The definition of an AI PC is now even muddier

Computex Analysis The dominant theme at this year’s Computex conf in Taiwan is that tens of millions of “AI PCs” will sell this year, and more the year after. But despite all the enthusiasm, the qualities of an AI PC have become even more uncertain – and clarity is many months away.

The Register first became confused by the term “AI PC” earlier this year when Intel was bandying it about, so we asked how the silicon giant defined the term and were told it described any laptop packing a Core Ultra CPU that includes an NPU. The moniker was not, however, used by Intel to describe a more comprehensive bill of materials and qualities in the same way that it used brands like “Centrino” or “Ultrabook” to outline a set of specs to which hardware manufacturers were required to conform.

A couple of weeks ago Microsoft muddied the waters by ignoring “AI PC” and introducing the term “Copilot+ PC” to describe machines with an NPU that perform at 40 TOPS or more.

At Computex this week, CPU- and PC-makers have used the terms “AI PC” and “Copilot+ PC” interchangeably.

Developments at the conference suggest the 40 TOPS definition won’t be helpful for long. Intel, for example, used Computex to reveal that its forthcoming Panther Lake SoCs will boast stats that “dwarf” the 48 TOPS produced by its newly announced Lunar Lake processors. Qualcomm and AMD have assured attendees they’re just getting started with NPU-packing processors.

Further complicating matters is that “AI PC” and “Copilot+ PC” are terms that describe only a PC’s processor – an obvious limitation, as myriad components can influence a machine’s performance. Further, NPUs are measured in different ways because they’re tuned to perform different sorts of math. Some are rated on INT8, others on FP16 … yet all get a TOPS score that a buyer is being asked to understand when comparing machines.

Which means a year from now a Panther Lake unit with perhaps 100 TOPs could enjoy the same branding as a 40 TOPS unit. Pity the punter or enterprise PC procurement posse that goes shopping under those circumstances.

Intel CEO Pat Gelsinger thinks this is not good for buyers.

In a Q&A session at Computex yesterday, he said “AI PC benchmarks are needed. We don’t have proper comparisons yet.”

That matters to buyers, but also to Intel as it faces a new and boisterous competitor in the form of Qualcomm, which is happily telling world+dog its new Elite X SoCs are the finest PC powerplants on the planet.

Gelsinger yesterday rubbished that claim.

But as things stand, he’ll struggle to prove it to most would-be buyers.

The Intel CEO suggested it could be a year before comparative performance of AI PCs is easy to understand.

“It will be six months or a year before there is any coherence in the benchmarking space,” he predicted, calling for the Standard Performance Evaluation Corporation to deliver certainty by developing a benchmark.

It’s hard not to agree with Gelsinger, as AI PCs are being hailed as the biggest thing to happen in tech since the internet went large in the late 1990s. In the words of Microsoft CEO Satya Nadella, they mark the moment at which “Computers can understand us instead of us having to understand computers.”

But as things stand, understanding how well an AI PC will be able to understand you is almost impossible to understand.

And that’s a problem for the entire AI ecosystem. It means the devices people will buy to experience the latest thing may disappoint.

Which leaves us all in a situation where the PC industry is telling the world its extreme cleverness has delivered its finest-ever products while it can’t even agree on what they should be called or properly differentiate the machines.

Buyers need more than hype that these machines are marvellous.

We all deserve better – from AI PCs, and the AI that will run on them. ®