TSMC has excluded CPUs from its AI accelerator growth forecast, a decision that clarifies how the world's largest contract chipmaker defines the market it counts on for its next leg of expansion. The distinction matters because it shapes how investors, customers and rivals read the foundry's outlook for advanced-node demand.

The report comes from DigiTimes, which flagged the classification move without attaching new capacity or revenue figures to it. In plain terms, TSMC will count only AI accelerators — the high-throughput chips built to train and run large models — when it sizes its AI-driven growth. General-purpose CPUs, even those with AI instructions bolted on, sit outside that forecast.

Why does the definition matter?

A growth forecast is only as useful as its boundary. If CPUs were folded into the AI bucket, almost every advanced processor TSMC makes could claim a share of the AI story, and the forecast would say little about genuine accelerator demand.

By excluding CPUs, TSMC narrows the metric to chips whose primary job is AI computation. That gives a cleaner signal on demand for the silicon that carries the heaviest AI workloads.

The takeaway is simple: the foundry wants its AI numbers to measure one thing, not everything.

Where do CPUs sit in TSMC's lineup?

TSMC does not sell CPUs under its own name. It manufactures them for customers such as AMD, Intel through its foundry arrangement, and Apple, whose Mac processors come off TSMC lines. Those chips increasingly include dedicated blocks for AI inference.

Even so, TSMC treats them as a separate product family from AI accelerators, the category that includes the graphics-processing-unit-based chips used for training large language models. The exclusion signals that the foundry sees accelerator demand as a distinct commercial force rather than an extension of the CPU business.

The line between the two product families, in TSMC's accounting, is now explicit.

What does this mean for reading TSMC's outlook?

Forecasts that exclude CPUs are easier to audit. When TSMC later reports AI revenue growth or demand strength at advanced nodes, analysts can map those statements directly onto accelerator shipments rather than a blended figure that also absorbs CPU volumes.

The move also sets expectations for customers. CPU designers gain no reflected glory from the AI accelerator forecast; their demand contributes to TSMC's overall advanced-node business, but not to the AI-specific number the foundry chooses to highlight.

In short, TSMC has made its AI story harder to inflate and easier to track.

How does this fit the competitive picture?

AI accelerators have been the strongest driver of demand for TSMC's leading-edge capacity, and the foundry's advanced packaging operations have become a bottleneck for customers racing to ship accelerator products. A tighter definition of the AI category sharpens the picture of where that pressure sits.

It also gives competitors and customers a clearer benchmark. Any rival claiming equivalent AI-driven growth now faces a comparison against a category that excludes routine processor volumes.

The definition, once applied consistently, becomes a yardstick for the whole supply chain.

What comes next?

DigiTimes did not attach new shipment targets or capacity figures to the revised classification, so the commercial impact will only become visible in TSMC's subsequent guidance and earnings commentary. Watch whether the foundry reports accelerator-driven growth as a standalone figure in coming quarters — if it does, the market will finally have a clean number for the silicon at the center of the AI build-out.

By Tom Whitfield, Chip Dispatch