Rising AI Inference Demand Drives CPU Recovery, TSMC Benefits Across Architectures
N.R. Finch
TSMC Chairman C.C. Wei said on the Q2 earnings call that growing AI inference workloads are restoring CPU importance in data centers, and because virtually all CPUs — x86 or RISC-V — are fabricated by TSMC, the rebound translates directly into incremental orders.
AI is shifting from training to inference — why does that help CPUs?
Generative AI's center of gravity is moving from training to inference and agentic AI — systems that autonomously execute multi-step tasks.
Training leans almost entirely on GPUs for raw compute; inference demands more CPU involvement in scheduling, data movement, and pre/post-processing.
This means → CPUs are no longer a sidecar in AI data centers — they are again an essential link in the compute chain.
Why does TSMC benefit, rather than a specific CPU camp?
Wei highlighted a key fact: whether a customer picks x86 or RISC-V, the vast majority of CPUs are fabricated by TSMC.
In plain terms = whichever design wins in the market, the chips still get made at TSMC — the company sits on both sides of the architecture divide.
This reflects a competitive moat tied not to any single architecture but to leading-edge process technology itself.
What does broadening AI-compute demand mean?
For years the market nearly equated AI compute with GPUs — primarily Nvidia's. Now demand is spreading to CPUs, custom chips, and other categories.
TSMC added that it is working closely with CPU clients, offering the most advanced process nodes and ample capacity to capture AI growth.
This means → TSMC's order-growth story is evolving from a single-GPU leg into a multi-category, cross-architecture tailwind — a wider base of demand.
Content is for reference only, not financial advice.