🧔♂️ A friendly human may check it before it goes live. More news here
Chinese chip stocks test traders’ patience as valuations soar
Chinese chip stocks have surged 55% since the end of June, outpacing gains of 20% to 30% in the US, Japan, South Korea, and Taiwan, according to Bloomberg data.
This rally comes amid Beijing’s push for tech self-sufficiency and encouragement for firms to move away from using Nvidia processors.
Companies like Huawei and Alibaba have introduced their own chip designs, while Semiconductor Manufacturing International Corp. (SMIC) is reportedly testing China’s first domestically produced advanced chipmaking equipment.
Valuations for Chinese chip firms have soared, with SMIC trading at over 160 times estimated forward earnings, and Cambricon Technologies at 150 times, compared to Nvidia’s 33 times.
Analysts warn that such high valuations leave little room for error if growth slows.
🔗 Source: Bloomberg
🧠 Food for thought
Implications, context, and why it matters.
SMIC’s domestic tooling tests lack specifics on yields and timelines that justify current valuations
– SMIC is testing China-built tools for etch and deposition. It is also trying clean and Chemical Mechanical Planarization (CMP) gear. Vendors include AMEC, Naura, and Piotech. ACM Research’s Shanghai arm plus Hwatsing and Kingsemi also ship tools. Reports give no data on yield, uptime stability, or timelines for volume use 1.
– The hurdle is cluster runs that keep recipes stable at acceptable yields. No concrete performance numbers have surfaced 1.
– Lithography blocks progress since SMEE tools remain unproven for high-volume advanced nodes. A plan for 70 to 80 percent local tools by 2025 to 2027 may not support SMIC’s valuation premium 1.
– Multi-patterning with Deep Ultraviolet (DUV) to hit 7 nm adds cost and cycle time. It also raises defect risk versus Extreme Ultraviolet (EUV) chips, creating doubt about economics at scale 1.
Software integrators can capitalize on CUDA-to-domestic stack migration as Chinese platforms become first-class targets
– DeepSeek launched its V3.2-Exp model with tuning for Huawei Ascend and the Compute Architecture for Neural Networks (CANN). Cambricon, Hygon, and SGLang added support fast. That signals Chinese accelerators rank as first-choice targets 2.
– Software firms and systems integrators can build migration tools, framework support, and tuning services. Beijing is pushing groups to use Huawei Ascend and Cambricon over Nvidia’s Compute Unified Device Architecture (CUDA) 1.
– Fast uptake across Ascend, Cambricon MLU, and Hygon DCU marks growing maturity in non-CUDA stacks. That shift creates demand for cross-platform compatibility layers and tuning know-how 2.
– An antitrust probe in China and procurement rules that steer pilots to domestic accelerators help integrators who can move workloads off CUDA 1.
