Nvidia’s China market collapses as local AI chips surge

The gist
US export controls have torpedoed Nvidia’s China dominance, unleashing a homegrown AI chip blitz that’s rewriting the global semiconductor playbook.
What to know
- Nvidia’s China market share is forecast to nosedive from 40% in 2025 to just 8% in 2026 as domestic players like Huawei capture nearly half the market.
- Chinese chipmakers such as Moore Threads and Biren are reporting explosive growth—147% and 20x jumps—thanks to government rules demanding 80% local chips in state-backed data centers.
- China is side-stepping US bans by renting Nvidia GPUs via Southeast Asian cloud providers, exposing major loopholes in export controls and keeping the Nvidia pipeline alive despite restrictions.
Cloud Loopholes Undermine Bans
Chinese AI firms are sidestepping US chip bans by remotely renting Nvidia GPUs from Southeast Asian clouds, exposing regulatory blind spots and sustaining demand despite plummeting official market share.
US export controls on Nvidia's advanced GPUs have created a complex regulatory landscape that Chinese firms have adeptly navigated by leasing compute capacity in overseas data centers, particularly in Southeast Asia. Companies like Moonshot AI reportedly accessed around 20,000 Nvidia H200 processors through Alibaba and leveraged cloud providers in Thailand, Malaysia, and Japan to circumvent direct hardware sales bans, exposing significant loopholes in the export regime. This remote renting strategy, highlighted by Michael Kratsios as a seam in the system, complicates Nvidia's compliance efforts and enforcement, as the controls govern physical chip shipments but not cloud-based access.
Despite these export restrictions, Nvidia's market presence in China has weakened dramatically, with its AI chip market share forecasted to plummet from 40% in 2025 to just 8% by 2026, while domestic giants like Huawei surge to nearly 50%. This shift is driven by tightened US controls blocking sales of Nvidia's most advanced Blackwell generation chips and critical manufacturing equipment, prompting Beijing to double down on indigenous chip development and procurement from local suppliers. Nvidia CEO Jensen Huang acknowledged this dynamic tension, noting the company once held about 95% market share in China before export controls and criticizing the policies for accelerating China's chip self-sufficiency and pushing buyers away from the US ecosystem.
The export control regime’s focus on physical chip ownership has inadvertently created a regulatory blind spot, as Chinese AI firms exploit cloud-based loopholes and smuggling to maintain access to Nvidia technology, sustaining demand despite official restrictions. This has led to documented cases of chip smuggling and indirect usage through intermediaries and foreign shell companies, complicating enforcement and signaling broader geopolitical risks for enterprises reliant on Nvidia hardware. Meanwhile, Nvidia continues to generate substantial revenue by selling GPUs to hyperscalers outside China and renting GPU capacity, underscoring the paradox of compliance amid eroding market share.
In response to these challenges, Nvidia is pivoting strategically by developing China-specific AI chips that comply with export restrictions, such as limiting high-bandwidth memory and advanced packaging technologies, to regain a foothold in the market. Concurrently, Chinese firms like SMIC and Alibaba are capitalizing on the export controls to expand domestic manufacturing and innovate alternative AI architectures, with SMIC’s revenue surpassing $3 billion and Alibaba advancing RISC-V based processors that reduce GPU dependence. This evolving landscape reflects a contested battleground where Nvidia’s operational strategies and US policy efforts to tighten controls, including pending legislation like the Remote Access Security Act, are continuously tested by rapid market and technological shifts.
Domestic Chipmakers Surge Ahead
Government mandates and rapid technological advances have fueled explosive growth for Chinese AI chipmakers, transforming export restrictions into a catalyst for a robust and self-reliant domestic ecosystem.
China’s AI chip industry has rapidly accelerated its push toward self-sufficiency, spurred by US export controls that restrict access to high-end foreign GPUs. This has catalyzed a maturing domestic ecosystem with advances in critical technologies such as advanced foundry nodes, packaging, and thermal management, enabling a strategic shift toward a dual-track AI hardware model that combines indigenous GPUs with proprietary ASICs. Government policies actively prioritize domestic champions, fostering a supply chain less reliant on foreign technology and reinforcing Beijing’s broader tech sovereignty ambitions.
Chinese AI chipmakers have posted staggering revenue growth fueled by export restrictions that effectively created captive domestic markets. For instance, Moore Threads reported a 147.42% revenue surge in H1 2026, while Biren’s revenue skyrocketed over 20-fold, buoyed by government mandates requiring 80% of chips in state-funded data centers to be domestically sourced. This policy-driven market protection, combined with strategic foundry shifts from TSMC to SMIC, underscores a rapid realignment of China’s AI semiconductor landscape despite ongoing technological constraints.
Huawei has emerged as a dominant force in China’s AI chip market, with its Ascend 950 series achieving performance parity with Nvidia’s H200, enabling it to capture nearly 50% market share as Nvidia’s presence dwindles from 40% to 8%. This shift is reinforced by extensive deployment of Huawei’s chips in large AI computing clusters and a procurement culture increasingly confident in domestic supply self-sufficiency, a trend noted by analysts like He Hui from Omdia. Despite Nvidia’s declining official market share, demand for its advanced chips persists, sometimes circumventing export controls through smuggling, highlighting the complex interplay of competition and regulation.
Beyond GPUs, China’s semiconductor ambitions extend to memory and logic chips, with companies like CXMT achieving a market capitalization surpassing Tencent at $613 billion, and SMIC leading logic chip manufacturing despite limitations imposed by restricted access to advanced lithography equipment. This multifaceted industrial strategy reflects a comprehensive government-backed effort to cultivate national champions across the semiconductor value chain, illustrating how export controls have inadvertently accelerated China’s drive toward technological autonomy and global competitiveness.
Beijing's Dual-Track Chip Play
China is reshaping global AI supply chains by combining strict procurement rules with a national security push for homegrown chips, guaranteeing local champions a protected market and attracting surging investment.
China’s government has aggressively advanced a dual-track hardware strategy that combines domestic GPUs with proprietary ASICs to reduce reliance on foreign technology and accelerate AI hardware self-sufficiency. This approach is supported by a maturing domestic ecosystem excelling in advanced foundry nodes, packaging, and thermal management, which underpins Beijing’s push for hardware sovereignty and reshapes the AI infrastructure landscape away from foreign GPUs.
To solidify this strategy, Beijing has implemented stringent procurement mandates requiring that 80% of chip purchases in new state-funded data centers come from domestic suppliers, while also restricting acquisitions of foreign chips like Nvidia’s H20—even those designed to comply with US export controls. The inclusion of Chinese AI chips on official government procurement lists since December 2025 not only signals political endorsement but also provides private firms with the commercial incentive and political cover to prioritize homegrown hardware.
This government-led framework treats semiconductor self-reliance as a national security imperative, guaranteeing a captive domestic market for targeted companies such as Biren, which saw a 22-fold revenue surge after being cut off from foreign foundries and pivoting to SMIC. Such policies create a structural growth story in China’s semiconductor industry, attracting investment exemplified by the launch of VanEck’s China Semiconductor ETF (SMHC) in mid-2026, and fundamentally reshape global supply chains by fostering a protected yet rapidly evolving domestic AI chip sector.
Complementing hardware strategies, China is heavily investing in renewable energy infrastructure and strategically locating data centers near power generation sites to deliver AI compute at a fraction of US costs—around 2-3 cents per kilowatt hour. However, despite these infrastructure advances, domestic AI compute capacity remains constrained by limited access to advanced chips and EUV lithography machines, prompting intensified government support through funding and resource allocation to accelerate indigenous GPU development and reduce foreign dependency.
Nvidia’s Strategic Crossroads
Facing shrinking access to China, Nvidia is pivoting its business model and developing export-compliant chips while openly warning that US controls are accelerating China’s chip independence and intensifying global competition.
Nvidia CEO Jensen Huang has openly criticized US export controls on advanced AI chips to China, labeling them a strategic failure that inadvertently accelerates China’s indigenous chip development and drives Chinese buyers away from the US hardware and software ecosystem. This critique underscores Nvidia’s recognition that these restrictions not only hamper its market access but also intensify competition by empowering China’s domestic chipmakers, a dynamic Huang links directly to the broader geopolitical contest over AI compute infrastructure and economic power.
In response to evolving market and regulatory challenges, Nvidia is redefining its business model by positioning AI factories as the new engines of revenue, emphasizing that compute capacity is no longer a mere cost but a direct monetization lever in the AI economy. Huang highlights Nvidia’s dominant hold on 80-90% of the AI accelerator market and stresses the critical need for a comprehensive AI infrastructure stack—including advanced chips, packaging, memory, and networking—while grappling with supply chain constraints like TSMC’s sold-out packaging capacity and persistent high-bandwidth memory shortages through 2027.
Amid these headwinds, Nvidia is strategically developing a new inference-focused chip based on Groq’s Language Processing Unit (LPU) technology, designed specifically to comply with US export controls by limiting features such as high-bandwidth memory capacity and advanced packaging techniques like CoWoS. This move aims to sustain Nvidia’s foothold in the Chinese AI market despite stringent export restrictions, although Nvidia officially denies current LPU sales in China, signaling ongoing uncertainty and the complex balancing act required to navigate regulatory constraints while maintaining competitive relevance.




