Why is the success of Huawei Ascend 950 important to the GX China semiconductor ETF (3191)?

With the daily influx of news related to Chinese semiconductors, it’s difficult to discern what is key issue. In that case, it’s worth paying attention to Huawei’s “Ascend 950” project, scheduled for release in 2026.

The success of Huawei’s new AI chip, the Ascend 950, is crucial. If successful, it will open up a huge market for Chinese semiconductor companies.

The Significance of the Ascend 950

The reason China semiconductor industry is betting on the Ascend 950 is clear.

Unlike “AI training,” which requires extreme computational performance, “AI inference,” the actual service implementation stage, can achieve outstanding performance even with a 7nm process, not a 3nm process.

AI training and AI inference are fundamentally different. AI training requires high-precision computational power, as it involves constantly modifying hundreds of billions of parameters ​​until the right answer is found. AI Inference, on the other hand, involves inputting data and deriving results from an existing AI model, and does not require extreme computational power. This means that Huawei’s (or Cambricon’s) 7nm chip is fully capable of entering the AI ​​inference chip market.

From a commercial perspective, AI inference chips are even more important, as the AI ​​chip market leadership is likely to shift from AI training chips to AI inference chips. While AI training chips accounted for the majority of the AI ​​chip market until 2025, this is expected to completely shift by 2030, with inference chips expected to account for nearly 80% of global AI chip sales.

Furthermore, the Ascend 950’s economic advantages are clear. Let’s look at its price-to-performance ratio. The NVIDIA H100 is approximately 1.3 times more powerful than the Ascend 950, but its price is estimated to be 3-5 times higher than the Huawei chip. For enterprise customers, the Ascend 950 is attractive when considering its “data throughput” relative to its budget.

If Ascend 950 is successful, the market size available to Chinese semiconductor companies will expand significantly. China’s domestic demand for AI semiconductors is expected to expand from approximately $40 billion in 2025 to approximately $100 billion in 2030. In 2025, Chinese companies held a 20% market share in local market and generated $8 billion in revenue. If Chinese companies secure a 50% domestic market share by 2030, their revenue is projected to increase 6 times from $8 billion in 2025 to $50 billion in 2030.

Therefore, the success of the Ascend 950 will provide a solid revenue base for the entire Chinese semiconductor supply chain and will also benefit the constituents of Global X China Semiconductor ETF.

Four Challenges for the Huawei Ascend 950

1) Foundry Segment

Unable to import EUV lithography equipment due to US sanctions, China is taking on the extreme challenge of utilizing its predecessor, deep ultraviolet (DUV) lithography. Currently, SMIC is the only company in the world attempting to achieve 5nm-level processing using only DUV equipment. The core of this strategy is multi-patterning, a process that uses DUV lithography equipment to overlap circuits four or more times. This increases the number of process steps by more than three times compared to using EUV equipment, inevitably leading to lower yields.

SMIC is circumventing this issue with a design-process optimization strategy. Huawei’s design team intentionally adjusts chip designs to account for the foundry’s lower process precision. This increases the physical size of the chip somewhat, but ensures a meaningful yield.

2) HBM segment

This aims to achieve self-sufficiency in High Bandwidth Memory (HBM), the import of which is banned due to US sanctions. Huawei is pursuing the establishment of a system to procure HBM3 from domestic companies such as Changxin (CXMT) and Fujian Jinhua (JHICC) for mass production of the Ascend 950.

China’s HBM technology is expected to enter the fourth-generation (HBM3) mass production stage by 2026. A technological gap exists compared to Korean companies that have already entered sixth-generation (HBM4) mass production.

However, HBM3 alone will be sufficient for the development of AI inference chips. CXMT, a leader in China’s DRAM industry, recently announced that it has increased the yield of HBM3 to approximately 50%.

3) Packaging segment

The final hurdle for Ascend 950 production is securing advanced packaging technology, namely CoWoS (Chip on Wafer on Substrate), which combines the GPU and HBM. SiCarrier and Tungfu Micro (TFME) are working together to address this challenge. SiCarrier, a subsidiary of Huawei, developed the interposer exposure and etching solution, a key component of packaging. Tungfu, China’s largest OSAT company, is responsible for final manufacturing based on this solution.

4) Equipment segment

The growth of equipment suppliers such as NAURA, AMEC, and Hwatsing is also a key driving force. NAURA’s etching and deposition equipment has become an essential component of China’s semiconductor production lines.

Building a Self-Reliant Ecosystem by 2026

The successful launch of the Huawei Ascend 950 in 2026 will be more than just a new product announcement. It symbolizes the establishment of China’s high-end semiconductor supply chain, spanning SMIC (manufacturing), CXMT (memory), Tongfu (packaging), and NAURA, AMEC (equipment). This ecosystem will not only facilitate the shared growth of companies within the Huawei alliance but also serve as a catalyst for all chip design companies like Cambricon.