NVIDIA CEO Jensen Huang
AsianFin -- Over the past week, Nvidia CEO Jensen Huang has stepped up his commentary on China's progress in AI chips, responding directly to Huawei founder Ren Zhengfei's recent remarks downplaying concerns over China's chip capabilities.
Ren had argued that China's AI development is not hindered by lagging chip performance, pointing out that techniques such as stacking and clustering can effectively offset hardware disadvantages. "There's no need to worry too much about the chip issue," Ren said.
"Single-chip performance isn't everything—what matters is delivering the required computing results. Huawei isn't as strong as some claim; in fact, we remain a generation behind the U.S. in single-chip performance."
Speaking at an event on June 13, Huang acknowledged that Huawei's chip technology is trailing Nvidia's, but he echoed Ren's assessment of parallelism in AI computing. "Our technology is a generation ahead," Huang said. "But AI is inherently parallel—if one computer isn't powerful enough, you just use more. Ren's point is that China has sufficient energy and scale to make that work."
Huang added that while China may no longer rely on U.S. chips, its own tech stack could be "good enough" to support domestic and even international markets.
If the U.S. chooses not to participate, Huawei can meet the needs of China and beyond, he said.
The exchange comes as Nvidia grapples with the fallout from tightened U.S. export controls, which have effectively shut off sales of its H20 AI chips to China. In late April, U.S. authorities extended licensing requirements on high-performance AI chips indefinitely. This move abruptly halted Nvidia's shipments, including about $18 billion in orders for the H20.
The ban has cost Nvidia an estimated $15 billion in lost revenue and inventory writedowns in the first half of 2025. In the fiscal first quarter alone, Nvidia reported $4.5 billion in inventory-related charges and an additional $2.5 billion in lost H20 sales. Analysts expect a further $8 billion impact in the second quarter.
Despite these headwinds, Nvidia's latest earnings remain strong. For the fiscal first quarter of 2026, the company posted $44.1 billion in revenue—up 12% sequentially and 69% year-over-year—with a net profit of $18.8 billion. Data center sales surged 73% to $39.1 billion, driven by robust demand for Nvidia's B100 and B200 GPUs. Excluding the H20 write-down, non-GAAP gross margin would have reached a record 71.3%.
Still, Huang has made clear that Nvidia no longer includes revenue from China in its forward-looking guidance. Nvidia has adjusted the financial forecasts to exclude China, he told investors. If export controls are lifted, that would be a positive surprise, he added.
While Nvidia exits the Chinese market, Huawei is stepping into the vacuum. According to Huang, Nvidia's market share in China has dropped from 90% to 50%, largely due to Huawei's rise. "Whether with or without American chips, China's AI will continue to develop," Huang warned. "The question is whether it will run on American platforms."
Huawei's 2024 annual report shows it is investing heavily to close the gap. The company spent 179.7 billion yuan ($24.8 billion) on R&D—20.8% of revenue—bringing its 10-year cumulative R&D spending to over 1.24 trillion yuan. Huawei posted 862.1 billion yuan in 2024 sales and a net profit of 62.6 billion yuan, with operating cash flow up 26.7% to 88.4 billion yuan.
The company now has over 208,000 employees, including 113,000 in R&D, accounting for 54.1% of the total headcount. Huawei's Ascend chip line and its cloud-based Pangu large models are increasingly central to its AI strategy.
While Huawei has not led in global AI benchmarks, its focus on industrial AI is yielding results. The company's Pangu 5.0 model, launched in 2024, targets over 400 real-world applications across 30 industries, from pharmaceuticals and railways to coal mining and weather forecasting.
Ren has emphasized that real-world AI adoption in China depends more on industry experts—like power, coal, and pharmaceutical specialists—than IT professionals. "In China, it's the domain experts who drive AI breakthroughs," he said.
Huawei is also collaborating with Chinese LLM leaders like DeepSeek. The DeepSeek-R1 model now supports efficient inference on Huawei's Ascend chips through quantization techniques that cut memory usage. DeepSeek V2 reportedly outperforms rival models on Ascend, signaling rapid progress in China's AI software-hardware integration.
As it loses ground in China, Nvidia is accelerating its transition from a GPU maker to a full-stack AI infrastructure company. At Computex 2025, Huang introduced NVLink Fusion—a modular interconnect system that lets customers integrate their own CPUs into Nvidia's GPU ecosystem. This approach mimics custom ASIC development, enabling broader adoption of Nvidia infrastructure in global enterprise and cloud settings.
Huang's messaging has become increasingly pointed. In May, he warned that U.S. export controls are backfiring. China's AI market will not disappear, he said. It must compute, and if it doesn't run on American chips, it will run on Chinese chips, he noted.
According to industry insiders, Huang's openness may also reflect an effort to ease mounting regulatory scrutiny from U.S. antitrust agencies. By acknowledging Huawei's progress, Huang potentially deflects accusations of monopolistic dominance amid a growing tech cold war.
Nvidia's current valuation stands at $3.46 trillion. Despite geopolitical risks, the company's ability to deliver strong margins and sustain global leadership in AI computing remains intact. However, the once-lucrative Chinese market—worth over $17 billion annually to Nvidia—is increasingly slipping from its grasp.
If U.S. export controls remain in place and China's AI infrastructure continues to scale, Jensen Huang's strategic pivot may come to define the next era of global AI competition—not just in technology, but in markets, ecosystems, and geopolitical alignment.
