I would focus on TSMC’s capacity execution and Nvidia’s order trajectory rather than wait for perfect clarity on AI capex returns.

TSMC’s strong revenue, profit and 58.7% operating margin suggest AI demand is still translating into real earnings, not merely investment promises. Higher capex is therefore not automatically bearish if new capacity remains highly utilised and generates attractive returns.

The bigger risk is downstream. If hyperscalers keep spending heavily but AI monetisation fails to catch up, Nvidia orders could eventually weaken and TSMC’s new capacity could face lower utilisation.

I would watch three signals closely: Nvidia order growth, TSMC advanced-node utilisation and hyperscaler AI returns. As long as all three remain healthy, the AI semiconductor thesis remains credible. The share-price declines look more like concern over future returns than evidence that current demand has broken.