Key Points
- ByteDance, Alibaba, and Tencent have ordered over 3 million Nvidia RTX PRO 5500 GPUs valued at approximately $35 billion (48 trillion KRW).
- The story is drawing significant attention in Korean tech and business news because Nvidia can only supply a fraction of the demand as orders exceed 6 million units against a tight supply chain.
- Chinese firms are pivoting to workstation-grade GPUs compliant with U.S. export limits specifically to power AI inference rather than resource-heavy model training.
Summary Three of China's largest technology conglomerates—ByteDance, Alibaba, and Tencent—have reportedly placed massive orders exceeding 3 million units for Nvidia's RTX PRO 5500 graphics processing units (GPUs). With each unit priced at around $11,400 (roughly 15 million KRW), the total order value reaches an astonishing $35 billion (approximately 48 trillion KRW), highlighting China's voracious appetite for AI hardware despite stringent U.S. export restrictions.
Unlike ultra-high-end enterprise processors designed for massive data centers, such as Nvidia's latest Blackwell-based B300 (which faces a total ban on sales to China) or Hopper-based H200 chips (which require individual export licensing), the RTX PRO 5500 is a workstation-grade GPU. As it stands, it represents the most capable Nvidia processor that Chinese firms can legally acquire without violating current U.S. export control thresholds. The hardware offers performance roughly on par with the custom-modified RTX 6000D released previously. While Chinese tech firms initially hesitated to purchase these chips due to less favorable price-to-performance ratios while hoping for loosened sanctions, escalating computational demands have forced them to commit capital to secure available legal stock.
However, satisfying this monumental demand appears nearly impossible. Sources indicate that Nvidia may only be able to fulfill 5% to 10% of the requested volume. While Nvidia's total production capacity for the RTX PRO 5500 is estimated at around 3 million units for 2027, cumulative orders from Chinese buyers have already surged past 6 million units. The bottleneck is poised to tighten further as Nvidia shifts TSMC production lines toward next-generation Rubin architecture data center platforms in 2027.
The acquisition underscores how Chinese tech giants are adapting under regulatory pressure. Rather than training colossal foundation models from scratch—which demands premier data-center chips—these companies plan to deploy the workstation chips primarily for AI inference tasks, such as generating chatbot responses from already trained models, which requires comparatively less raw compute. Even after Nvidia CEO Jensen Huang noted that export curbs virtually reduced the company's once-dominant Chinese data center market share to zero, Chinese tech giants continue to find pragmatic, workaround paths to keep their AI engines running.
Sponsored · Ad