TSMC's cautious capex is averting an AI bubble, says investor

TSMC’s restrained capex isn’t caution—it’s a calibrated hedge against AI infrastructure overbuild. Technically, this delays HBM3e/4 and advanced packaging ramp-ups, forcing NVIDIA and AMD to prioritize architectural efficiency over node shrinks. Compliance-wise, U.S. CHIPS Act mandates have already inflated TSMC’s Arizona fab operating costs by 15–20%; slowing expansion mitigates subsidy-linked execution risk. Samsung may exploit the gap in mid-tier AI foundry services, but yield issues prevent meaningful CoWoS ecosystem disruption. Over the next 12–24 months, expect a deflationary correction: capex concentration intensifies among top-tier foundries, while second-tier players pivot to automotive or industrial MCUs. AI chip valuations will likely reset downward by 20–30% as speculative demand evaporates.