nvidia ceo warns us lawmakers: huawei ai chips gaining ground

2025-05-02 nvidia

Santa Clara, Friday, 2 May 2025.
nvidia’s ceo, jensen huang, has voiced concerns to us lawmakers regarding huawei’s growing ai capabilities. huang fears that export restrictions on nvidia’s chips to china could inadvertently boost huawei’s competitiveness. this could lead to a global market demanding huawei chips. us restrictions already led to nvidia being forced to halt sales of its h20 chip in china. huawei is now stepping in to fill the gap, preparing to ship ai chips to chinese customers.

Competitive pressures in the ai chip market

Huang’s discussion with the House Foreign Affairs Committee on Thursday, May 1, 2025, highlighted the strategic importance of AI and the necessity of investing in U.S. manufacturing [1][6]. He also reiterated nvidia’s support for the U.S. government’s efforts to promote american technology globally [1]. However, nvidia’s ai chips, crucial for developing advanced systems, have faced export controls since the trump administration [1][2]. This has prompted nvidia to create compliant chips for the chinese market, but these are also facing restrictions [1].

Huawei’s ai chip capabilities and challenges

Huawei is actively working to fill the void left by nvidia in china [2]. The company plans to begin mass shipments of its new generation Ascend 910C AI chips to chinese customers this month [6]. Huawei is also reportedly engaging with chinese tech firms to test the feasibility of its Ascend 910D chip as a replacement for some of nvidia’s high-end products [6]. Huawei’s CloudMatrix 384 AI chip cluster, while powerful, consumes 3.9 times more power than nvidia’s GB200 NVL72, leading to higher electricity costs [3].

Cost and efficiency comparison

The CloudMatrix 384 has a system-wide power consumption of approximately 559 kW, including networking and storage, compared to nvidia’s GB200 NVL72 at 145 kW [3]. This makes nvidia’s solution 2.3 times more energy-efficient than huawei’s [3]. Furthermore, the huawei software system demands more maintenance from experienced engineers, increasing operational labor costs by three to five times compared to nvidia’s CUDA platform [3]. The CloudMatrix 384 is priced around 60 million yuan (approximately $8.2 million) per set, while nvidia’s NVL72 is estimated to cost around $3 million [3].

Market dynamics and nvidia’s response

Despite huawei’s advancements, analysts suggest that huawei’s chips may not be very appealing outside of china [5]. Huang has urged the trump administration to revise ai technology export regulations to allow american companies to better seize future opportunities [4][6]. He believes that accelerating the global diffusion of american ai technology requires supportive government policies and incentives [6]. Huang has also expressed confidence in nvidia’s ability to manufacture chips domestically using u.s. resources [4].

Executive compensation reflects ai leadership

Huang’s compensation reflects nvidia’s leadership in ai [7]. Huang received his first base salary adjustment since 2013, increasing his annual salary from $1 million to $1.5 million, a 50 percent increase [7]. His performance bonus also increased to $3 million, up $1 million from the previous year [7]. Furthermore, his stock awards were valued at $38.8 million, bringing his total compensation for the 2024 fiscal year to approximately $49.9 million [7].

Bronnen

AI Chips Export Restrictions