The “AI PC” scam — Dell’s CES 2026 wake-up call

Something shifted at CES 2026. For the first time, the people selling you “AI PCs” admitted — in public, on a panel — that virtually nobody actually wants the AI features they’ve been marketing for eighteen months. Dell said it. Others nodded along. The room got quiet in a way that feels significant in retrospect.

You’ve been paying $300 to $500 more for a chip — the NPU — that has no meaningful consumer application today. Not one. And the industry knew it the whole time.

What actually happened at CES 2026

The pitch was always the same: local AI processing would make everything faster, keep your data private, and transform how you work. It sounded good. It still sounds good. The problem is that after eighteen months and billions in R&D and marketing, most users still can’t name a single task where the NPU made a real difference in their day.

Dell’s admission wasn’t a scandal — it was just honesty arriving late. Consumer interest in AI-specific hardware features is, in their own words, practically non-existent. We were sold a solution to a problem nobody had.

NPU marketing vs reality — billions in spend, zero breakthrough apps to show for it.

The “AI tax” and where your money actually goes

It gets worse when you follow the money. The high-performance silicon that enthusiasts and power users actually want — the chips that make a real difference in gaming, rendering, and workstation tasks — is being quietly redirected. Not to you. To enterprise data centers, where margins are higher and demand is insatiable.

We covered this in detail in our companion piece on NVIDIA’s Blackwell silicon shortage. The “AI PC” label is part of the same system. It justifies premium pricing on consumer hardware while the parts you’d actually use go somewhere else. You’re not buying better performance — you’re subsidizing enterprise compute.

Privacy or prison? The lock-in nobody talks about

The privacy angle was always the most appealing part of the “AI PC” pitch. Process everything locally. Keep your data off the cloud. Own your workflow. It’s a compelling idea — and for some people, it’s the main reason they considered upgrading.

Here’s the uncomfortable reality: many of the “local” AI features in 2026 machines still require cloud verification to stay active. Some require subscriptions. The hardware is yours, technically. But the full functionality? That’s increasingly rented.

So where are all the apps?

If you bought an AI PC today, here’s what you’d get for that NPU premium: a Copilot key you could have remapped yourself, background blur on video calls, and a handful of demos that run smoother than they do on your GPU — barely. That’s it. That’s the ecosystem after eighteen months of “this changes everything” marketing.

Developers aren’t rushing to build for NPUs because the install base is fragmented, the tooling is immature, and consumers aren’t asking for it. The killer app doesn’t exist yet. It may never exist in a way that justifies what you paid. And the community is starting to notice — the “AI PC” badge has gone from aspirational to a warning sign almost overnight.

What we should be demanding instead

This isn’t about being anti-AI or anti-progress. Local AI processing could genuinely be valuable — eventually. The problem is being charged for a future that hasn’t arrived, while the hardware you actually need today gets more expensive and harder to find because it’s been redirected elsewhere.

Buy on performance. CPU power. GPU capability. RAM. Storage speed. These are the things that will make your machine faster tomorrow, next year, and in five years. If meaningful NPU applications emerge later, your GPU will most likely handle them just fine. Don’t pay today for a promise that has no delivery date.

The verdict