This information is sourced from company announcements, related news, publicly available research reports, and social media networks. It does not constitute investment advice regarding the industries and stocks mentioned in the text. If there are any copyright infringements or violations of information disclosure, please contact us for removal.
Summary of the main content of the document
1. Industry Background and Trends in Self-Developed ASICs by Major Companies
1. Core Driving Forces
Surge in computing power demand: Google discovered early on that if 100 million Android users used voice-to-text services for 3 minutes daily, the required computing power would reach twice that of the total computing power of data centers. Traditional CPU/GPU efficiency is insufficient,
driving the demand for self-developed ASIC chips.
Cost optimization: Self-developed ASICs can reduce the procurement cost of A1 chips (e.g., Google’s TPU contributes $4,400 per unit, and the leasing model has greater profit potential).
2. Domestic Demand Factors
Self-sufficiency: Geopolitical factors are prompting domestic major companies to reduce dependence on overseas chips.
Price wars force cost reductions: The high costs of training large models make self-developed ASICs a key path for cost reduction.
2. Development History of International Major Companies’ Self-Developed ASICs
1. First Stage: Google TPU Pioneers (2013-Present)
Technological iteration: TPU V1 was launched in 2015, and it has now evolved to V6, specifically optimized for TensorFlow/PyTorch, reducing high-precision units (FP32) and focusing on FP16/INT8.
Business model: Primarily for internal use (70%-80% for internal search, advertising, and other scenarios), with 20%-30% leased externally; over 60% of AI startups and 90% of generative AI unicorns use Google Cloud TPU.
2. Second Stage: Multiple Vendors Follow (2020-2024)
Amazon: Initiated self-developed chips in 2020, with Meta and Tesla accelerating their layouts. Apple: Disclosed in July 2024 the use of TPU V5 (204 chips) to train a 2.73 billion parameter language model V4 (8192 chips) for private cloud MLU.
3. Third Stage: Multi-Core + Stack Era (Outlook for 2025)
Domestic manufacturers lead: ByteDance (CAPEX of 90 billion for A1 cards in 2025), Alibaba (Pingtouge).
Baidu (Kunlun Chip) leads, benchmarking against overseas technology paths.
3. Core Demands of Domestic Major Companies for Self-Developed ASICs
1. Geopolitics and Supply Chain Security: Reducing dependence on overseas manufacturers like NVIDIA and building a self-sufficient ecosystem.
2. Balance of Cost and Performance: Enhancing energy efficiency through ASIC customization to cope with the high costs of training large models (e.g., parameter scales reaching hundreds of billions require higher computational density).
4. Analysis of Key A-Share Targets
1. Chipone Technology
Business Highlights: One-stop chip customization and IP licensing, covering GPU/NPU/VPU/DSP/ISP across all fields.
Technical Advantages:
VPU: Six products in mass production globally, supporting multiple mainstream video formats and codecs, applied in data centers, security, and other fields.
NPU: Over 82 customers, with shipments exceeding 100 million, optimized for large model architectures.
Market Opportunities: Entering the supply chain of leading internet and automotive companies, expected to benefit from the explosion in ASIC demand by 2025.
2. Aojie Technology
Business Layout: Expanding from IoT SoCs to mobile phone SoCs, with initial performance of 4G chips emerging.
Progress: Engaging with leading internet companies in 2024, with stock price adjustments providing attention value.
5. Investment Recommendations
Chipone Technology: Strong technical accumulation, accelerated customer expansion, significant elasticity, recommended for close attention.
Aojie Technology: Dual-line layout of IoT and mobile chips, potential order releases are expected.
Risk Warning: Customer order landing rhythm, technology iteration not meeting expectations, etc.
6. Conference Conclusion
DW analysts emphasize that 2025 is a critical node for the ASIC industry and recommend investors seize the opportunity for domestic substitution.
——————
[Zhongtai Electronics] The trend of the ASIC industry is confirmed, pay attention to the emerging powerhouses in computing!
#Broadcom ASIC guidance exceeds expectations
Broadcom’s ASIC business is strong, and the company estimates that three major customers will bring a potential market scale of $60 billion to $90 billion by FY27. In terms of industry trends: due to new frontier models and technologies putting unexpected pressure on AI systems, single system designs are difficult to meet the needs of all model clusters, confirming the necessity of ASICs.
#Domestic capital begins to accelerate iteration
The trend of domestic internet ASIC three-party cooperation is confirmed, following Alibaba’s significant increase in capital expenditure and Tencent’s acceleration of capital expenditure in 2025. With the overseas situation and potential pressure from H20 being banned, the demand for self-developed chips by major companies is high. Companies with rich IP licensing capabilities, advanced process tape-out capabilities, and stable technical teams have forward-looking positioning capabilities, warranting close attention!
2024 is the zero-to-one year for domestic computing power, and 2025 will enter the one-to-ten volume stage. The trend of self-developed ASICs is confirmed, and companies with forward-looking positioning are scarce, recommended for close attention: [Aojie Technology], [Chipone Technology]!
———————-
[Minsheng Electronics] Chipone Technology: Continue to strongly recommend the leading domestic ASIC company
🌹Hello leaders, continue to emphasize Chipone Technology, the leading company in the ASIC sector, with deep cooperation with global wafer fabs, and the scarcity of industry positioning. Here is an update on the company’s recent situation:
#Leading ASIC R&D capabilities
1⃣ Leading IP + ASIC capabilities: The second largest number of IP types globally, seventh in IP licensing revenue globally, with a full layout of VPU + GPU + NPU IP.
In 2024, the company’s ASIC business revenue is expected to reach 1.6 billion yuan, mainly blooming in the second half of the year. 2H24 revenue is 969 million yuan, with a year-on-year increase of 36.7%, and the company’s Q4 ASIC revenue has increased by 81% year-on-year!
2⃣ Strong executive background: Chairman Dr. Dai has entrepreneurial experience in multiple semiconductor companies, and the company’s executives have backgrounds in leading global ASIC companies like Broadcom and Marvell;
3⃣ Sufficient R&D personnel: ASIC design typically requires teams of hundreds or even thousands of people. Chipone has continued to expand recruitment during the industry downturn, reaching 1,640 R&D personnel by 1H24, with ample personnel reserves;
#First-tier layout of domestic AI chips
1⃣ Chipone has years of custom design capabilities for GPGPU and NPU, with over 100 million AI chips shipped globally;
2⃣ Continuously developing GPU-AI computing chips for data centers, supporting 16~32 TFLOPs FP32 computing power, with 128~1536 Texel/cycle texture processing capability and 32~384 Pixel/cycle pixel fill capability;
3⃣ Previously mass-produced AI chips based on Chiplet architecture for clients like Blue Ocean Intelligence, with ample experience, providing a competitive edge in future cloud vendor client competition.
#Blooming in multiple fields such as smart driving and AI terminals
1⃣ In smart driving, the company provides 5nm automotive-grade process autonomous driving chips for new energy manufacturers;
2⃣ In AI edge devices, the company’s NPU IP can already support mobile phones, AI PCs, and other edge devices;
3⃣ Based on previous one-stop customization services for global leading manufacturers’ AR glasses chips, the company is laying out smart glasses chips.
Performance inflection point is evident: The industry’s downturn in 2023-2024 has led to poor performance, but the company is expected to reach an inflection point across various businesses. As of the end of 2024, the order backlog is 2.4 billion, with projected revenues of 3.17 billion and 4.01 billion yuan for 2025-2026, maintaining a positive outlook.
———————–
[Chipone Technology] Value is severely underestimated, progress exceeds expectations (Cinda Electronics)
#Self-developed chips by major companies are a trend, combining adaptability and economy, with cost savings of 67%, making it imperative.
# The volume of self-developed chips is substantial, with this year’s total order volume for computing power reaching 800,000 to 900,000 units, and inference demand growing rapidly, with annual self-developed demand of no less than 500,000 units. Assuming strict control, there is a greater demand increase.
# Tape-out is crucial, and there must be at least one channel; currently, two channels have been confirmed.
# The value is significant; apart from core IP, everything else is outsourced, estimating that each unit will bring suppliers a gross profit of 4,000 yuan.
Currently, xygf440 billion, with 2 billion in revenue, assuming a 50% share after cooperation is confirmed, an annual revenue growth of 5 billion and a gross profit of 1 billion, considering other industry and customer demands, it is assumed to bring in 1 billion in revenue annually. By 2026, revenue may reach fourfold growth, with 8 billion in revenue * 20 times PS = 160 billion market value. The recently widely circulated $600 is severely underestimated (how could a large chip be $600? A single die is around $600, and a chip has four dies combined), an excellent entry point.
———————-
The public account has been revamped, and everyone may not receive article notifications in time. Please be sure to click “Like” and “See First” at the bottom of the article, or click the three dots in the upper right corner of the public account homepage to set the public account as “Starred” for easy access (Your likes and views are my motivation for updates).