Nvidia chief says US chip curbs speed up China’s domestic semiconductor push

Speaking at the company’s Berlin event, the chief executive argued Washington’s trade restrictions hastened Beijing’s push to build its own semiconductor industry, and he predicted memory chips will not face a glut before 2029.

Nvidia chief executive Jensen Huang has acknowledged that American restrictions on chip exports have inadvertently strengthened China’s determination to develop its own semiconductor capabilities, describing the policies as a driving force behind Beijing’s recent momentum.

In an interview on the sidelines of the company’s GTC conference in Berlin, Huang said the measures had effectively nudged China toward faster domestic production and greater technological independence. He also played down fears that the memory chip market is heading for oversupply.

Curbs and workarounds

Under existing US rules, advanced processors, including Nvidia’s most powerful AI graphics chips, cannot be shipped to China. Yet reports suggest that several Chinese organisations have still found ways to tap that computing power, renting capacity from third-party data centres located in South-East Asia. Analysts say such access has been important to the improving quality of AI models built in China.

Huang did not dispute that the restrictions have had an unintended effect. Washington’s actions, he said in substance, had sped up China’s efforts to bring chipmaking under national control, along with the wider industrial backing the sector now receives. In his words, there was “no question” the policy had accelerated those initiatives.

Despite that, the Nvidia chief voiced confidence in the position of American technology companies. He described them as strong and well placed to handle competition from any rival, Chinese or otherwise.

Memory demand to stay strong

On the question of memory, Huang dismissed suggestions that supply will outstrip demand in the near term, indicating that the current surge in hardware spending shows no sign of ending suddenly. He said a glut was unlikely to emerge by 2029.

His reasoning was simple: there is always a use for additional memory. That applies in particular to high-bandwidth memory, the specialised RAM that AI accelerators depend on, which he expects to remain in wide use for years.

Huang also pointed to a technical shift at Nvidia. The company, he said, is the first to bring LPDDR memory, the type found in smartphones, into data centres. He claimed the approach delivers roughly twice the energy efficiency of anything currently available, and that Nvidia intends to keep buying large volumes of it.

The remarks come as the AI hardware industry navigates an uneasy mix of geopolitical limits and booming demand. For Nvidia, China remains both a sensitive market and a source of emerging competition, a tension Huang’s comments appear to acknowledge.