What's Really Inside Huawei's 'Domestic' AI Chips?
News Summary
A TechInsights teardown reveals that Huawei's latest AI chips, despite being promoted as a milestone in China's technological self-reliance, continue to rely on critical components from Taiwan Semiconductor Manufacturing Co. (TSMC), Samsung, and SK Hynix. The report indicates that TSMC produced the dies powering Huawei's Ascend 910C processors, while older-generation high-bandwidth memory (HBM2E) used in separate chip samples came from Samsung and SK Hynix. Both Korean companies emphasized they halted sales to Huawei following U.S. export controls in 2020 under President Donald Trump and remain compliant with American regulations. U.S. authorities have tightened controls on AI chips, HBM memory, and their manufacturing equipment, aiming to limit Beijing's ability to challenge Nvidia in advanced computing. Despite these restrictions, Huawei has leveraged stockpiled inventory; analysts estimate the company acquired nearly 3 million TSMC dies through intermediary Sophgo before sanctions severed the channel, which continue to power the 910C chips that entered mass shipment earlier this year. However, supply risks are mounting, with SemiAnalysis projecting Huawei could face high-bandwidth memory shortages by the end of 2025. Pressure on Huawei's supply chain intensified after reports in December 2024 suggested the U.S. administration was considering blacklisting Sophgo. Huawei is pressing ahead with efforts to rival Nvidia's AI performance; its rotating Chairman Eric Xu unveiled a three-year plan at Huawei Connect to cluster Ascend processors via a new UnifiedBus interconnect system, claiming it can transfer data up to 62 times faster than Nvidia's upcoming NVLink144. This move underscores Beijing's strategy of supporting Huawei as a national champion amid U.S. curbs, even though Huawei's chips still trail Nvidia's in raw processing power and remain at 7-nanometer designs.
Background
Since 2019, the U.S. government has placed Huawei on its Entity List, citing national security concerns and restricting its access to American technology. Under the Trump administration, these restrictions included prohibitions on U.S. companies selling chips to Huawei and using semiconductor manufacturing equipment with U.S. technology. This policy aimed to curb China's advancements in critical technological areas, particularly 5G and advanced semiconductors. In response, China initiated a "domestic substitution" strategy, heavily investing in its indigenous semiconductor industry with the goal of achieving technological self-reliance. Huawei, as a leading Chinese tech giant, is seen as a central player in this national strategy. The U.S. has since further tightened export controls on AI chips and high-bandwidth memory (HBM) to limit China's progress in artificial intelligence and high-performance computing, directly impacting the ability of companies like Huawei to acquire high-end chips.
In-Depth AI Insights
What are the deeper implications of Huawei's continued reliance on foreign components regarding the long-term effectiveness of US sanctions? While US sanctions have undoubtedly pressured Huawei by cutting off critical supply chains, forcing it to deplete inventory and seek alternatives, the company's continued survival and promotion of "domestic" chips reveal inherent contradictions in the sanctions' efficacy: - Persistent "Cat and Mouse" Game: The complexity of global supply chains means loopholes and intermediaries, as seen with Sophgo, will persist. Sanctions require continuous monitoring and escalation, increasing enforcement costs and difficulty. - Accelerated, but Incomplete, Indigenous Innovation: Sanctions have undeniably spurred China's domestic semiconductor industry. However, the complexity of advanced chip manufacturing (7nm and below) and bottlenecks in critical components like HBM demonstrate that full self-sufficiency is distant and costly. - Long-Term Geopolitical Struggle: This is not a short-term tech battle but a manifestation of a protracted geopolitical competition. The US aims to curtail China's ability to challenge its technological hegemony, while China views it as an existential challenge, pursuing breakthroughs at all costs. How might the sustained pressure from the Trump administration, particularly on intermediaries like Sophgo, reshape the global semiconductor supply chain and foster new avenues of circumvention or genuine domestic breakthroughs? The Trump administration's continued targeting of intermediaries signals an evolving, deepening US sanctions strategy with multiple impacts on the global semiconductor supply chain: - Accelerated Supply Chain "De-risking": Non-Chinese firms will increasingly seek to decouple production and sales networks from China to avoid entanglement in geopolitical conflicts, leading to further fragmentation of supply chains. - Increased Cost and Complexity of Circumvention: As the US escalates its crackdown on circumvention, the cost and risk of acquiring restricted components via intermediaries will surge, forcing companies to pursue more clandestine, complex, and expensive routes. - China's Domestic Alternative as a "Double-Edged Sword": On one hand, external pressure will catalyze greater investment and policy support for China's indigenous semiconductor industry, accelerating technological breakthroughs by firms like CXMT. On the other hand, technological barriers and timeframes mean these breakthroughs may remain concentrated in lower-to-mid-tier segments or fail to fully meet high-end demand in the short term, creating a "progress but not enough" scenario. Given Huawei's aggressive push with its UnifiedBus interconnect despite lagging process nodes, what is the strategic rationale and potential investment implication for Nvidia and other AI chip leaders? Huawei's move highlights its strategic adaptation under manufacturing constraints, shifting from solely pursuing advanced process nodes to focusing on system integration and ecosystem building. This has profound implications for Nvidia and other AI chip leaders: - "System-Level" vs. "Process-Level" Competition: Huawei recognizes that where process leadership is unattainable, it can enhance overall AI computing capabilities by optimizing interconnects, software ecosystems, and customized solutions to meet specific application needs. This could create a viable "non-Nvidia" tech path within the Chinese market. - Risk of Chinese Market Fragmentation: Even if Huawei chips lag in general performance, their deep integration with local hardware and software, coupled with national support, could allow them to dominate specific sectors in China (e.g., government, military, state-owned enterprises), thereby eroding Nvidia's market share in China. - Long-Term Strategic Challenge for Nvidia: Nvidia must be wary of this "differentiated" competition. While its technological lead is unassailable in the short term, if Huawei can establish a mature Ascend ecosystem within China, attracting developers and applications, it could pose a substantial long-term threat to Nvidia's near-monopoly in one of the world's largest markets.