China’s $10B+ bet to build AI chips faces 4 major hurdles

U.S. export controls have effectively cut China off from the world’s most advanced artificial intelligence chips, forcing Beijing to build an entire domestic semiconductor ecosystem from scratch. This isn’t simply about replacing a few high-end processors—it requires mastering four interconnected technological domains that took decades for the West to develop.

The challenge extends far beyond chip design. U.S. restrictions target every link in the semiconductor supply chain, from the specialized manufacturing equipment needed to produce chips to the memory components that make them functional. China has mobilized tens of billions of dollars to bridge these gaps, achieving some breakthroughs through sheer financial force while still facing fundamental technological hurdles.

“U.S. export controls on advanced Nvidia AI chips have incentivized China’s industry to develop alternatives, while also making it more difficult for domestic firms to do so,” explains Paul Triolo, partner and senior vice president for China at advisory firm DGA-Albright Stonebridge Group.

Understanding China’s progress requires examining how the country stacks up in four critical areas that form the backbone of any AI chip ecosystem.

1. AI chip design

Graphics processing units (GPUs) serve as the computational engines that power artificial intelligence systems, handling the massive parallel calculations required for training AI models and running inference tasks. Unlike traditional computer processors that excel at sequential tasks, GPUs can perform thousands of calculations simultaneously—making them indispensable for AI workloads.

Nvidia dominates this market with GPU designs that have become the industry standard. However, the company doesn’t manufacture these chips itself. Instead, Nvidia creates the architectural blueprints and sends orders to specialized foundries—contract manufacturers that produce semiconductors for other companies. This design-focused approach has allowed Nvidia to concentrate on innovation while leveraging the manufacturing expertise of partners like Taiwan Semiconductor Manufacturing Company (TSMC).

U.S. export controls have significantly constrained Nvidia’s ability to serve Chinese customers. The company revealed in April that additional restrictions prevented it from selling its H20 processor to Chinese clients. The H20 represented a deliberately downgraded version of Nvidia’s flagship H100 processor, specifically engineered to comply with previous export controls while still providing meaningful AI capabilities. Despite these limitations, experts note that even Nvidia’s restricted chips remained more advanced than any domestic Chinese alternatives.

China’s response has involved mobilizing multiple domestic semiconductor companies to fill the GPU void left by Nvidia’s absence. Startups like Enflame Technology and Biren Technology have emerged, seeking to capture billions of dollars in unmet demand. However, no Chinese firm appears closer to providing a genuine Nvidia alternative than Huawei’s chip design division, HiSilicon.

Huawei’s most advanced GPU currently in mass production is the Ascend 910B, with the next-generation Ascend 910C reportedly expected to begin large-scale shipments. According to Dylan Patel, founder and chief analyst at SemiAnalysis, a semiconductor research firm, Huawei has made significant progress in closing the performance gap with Nvidia.

“Compared to Nvidia’s export-restricted chips, the performance gap between Huawei and the H20 is less than a full generation,” Patel explains. “Huawei is not far behind the products Nvidia is permitted to sell into China.” He notes that while the 910B was approximately two years behind Nvidia’s capabilities as of last year, the Ascend 910C represents only a one-year lag.

This progress in GPU design capabilities represents genuine advancement, but chip design constitutes just one piece of a much larger technological puzzle.

2. AI chip fabrication

Designing sophisticated chips means nothing without the ability to manufacture them at scale. This reality has created China’s most significant bottleneck in building an independent AI chip ecosystem.

TSMC, the world’s largest contract chip foundry, produces most of the globe’s advanced semiconductors, including Nvidia’s GPUs. The Taiwanese company operates at the technological frontier, currently manufacturing chips using 3-nanometer processes. In semiconductor terminology, smaller nanometer measurements indicate more advanced technology—3-nanometer chips pack transistors more densely than 7-nanometer chips, delivering superior performance and energy efficiency.

However, TSMC complies with U.S. chip controls and cannot accept orders from companies on America’s trade blacklist. Since Huawei was added to this list in 2019, Chinese chip designers have been forced to rely on domestic foundries, primarily Semiconductor Manufacturing International Corporation (SMIC).

SMIC represents China’s most capable chip foundry, but it operates several generations behind TSMC’s cutting-edge capabilities. The company can officially produce 7-nanometer chips, though this requires less advanced technology than TSMC’s 3-nanometer production. Signs suggest SMIC has made unexpected progress—the company reportedly produced a 5-nanometer chip for Huawei’s Mate 60 Pro smartphone in 2023, surprising industry observers and raising questions about the effectiveness of U.S. export controls.

Despite this breakthrough, SMIC faces substantial challenges in mass-producing advanced GPUs cost-effectively. According to independent chip analyst Ray Wang, SMIC’s operational capacity remains dwarfed by TSMC’s scale and efficiency. “Huawei is a very good chip design company, but they are still without good domestic chipmakers,” Wang observes, noting that Huawei is reportedly developing its own fabrication capabilities to address this gap.

The fundamental constraint limiting both SMIC and Huawei’s manufacturing ambitions lies in accessing critical production equipment.

3. Advanced chip equipment

The Netherlands may lack prominent semiconductor designers or manufacturers, but it houses ASML, the world’s monopoly supplier of the most advanced chipmaking equipment. ASML’s machines use precisely controlled light beams to transfer intricate circuit patterns onto silicon wafers, forming the foundation of modern microchips.

The most sophisticated of these machines employ extreme ultraviolet (EUV) lithography—a technology so complex that ASML remains the sole company capable of producing it at commercial scale. EUV systems cost over $200 million each and represent the only viable method for manufacturing the most advanced chips cost-effectively.

In accordance with U.S. export controls, the Netherlands has agreed to block sales of ASML’s EUV machines to Chinese companies. This restriction creates a fundamental barrier to China’s advanced chip production capabilities.

“EUV is the most significant barrier for Chinese advanced chip production,” explains Jeff Koch, an analyst at SemiAnalysis. “They have most of the other tooling available, but lithography is limiting their ability to scale towards 3-nanometer and below process nodes.”

SMIC has developed workarounds using ASML’s less advanced deep ultraviolet lithography systems, which face fewer export restrictions. This “brute force” approach enables 7-nanometer chip production, but with poor yields and high costs. Koch notes that “at current yields it appears SMIC cannot produce enough domestic accelerators to meet demand.”

Chinese companies are attempting to develop domestic lithography alternatives. SiCarrier Technologies, reportedly linked to Huawei, is working on lithography technology. However, replicating existing EUV capabilities could require years or decades to achieve. Instead, China may pursue alternative lithography techniques or entirely different technological approaches to circumvent current limitations.

4. AI memory components

Even the most sophisticated GPUs cannot function in isolation—they require specialized memory chips to store and rapidly access the vast amounts of data involved in AI computations. In AI applications, high bandwidth memory (HBM) has emerged as the industry standard due to its ability to feed data to processors at the speeds modern AI workloads demand.

South Korea’s SK Hynix leads the HBM market, with Samsung and U.S.-based Micron Technology also competing in this space. These memory chips stack multiple layers of memory cells vertically, connected by thousands of microscopic wires that enable extremely fast data transfer rates.

“High bandwidth memory at this stage of AI progression has become essential for training and running AI models,” explains analyst Wang. The memory chips must be precisely integrated with GPUs to create functional AI accelerators, requiring sophisticated packaging and testing capabilities.

South Korea began complying with fresh U.S. restrictions on HBM sales to China in December, creating another supply chain vulnerability for Chinese AI development. In response, Chinese memory chip manufacturer ChangXin Memory Technologies (CXMT), partnering with chip-packaging company Tongfu Microelectronics, has begun early-stage HBM production.

According to Wang, CXMT remains three to four years behind global leaders in HBM development while facing significant obstacles including export controls on essential chipmaking equipment. SemiAnalysis estimated in April that CXMT remained approximately one year away from achieving reasonable production volumes.

Chinese foundry Wuhan Xinxin Semiconductor Manufacturing is reportedly constructing a factory dedicated to HBM wafer production. Reports suggest Huawei has partnered with the firm for HBM chip production, though neither company has confirmed this relationship.

Meanwhile, Huawei has relied on HBM stockpiles from suppliers like Samsung for use in their Ascend 910C AI processor. As SemiAnalysis noted in an April report, while the chip was designed domestically, it still depends on foreign components obtained before or despite current restrictions.

“Whether it be HBM from Samsung, wafers from TSMC, or equipment from America, Netherlands, and Japan, there is a big reliance on foreign industry,” SemiAnalysis concluded.

The interconnected challenge

China’s quest to build a complete AI chip ecosystem reveals the intricate interdependencies within the global semiconductor industry. Progress in one area—such as Huawei’s advancing GPU designs—can be undermined by limitations in another, like SMIC’s manufacturing constraints or restricted access to EUV lithography equipment.

The country’s substantial financial investments have yielded tangible results, particularly in chip design capabilities where companies like Huawei have demonstrated they can compete with international standards. However, the manufacturing and equipment challenges represent more fundamental technological barriers that cannot be easily overcome through increased spending alone.

For China to achieve true independence in AI chip production, it must simultaneously master advanced chip design, develop cutting-edge manufacturing processes, create sophisticated production equipment, and build reliable memory chip capabilities. This represents one of the most complex technological undertakings in modern industrial history, with implications extending far beyond China’s borders as the global AI revolution continues to unfold.