- Intraday data tracking shows that the Semiconductor Equipment ETF, Zhaoshang (561980:CH), opened significantly higher in early trading and maintained its strong momentum, with gains at one point reaching 4.46%. The underlying component stocks showed a general upward trend, with Haiguang Information (688041:CH) rising over 6%, Zhongke Feice (688361:CH) climbing 8%, and key weight stocks such as ACM Research and Advanced MicroFabrication Equipment gaining more than 5% each.
- DeepSeek released version V4 of its large language model and explicitly announced support for Huawei's Ascend 950 chip. Meanwhile, local computing power architectures such as Cambricon and Moore Threads have also completed adaptation, providing substantive verification of the fully autonomous and controllable process of domestic AI chips at the model end.
- Supply chain data indicates that the delivery cycle of server CPUs from Intel (INTC:US) and AMD (AMD:US) for Chinese customers has been extended to six months, with an accumulated price increase of over 10%. This supply-side contraction is significantly accelerating the market share penetration and value re-evaluation of domestic CPU manufacturers.
Fund Flows and Micro Pricing Structures
During the Asia-Pacific session on this trading day, there was a significant switch of funds among different segments within the Chinese A-share technology sector. As the congestion in certain popular tracks on the Growth Enterprise Market reached a high level, macro liquidity quickly moved towards upstream hard technology fields with performance certainty. The Semiconductor Equipment ETF, Zhaoshang, as a broad-based tool covering the entire industry chain, reflects a 4.46% daily increase, indicating that institutional funds are establishing long positions through indexed investments. The CSI Semi-conductor Index tracked by this ETF has risen over 284% since 2020, and during the current peak disclosure period for the first-quarter reports, the highly determinable capital expenditures and order conversion rates of its underlying assets provide robust fundamental support for market valuations.
Marginal Variables of Computing Power Infrastructure
The technological evolution of large AI models is reshaping the underlying hardware allocation logic. According to brokerage research data, the internal CPU-to-GPU structure ratio of AI data centers is converging from the traditional 1:8 to 1:4, with an industry trend towards 1:1 in the mid to long term. This marginal change implies that the widespread application of AI inference and agents not only drives demand for graphics processors but is also bringing about the value return of general-purpose processors. Against the backdrop of tight supplies and continuous price hikes from leading overseas suppliers, the construction cost curve of domestic data centers is being passively elevated, forcing downstream customers to expedite order shifts to local hardware suppliers.
Commercialization Turning Point for Domestic Alternatives
The release of DeepSeek V4 model acts as a catalyst for the local computing power ecosystem. This model, with its capability to process millions of contextual instructions, establishes an advantage in the open-source track by its native adaptation to Huawei’s Ascend 950 chip and other domestic AI architectures, breaking the past strong reliance of software ecosystems on single overseas hardware. For local enterprises like Haiguang Information, the next-generation C86 processor plan aims for a 17% IPC performance improvement, directly benchmarking mainstream overseas products. If the yields and capacity ramp of such domestic chips meet expectations, they are poised to accommodate large-scale server upgrade needs within the year, thereby driving a new round of "Davis Double Kill" for related semiconductor design and manufacturing sectors.
