline. AI running on the device responds in milliseconds — far faster than a round trip to a server. And it keeps working with no signal at all. For things like background blur on a video call or medical assistive tools, even a fraction of a second of delay is unacceptable.

Fourth: battery. Microsoft insists its AI runs on the NPU, not the main processor — so the battery doesn’t drain fast. The NPU is purpose-built for AI math, so it sips power.

But two honest caveats.

First, the future isn’t “all AI moves to the device”. Even Satya Nadella, Microsoft’s CEO, says what’s coming is a combination of both: some on the device, some in the cloud. Giant models that need constant updating stay in the cloud; routine, sensitive, or latency-critical tasks run on the device.

Second, the “AI PC” label that was hyped so loudly isn’t selling. Microsoft and Qualcomm just stopped using the term “Copilot+” — because the data shows people buy laptops because they need a new laptop, not because they’re drawn to an AI label. The hardware keeps moving forward, but the marketing no longer leans on the AI badge.

So what’s the urgency? On-device AI matters at four points — cost, privacy, speed, and battery. But as a complement, not a replacement. And for us ordinary users, it simply means this: more and more smart things our laptops and phones can do — even with no internet at all.

Chips like this evolved into the “AI brains” of modern devices — from graphics processors to dedicated NPUs. Photo: Wikimedia Commons.