NVIDIA's High-Stakes Game in China's Shifting AI Landscape
High-performance AI chip landscape is a turbulent one, and for NVIDIA, the gales of transformation are blowing particularly strongly in China. Recent comments by CEO Jensen Huang at Computex stand in stark contrast: US export regulations, and particularly those introduced under the Biden administration, have dealt a hard hit to NVIDIA's dominance of the Chinese AI market, a market Huang himself has described as a staggering $50 billion opportunity.
A Losing Grip on a Strategic Market
Huang writes that NVIDIA's market share in China has plummeted from a leading 95% to around 50%. This is not a short-term drop; it is a classic shift. The regulations have choked NVIDIA's ability to sell its GPUs, like the H100 and A100 series, which it has been selling in gigantic quantities to Chinese computer giants. The implications are harsh: a significant source of income is in peril, and many decades of market leadership are being jeopardized.
Why is this occurring. The core issue is US concern that sophisticated AI technology could end up in the possession of states considered hostile and thereby present a threat to national security. This has led to a sequence of increasingly stringent regulations designed to limit China's access to these enormously powerful chips.
The Rise of a Fierce Challenger: Huawei
Nature abhors a vacuum, and technology is no different. With NVIDIA's ability to deliver its top chips curtailed, a fierce homegrown contender has stepped into the breach: Huawei. China's technology giant has been making tremendous strides in developing its own AI chip technology, with offerings like the Ascend 910C and 910B creating waves. Tencent, Baidu, and ByteDance have been among the leading Chinese companies reportedly adopting these chips, particularly for inference workloads. Even Huawei is launching rack-scale products like the CloudMatrix 384 to directly compete with NVIDIA's Blackwell GB200 NVL72 offerings.
Jensen Huang has spoken about concerns that if China were to be compelled to develop its own complete AI ecosystem, then the US would risk losing its traditional leadership in the field. This suggests a feeling that Chinese technology is rapidly closing the gap with Western technology.
Navigating the Regulatory Maze and Market Sentiment
The landscape is dynamic and multidimensional. While initial controls were stringent, there have been suggestions of replacing sweeping AI diffusion regulations with a less extensive, chip-based international regime of licensing hammered out between states. This has given analysts some hope.
For instance, Cantor Fitzgerald analyst C.J. Muse reiterated an 'Overweight' rating on NVIDIA, forecasting a $200 price target. Muse sees near-term headwinds in the form of expected revenue hit from China-only licensing requirements for H20 GPUs. But he looks for a robust recovery in the second half of 2025 driven by Blackwell architecture ramp and robust rack-scale system shipment expansion. Muse is looking for NVIDIA's data center revenue to increase to $200 billion in 2025, well above current consensus estimates, and gross margins in the mid-70s.
Muse believes any near-term deficit in guidance from NVIDIA would be a welcome "clearing event," clearing the way for future growth as bearish catalysts fade.
NVIDIA's Plan: New Chips and Unwavering Partnerships
In spite of the issues, NVIDIA isn't ready to abandon the Chinese market. The company is reportedly developing new AI chips that will be compliant with US export controls. The new chips, said to be based on Blackwell architecture, will use GDDR7 memory instead of HBM to fit the limitations. Shipments are estimated to be as much as one million units by 2025 end, indicating NVIDIA's will to remain a significant player.
Even with these altered chips, NVIDIA is relying on its superior architecture and robust software ecosystem to offer a competitive edge over Huawei's current products. But the uncertainty of employing US-designed chips is irresistibly pushing at least some Chinese entities towards local solutions.
Meanwhile, NVIDIA's reliance on strategic manufacturing partners goes on unchecked. CEO Jensen Huang has publicly reaffirmed the company's very strong commitment to TSMC, particularly for its advanced CoWoS (Chip-on-Wafer-on-Substrate) packaging technology. Huang explained that, currently, NVIDIA has "no other choice" for this critical technology that has enabled the break of Moore's Law by enabling the stacking of multiple chips on top of one another to achieve higher performance. Despite speculations of collaboration with Samsung Foundry and Intel for future-generation packaging, to date, no concrete agreements have reportedly been signed, and TSMC is still the only top partner for NVIDIA's sophisticated chip manufacturing.
NVIDIA's history in China is an intriguing story of innovation, geopolitical tensions, and intense rivalry. How the company will develop its strategies and navigate the complex regulatory environment will be key to determining its future success within this strategic AI market.
