Yes, AI demand can realistically keep TSMC’s momentum intact into 2026, and the market can still be underpricing its AI leverage, but the upside depends on whether this cycle stays “tight” rather than “normalising”.

Why momentum can stay strong into 2026

TSMC is the bottleneck for leading-edge AI: Most meaningful AI compute still concentrates around advanced nodes (N3/N2, advanced packaging). Even if end-demand fluctuates, the strategic need to secure capacity stays high.

AI is not just GPUs: Beyond NVIDIA/AMD accelerators, AI demand spreads into CPUs, networking silicon, HBM controllers, custom ASICs (hyperscalers), and edge AI. That broadens TSMC’s growth base.

Margins accelerating is a powerful signal: When profits and margins rise alongside revenue, it implies pricing power + high utilisation + richer mix (advanced nodes + packaging). That is the “good kind” of beat.

Is the market still underpricing TSMC’s AI exposure?

Possibly, yes, because many investors still value TSMC like a cyclical foundry, not like the infrastructure backbone of AI. Two reasons:

AI visibility is higher than typical consumer cycles (multi-year capex roadmaps from hyperscalers).

TSMC monetises every winner: It does not need to pick the right AI brand. If the compute ships, TSMC earns.

What could cap the upside (key risks)

AI digestion quarters: A pause in accelerator orders can cause sentiment swings even if long-term demand remains intact.

Geopolitics / policy shocks: This is the biggest “non-fundamental” discount driver.

Valuation re-rating limits: Even with strong fundamentals, the market may refuse to pay “software-like multiples” for a manufacturer.

Bottom line

AI demand is likely sufficient to support TSMC’s growth into 2026, especially if advanced packaging and leading-edge utilisation stay tight. The market may still be underpricing TSMC’s AI exposure, but the re-rating will only fully unlock if investors gain confidence this is a multi-year AI infrastructure cycle, not a single wave.