TSMC has excluded CPUs from the growth forecast it publishes for AI accelerators, DigiTimes reports. The decision matters because TSMC's outlook statements move markets and set expectations for the whole silicon supply chain.

By Marcus Bennett

What did TSMC actually do?

TSMC reports expected growth for AI accelerators as a product category. It has now made clear that central processors do not count toward that number. The two chip types sit in different buckets in the company's forecasting.

This is an accounting and definitions change, not a new product. But definitions drive how analysts read a foundry's momentum. When the world's largest contract chipmaker separates CPUs from AI accelerators, every downstream model built on that forecast has to change too.

Why separate CPUs from AI accelerators?

The mechanism is simple. AI accelerators are dedicated math machines. They handle the training and inference workloads behind large AI models. CPUs are general-purpose processors. They run operating systems, application logic and everything else.

Vendors increasingly blur this line. Chipmakers now ship CPUs with AI processing units attached. Marketing copy often calls these parts AI chips. If TSMC counted such processors as accelerators, its growth number would inflate without any change in real accelerator demand.

Excluding CPUs keeps the forecast clean. It measures silicon built specifically for AI workloads, not general processors with AI features bolted on.

What does this change for customers and investors?

For investors, the change affects comparability. A forecast that excludes CPUs is narrower than one that includes them. Analysts who track TSMC's AI-related revenue must now check which definition each figure uses. Apples-to-apples comparisons across quarters may break where definitions shifted.

For chip designers, the signal is subtler. TSMC is telling the market that AI accelerator demand stands on its own. The category does not need CPU volumes padded into it to look strong.

For server buyers, expect clearer product segmentation. Vendors will have less cover to market hybrid CPU parts as full AI accelerators.

The bigger picture

TSMC sits at the center of the AI hardware buildout. It manufactures for the major accelerator designers, and its capacity plans shape the industry's supply. How it defines and forecasts product categories ripples outward to equipment makers, packaging suppliers and cloud capital spending.

A definition change at that scale is not trivia. It resets the baseline that suppliers, customers and shareholders use to judge whether AI demand is accelerating or cooling.

What to watch next

Watch TSMC's next earnings call for the growth figure under the new definition. Watch whether competitors adopt the same split between processors and dedicated accelerators. And watch how analysts revise their models once the narrower category takes hold.

The number TSMC reports will likely be smaller in scope than before. That smaller number may also be the more honest one.