admin
The landscape of the global semiconductor industry is undergoing a tectonic shift as China accelerates its drive toward technological self-reliance, with domestic artificial intelligence (AI) chip shipments now projected to reach 5 million units by 2026. According to recent data emerging from an expert consultation hosted by Deutsche Bank, the Chinese domestic chip sector is poised for a period of rapid expansion, fueled by a combination of stringent U.S. export controls and a concerted state-led effort to prioritize local hardware. This projection aligns with a broader trend of "de-Americanization" within the Chinese technology ecosystem, as firms pivot away from industry leaders like NVIDIA in favor of homegrown alternatives.
The details of the Deutsche Bank expert call suggest that the trajectory for domestic shipments is already well underway. Estimates indicate that local firms shipped approximately 2.5 million units in 2024, a figure that is expected to climb to 4 million units in 2025 before hitting the 5 million mark in 2026. This growth represents a compounded annual growth rate (CAGR) of roughly 30% over the next three years. Perhaps more significant is the shift in market composition; the share of the domestic market held by local manufacturers, led by the Semiconductor Manufacturing International Corporation (SMIC) and the Shanghai Huahong Grace Semiconductor Manufacturing Corporation, is expected to rise from 40% to over 50% in the same period.
The Catalyst of Export Restrictions and Regulatory Pressures
The primary driver behind this surge in domestic production is the escalating series of trade restrictions imposed by the United States. Since 2022, the U.S. Department of Commerce has implemented a series of export controls designed to limit China’s access to advanced logic chips and the equipment necessary to manufacture them. These sanctions have specifically targeted high-end AI GPUs, such as NVIDIA’s A100 and H100 architectures, which are essential for training large language models (LLMs).
While NVIDIA initially responded by designing "China-specific" chips—such as the H20, L20, and L2—that comply with U.S. performance thresholds, the Chinese government has reportedly begun discouraging local enterprises from purchasing these downgraded variants. Instead, Beijing is leveraging a "buy local" strategy to stimulate the domestic semiconductor industry. This policy serves a dual purpose: it mitigates the risk of future supply chain disruptions caused by foreign sanctions and provides the necessary capital and volume for domestic chipmakers to refine their manufacturing processes.
Divergent Projections from Global Financial Institutions
The optimistic outlook provided by Deutsche Bank is largely mirrored by other major investment banks, though timelines vary slightly. JPMorgan, in a recent analytical report, forecasted that domestic AI chip shipments could jump from 1 million units in 2025 to 5 million units by 2028. While JPMorgan’s timeline is more conservative than the Deutsche Bank expert’s 2026 target, both institutions agree on the massive scale of the impending volume.
JPMorgan’s analysis highlights Huawei and the government-backed Cambricon Technologies as the primary beneficiaries of this growth. Huawei’s Ascend series, particularly the 910B, has emerged as the most viable domestic alternative to NVIDIA’s high-end offerings. Meanwhile, Cambricon Technologies remains a significant player in the specialized AI processor market. Goldman Sachs previously estimated that Cambricon could ship as many as 2.3 million units of its AI chips by 2030. Interestingly, reports indicate that Cambricon currently derives approximately 80% of its revenue from ByteDance, the parent company of TikTok, illustrating the deep integration between domestic hardware providers and China’s largest internet conglomerates.
Case Study: Moonshot AI and the Kimi K3 Milestone
The feasibility of an all-domestic AI ecosystem was recently bolstered by an announcement from Moonshot AI, one of China’s leading artificial intelligence startups. The firm claimed that its latest model, the Kimi K3, was developed and trained exclusively using domestic silicon. This claim served as a significant "proof of concept" for the industry, suggesting that despite the lack of access to the most advanced Western lithography, Chinese software developers can still achieve high-performance results using local hardware.
The development of Kimi K3 is seen as a milestone for the "sovereign AI" movement in China. By proving that a competitive LLM can be trained without relying on NVIDIA’s ecosystem, Moonshot AI has provided a blueprint for other domestic firms facing similar procurement hurdles. This achievement is particularly noteworthy given that domestic chips often face challenges related to interconnect speeds and software optimization compared to the mature CUDA (Compute Unified Device Architecture) platform offered by NVIDIA.
Manufacturing Constraints and the Role of SMIC and Hua Hong
Despite the bullish shipment projections, the Chinese semiconductor industry continues to grapple with significant manufacturing headwinds. The U.S. and its allies have restricted the export of Extreme Ultraviolet (EUV) lithography machines, which are critical for producing chips at the 5nm node and below. Consequently, China’s leading foundry, SMIC, has had to rely on Deep Ultraviolet (DUV) immersion lithography to push the boundaries of its production capabilities.
Industry experts note that while SMIC has successfully produced 7nm-class chips using multi-patterning techniques, this process is often characterized by lower yields and higher production costs compared to EUV-based manufacturing. Nevertheless, the strategic importance of domestic production appears to outweigh these economic inefficiencies. The Shanghai Huahong Grace Semiconductor Manufacturing Corporation (Hua Hong) complements SMIC’s efforts by focusing on mature nodes and specialty processes, which are essential for the broader AI ecosystem, including power management and sensor interfaces.
The expansion of these foundries is supported by the "Big Fund" (China Integrated Circuit Industry Investment Fund), which recently launched its third phase with a massive capital injection of approximately $47.5 billion. This funding is intended to bridge the gap in high-end manufacturing and support the development of domestic supply chains for materials and chemicals used in chip fabrication.
Chronology of the US-China Chip Conflict
The current state of the market is the result of a multi-year escalation in technological competition:
- October 2022: The U.S. Bureau of Industry and Security (BIS) introduces comprehensive export controls on advanced computing and semiconductor manufacturing items to China.
- August 2023: Huawei releases the Mate 60 Pro, featuring a 7nm Kirin 9000s chip manufactured by SMIC, signaling a breakthrough in domestic production capabilities.
- October 2023: The U.S. updates its rules to close loopholes, further restricting the export of mid-tier AI chips like the NVIDIA A800 and H800.
- Early 2024: Reports emerge that the Chinese government is instructing local EV makers and AI firms to prioritize domestic chips over foreign alternatives.
- July 2024: Financial institutions like Deutsche Bank and JPMorgan revise their shipment forecasts upward, reflecting the accelerated adoption of local hardware.
Strategic Implications for the Global Supply Chain
The shift toward a 5-million-unit domestic shipment volume has profound implications for the global technology market. For NVIDIA, China has historically accounted for a significant portion of its data center revenue. The loss of market share in this region—or the forced transition to lower-margin, China-specific products—presents a long-term challenge to its revenue growth.
Conversely, for the global supply chain, China’s focus on self-sufficiency is leading to a bifurcated ecosystem. We are seeing the emergence of two distinct AI stacks: one built on Western standards (NVIDIA, AMD, Intel) and another built on Chinese standards (Huawei, Cambricon, Biren Technology). This bifurcation affects everything from software compatibility to the global talent pool, as engineers must increasingly specialize in one of the two ecosystems.
Furthermore, the 30% CAGR in domestic shipments suggests that China is not merely aiming for survival but for dominance in the next generation of AI applications. While the "leading edge" of 2nm and 3nm production may remain out of reach for the immediate future, the ability to produce millions of 7nm and 14nm AI chips provides a sufficient foundation for a wide array of industrial, military, and consumer AI applications.
Conclusion and Future Outlook
The projection of 5 million domestic chip shipments by 2026 marks a critical inflection point in the global chip war. While technological hurdles remain—specifically in the realm of advanced lithography and software ecosystems—the momentum generated by state support and localized demand is undeniable. The success of models like Moonshot’s Kimi K3 suggests that the software-hardware gap is narrowing, and the massive capital infusions from the "Big Fund" ensure that the infrastructure for this growth remains well-funded.
As the industry moves toward 2026, the focus will likely shift from sheer shipment volume to the qualitative performance of these chips. The ability of SMIC and Hua Hong to improve yields and the capacity of Huawei and Cambricon to challenge NVIDIA’s performance benchmarks will ultimately determine if China can sustain this growth or if the industry will hit a technological ceiling. For now, the data indicates a sector that is rapidly scaling to meet the challenges of a restricted global trade environment.
