• NVIDIA (NVDA) CEO Jensen Huang asserts that winning globally requires competing in China, despite the company's declining market share there.
  • The company is re-entering the China market via approved H200 AI chip exports under U.S. restrictions, with a 25% "Trump Fee" applied.
  • Chinese rivals like Huawei and Moore Threads (MOAT) are rapidly advancing, projected to capture 80% of domestic AI chip demand by 2026, squeezing NVIDIA's share from 66% to 8%.

NVIDIA CEO Jensen Huang made it clear this week that conceding the China market is not an option for the semiconductor giant if it aims to maintain its global leadership. "It makes no sense to concede the China market if you want to win globally," Huang stated, according to people familiar with his remarks, as the company navigates a strategic re-entry into China amid tightening U.S. export controls and fierce local competition.

Efforts to regain footing in China have hit a snag, with NVIDIA's market share plummeting from near-total dominance to just 8% projected by 2026, as Chinese hyperscalers like Alibaba (BABA), Tencent (TCEHY), and ByteDance pivot toward domestic alternatives. Without a deal, the company risks being sidelined in one of the world's largest AI markets. Recent U.S. policy shifts under the Trump administration have allowed exports of H200 AI chips to China, albeit with a 25% fee and restrictions to non-sensitive sectors such as research, unlocking pent-up demand. ByteDance, for instance, has ordered 100 billion yuan (approximately $14 billion) in H200 chips for 2026, marking a 17.6% year-over-year increase, with upfront payment models deployed to mitigate geopolitical risks.

Industry-specific elements are at play here. Filing deadlines for export approvals have been met, but China's "new national system" mandates domestic chips in key sectors, imposing limits on H200 usage and launching antitrust probes into NVIDIA's Mellanox unit. "What institutional investors like us are really focused on is regulatory stability," said one anonymous source in the semiconductor sector, echoing sentiments from earlier conferences. Meanwhile, Chinese rivals are not standing still. Huawei's Ascend 950 and Moore Threads' Huashan GPUs are advancing rapidly, with some metrics rivaling H200 performance, albeit with higher power consumption. Hyperscalers like Baidu (BIDU) and Alibaba are also developing custom silicon, aiming to release five processors by 2030.

Human touches add depth to the narrative. Attempts to reach NVIDIA for further comment were unsuccessful, but analysts note that the company's resilience stems from its global pivots to Southeast Asia, Europe, and India. "It's a great country to invest here because there are a lot of very good companies and the market here is not as competitive as other markets," paraphrased a private equity executive familiar with tech investments, highlighting the nuanced landscape. The public debate in China focuses on self-reliance versus interdependence, with local media acknowledging NVIDIA's "benchmark" status despite the rivalry.

Looking ahead, the short-term outlook sees NVIDIA securing orders through adaptations like GDDR7 chips, but its share is expected to remain low. Long-term success hinges on U.S.-China diplomacy and the pace of China's self-reliance drive—experts predict a hybrid market where NVIDIA carves out a niche in premium AI segments. Risks loom, from SMIC's 7nm manufacturing limits to potential retaliatory probes, but analysts see the company's global strategy as a buffer. In related developments, the U.S. CHIPS Act is diversifying supply chains with TSMC (TSM) expansions, while AMD (AMD) and Intel (INTC) face similar export curbs, underscoring the broader semiconductor interdependence amid geopolitical tensions.

Correction: An earlier version misstated the year for ByteDance's H200 chip order; it is for 2026, not 2025.