EXPERT Q&A

As tensions between the U.S. and China in technology continue to escalate, the semiconductor sector has emerged as a crucial element in a high-stakes geopolitical confrontation.

The recent wave of U.S. export bans has made even Nvidia’s H20 chip—previously seen as a compliant workaround—unavailable for Chinese purchasers. These changes provoke urgent inquiries regarding the trajectory of worldwide chip design, innovation, and market access.

To shed light on these swiftly changing circumstances, Valeria Bertacco provides her perspectives. A prominent authority in computer architecture and semiconductor design, she holds the position of Mary Lou Dorf Collegiate Professor of Computer Science and Engineering at the University of Michigan and leads the Applications Driving Architectures Research Center, backed by a consortium of leading semiconductor companies.

She investigates how U.S. policies are transforming the global semiconductor framework—and what implications this holds for innovation, energy efficiency, and the future of enterprises like Nvidia:

Nvidia developed the H20 chip specifically to adhere to U.S. regulations. Now, even this product is said to be unavailable to Chinese companies. What does the updated ban indicate about the shifting landscape of U.S. tech policy towards China?

The position of the U.S. administration regarding the People’s Republic of China has become increasingly challenging for the semiconductor sector, which is inherently a global industry—as demonstrated by the supply chain disruptions we all witnessed during the COVID-19 pandemic. Numerous tech firms are tirelessly striving to navigate the evolving policy landscape to comply with U.S. regulations while attempting to reduce their loss of customers in the global arena.

Nvidia’s CEO Jensen Huang mentioned in Beijing that Nvidia will ‘unwaveringly serve the Chinese market’. How feasible is this, considering the ongoing congressional investigation and tightening export restrictions? What forms could “serving” China take at this moment? Perhaps another variant of H20?

Currently, the hardware targeted by restrictions in certain international markets is solely focused on the highest-end chips, which represent a minor segment of the AI hardware market. Mid and lower-end hardware will still be accessible in those markets, as will edge-AI hardware, all of which are vital sectors well-served by U.S. firms and startups.

A significant requirement in the AI hardware domain is computational efficiency and energy availability. Current projections indicate that the energy consumed in AI computations by the decade’s end could amount to 21% of global energy consumption. Limiting a large segment of the AI community to lower-tier hardware has spurred advancements in AI efficiency. Recently, we observed a clear instance of this trend in the DeepSeek LLM model, which utilizes significantly fewer resources than previous large AI models.

Can U.S. chip manufacturers such as Nvidia and AMD sustain their innovative advantage if they permanently forfeit access to the Chinese market? How might reduced income, scale, or varied customer demands influence their long-term research and development?

Given that sales in these markets often finance technology companies’ R&D efforts, there exists a tangible risk that these companies may need to curtail their internal innovation and research investment. Support from the U.S. government could alleviate this potential loss of opportunities in innovation through research initiatives focusing on AI technology.

Another critical concern pertains to the accessibility that students from nearly every nation have to U.S. universities, with subsequent pathways into U.S.-based technology roles. For years, U.S. tech leadership has attracted the finest talent globally, training them within our academic institutions and propelling them into our tech sector. Should this flow be disrupted due to policy modifications, we would undoubtedly observe significant consequences on U.S.-led technological innovation, especially within the semiconductor and software realms.

Is there a possibility that excessively stringent restrictions could drive Nvidia or similar companies to relocate parts of their operations or innovation hubs overseas—perhaps to less restricted markets?

It’s possible. I do not believe this is a consideration at present, but the evolution of policy restraints is uncertain. If such shifts were to occur, I anticipate they would manifest gradually and incrementally. Additionally, reversing them would likely be a slow process.

How will this impact prominent Chinese tech companies like Alibaba, Tencent, ByteDance, and DeepSeek? Could a prolonged chip shortage hinder their innovation?

In the near term, it is likely to do so. However, the shortage may inspire innovators to pursue alternative avenues for new solutions—pathways that might not have been considered if high-end AI chips were readily available. For example, DeepSeek demonstrated that high-quality LLM training can be achieved with ten times less computing effort than was previously assumed necessary. Their breakthroughs may not have been explored if the looming challenge of restricted access to premium AI chips had not been present.

What implications does this have for China’s domestic alternatives, such as chips from Huawei or Cambricon? Do you believe they can scale sufficiently to fulfill domestic AI demand?

The PRC government possesses the resources to invest in the expansion of its semiconductor industry and to support its innovation initiatives, seeking to establish itself as a leader in the global semiconductor landscape. Restrictions from U.S. policies may accelerate their endeavors in this direction by intensifying unmet market needs within China.

With Nvidia now producing chips locally through TSMC in Phoenix and constructing supercomputers in Texas, how might this shift towards U.S.-based production alter the global AI supply chain? Additionally, what impact could it have on Taiwan’s strategic role in global chip production?

The enhancement of semiconductor manufacturing capabilities within the United States, propelled by TSMC and other manufacturers such as Intel, Micron, Samsung, and Global Foundries, represents a highly favorable trend that strengthens the U.S. against future vulnerabilities in the supply chain. These new developments are outcomes of the strategic CHIPS Act’s investments initiated in 2022. I do not foresee these trends diminishing Hsinchu and Taiwan’s significance within the global semiconductor context, as the demand for chip production continues to vastly surpass supply.