China faces an uphill battle to wean tech firms off Nvidia (NVDA) chips, according to FT

Reporting of this kind sits at the intersection of export controls and industrial policy, a space where the established pattern is that official substitution targets run well ahead of actual procurement behaviour. The mechanism is software lock-in as much as silicon: switching away from incumbent accelerator ecosystems means rewriting code, retraining engineers and accepting performance penalties, which is why past episodes of forced localisation have tended to show headline loyalty to domestic suppliers alongside quiet continued demand for foreign parts through whatever channels remain open. The distinction worth drawing is between training workloads, where the performance gap has historically been hardest to close, and inference, where domestic substitutes have found readier acceptance. Beijing's track record in this area is directive followed by uneven compliance, so the tells are procurement data, cloud capex disclosures from the large Chinese platforms, and any tightening or loosening of licensing on the US side. For the incumbent supplier, prior rounds of restriction have typically produced product redesigns to fit the rules rather than a clean revenue loss. As a press report without new policy, the signal is about the durability of demand rather than a fresh catalyst.