路透：字节、阿里、腾讯合计下单超300万张英伟达RTX PRO 5500合规显卡（标价超350亿美元），但英伟达称只能交付约5%–10%

Chinese tech giants ByteDance, Alibaba, and Tencent have collectively ordered more than 3 million units of Nvidia's new RTX PRO 5500 graphics cards, representing a total value exceeding US$35 billion, as reported by Reuters.

Designed as workstation chips rather than data center processors, these units are one of the few high-performance options compliant with US export controls. However, with total orders from Chinese buyers topping 6 million units and global production capped at 3 million cards through 2027, Nvidia has signaled that it can fill only a small fraction of requested deliveries.
At the same time, manufacturing partner Taiwan Semiconductor Manufacturing Co. (TSMC) will be redirecting foundry capacity toward Nvidia's next-generation Rubin architecture in the upcoming years, creating severe supply constraints for compliant chips.
The policy dilemma: sales vs. containment
The massive demand from Chinese buyers highlights an intense ongoing debate within Washington regarding hardware controls. US chipmakers including Nvidia argue that maintaining compliant export channels ensures foreign AI ecosystems remain dependent on American technological standards. This strategy also generates massive global revenues that chipmakers can funnel directly into domestic US research and development, while contributing significant revenues back to the US government.
Conversely, national security hawks on Capitol Hill continue to push for tighter trade restrictions on chips, arguing that any high-performance hardware sales risk accelerating foreign military and technological capabilities.
Firms such as OpenAI back export constraints. "We want the US to continue to have the most powerful AI models, maintain and widen its compute advantage, and lead in setting global standards," an OpenAI spokesperson stated. "US labs having a substantial lead in advanced chips is key to achieving those goals, and effective export control policies are one part of that broader strategy."
Domestic alternatives, market shift, and technical bottlenecks
Deprived of Nvidia's highest-end accelerators, such as the B300 and H200, Chinese firms are increasingly expanding their local capabilities. According to data tracking from IDC and reports from DIGITIMES, non-GPU accelerator shipments in China grew by nearly 80% in the first half of 2026, driven by domestic suppliers like Huawei, Alibaba's T-Head, Cambricon, and Baidu's Kunlunxin.
Major AI developers are actively shifting away from foreign dependency. According to investor disclosures by DeepSeek's CEO, the company aims to train its future models on domestic silicon, with Huawei's flagship Ascend 910C processor seeing record demand.
Despite these rapid domestic gains, Chinese chipmakers still face major technical hurdles in manufacturing yields, advanced packaging, and access to high-bandwidth memory. While domestic silicon is increasingly viable for running routine tasks, American chips remain far more cost-efficient for heavy, large-scale model workloads.
Ultimately, strict trade controls have accelerated China's drive toward self-reliance below the cutting edge. Reframing the original sentiment expressed when the policy was unveiled during the Biden administration in 2022, attempting to slow rivals is an incomplete strategy: "A far better strategy is to run faster." To maintain long-term global leadership, policy experts emphasize that the US must prioritize out-innovating competitors at the technology frontier rather than relying solely on trade restrictions to delay foreign advancement.