Here are the latest developments in the AI chip sector as of August 2025, covering technological breakthroughs, policy support, international competition, and market trends, combined with authoritative industry information and in-depth analysis of data:
1. International Giants Accelerate Their Layout in the Chinese Market
NVIDIA is developing two new AI chips for the Chinese market: the B30A (training chip) based on the Blackwell architecture and the RTX6000D (inference chip). The B30A features a single-chip design, with performance between the H20 and H100, equipped with HBM memory and NVLink technology, and samples are expected to be provided to Chinese customers in September. The RTX6000D, on the other hand, uses GDDR memory with a memory bandwidth of 1398GB/s (slightly below the U.S. limit of 1.4TB/s) and is expected to be priced lower than the H20, with plans for small batch testing at the same time. Despite the Trump administration’s previous indications of potential restrictions on Blackwell chip exports, NVIDIA is still attempting to maintain its market share in China through a technology downgrading scheme. However, Chinese regulatory authorities have tightened security reviews of NVIDIA chips, leading some state-owned enterprises to suspend purchases of the H20 and evaluate domestic alternatives.
2. Breakthroughs in Domestic Technology and Ecosystem Development
Milestone in EDA Tool Localization: Huada Jiutian has launched the first domestic full-process EDA solution supporting large-scale Flash/DRAM mass production, covering the entire chain from design to verification to mass production, breaking the monopoly of international giants like Synopsys and Cadence, and directly supporting advanced process R&D for companies like Yangtze Memory Technologies and Changxin Memory Technologies. This solution has already improved yield by 12% in Yangtze Memory’s 232-layer 3D NAND production line, and it is expected to increase the domestic EDA market share from 15% to 25% by 2025.
3. Policy-Driven Market Explosion
National-level Strategy Deepening: The Ministry of Industry and Information Technology has released the “2025 Work Points for the Integration of Information Technology and Industrialization,” clearly requiring the promotion of intelligent transformation for 3000 manufacturing enterprises, with a focus on supporting edge computing chip applications for AI quality inspection and predictive maintenance scenarios. Shanghai has simultaneously introduced a three-year action plan for “AI + Manufacturing,” offering up to 20% equipment procurement subsidies to enterprises using domestic AI chips, directly benefiting edge chips like Rockchip RK3588 and Tailin Micro TL7000.
Investment Strategy Recommendations:
◦ Short-term: Focus on the approval progress of NVIDIA’s B30A and targets accelerating domestic chip replacements, such as Digital China in the Ascend ecosystem and Zhongke Shuguang for Cambricon.
◦ Long-term: Invest in sectors with deep technological barriers like storage-computing integration (Juxin Technology), Chiplet (Lianqi Technology), and neuromorphic computing (Huawei Ascend), while being cautious of risks from changes in technological routes. Investors are advised to closely monitor events such as the National AI Industry Development Conference in September and Apple’s fall launch event, as these milestones may release new technical standards or application scenarios, further catalyzing the AI chip market.
Below is a list of AI chips expected to see mid-year performance increases in 2025, sorted by net profit growth rate based on technical strength, growth logic, and market performance (data as of August 22, 2025):
1. High-Growth Core Targets in AI Chips
1. Tianyu Shuke
◦ Expected Increase: Net profit year-on-year +369.52%~489.3% (mid-year forecast)
◦ Core Driver: The affiliated company Chipeng Intelligent focuses on spatial computing and AI chip design, with products applied in humanoid robots, AI glasses, etc. The metaverse and AI hardware boom have driven a surge in demand for spatial computing chips, with the company’s products entering the supply chains of leading manufacturers, coupled with cost reduction and efficiency improvement measures significantly enhancing profitability.
◦ Technical Highlights: Chipeng Intelligent’s AI chip supports high-precision indoor positioning (0.5 meters), outperforming Apple’s U1 chip (0.8 meters), and has been mass-produced for smart home and industrial sensor applications.
2. Tailin Micro
◦ Expected Increase: Net profit year-on-year +267% (mid-year forecast)
◦ Core Driver: Edge AI chips (such as the TL7000 series) support mainstream training models and are globally leading in the Bluetooth low-power SoC field. The explosive demand for low-power AI chips in IoT devices (smart homes, industrial sensors) has led to a surge in the company’s product shipments, coupled with accelerated domestic replacements.
◦ Technical Highlights: The world’s first chip manufacturer supporting BLE6.0 high-precision indoor positioning, with positioning accuracy reaching 0.5 meters, and the Matter protocol chip has a 15% market share in the North American smart home market.
3. Huazheng New Materials
◦ Expected Increase: Net profit year-on-year +271.02%~371.3% (mid-year forecast)
◦ Core Driver: BT packaging materials and CBF laminated insulating films are used in the packaging of GPU and other computing power chips, with bulk supply to Huawei Ascend’s supply chain. High-end GPUs like NVIDIA’s H100 and AMD’s MI300 adopt advanced packaging technologies, and the company’s materials directly benefit from the expansion of AI chip production capacity.
◦ Technical Highlights: Breaking the Japanese manufacturers’ monopoly on ABF substrate materials, the company has established an industrialization company in collaboration with the Shenzhen Advanced Electronic Materials International Innovation Research Institute, and its products have been validated by downstream manufacturers.
4. Guanghua Technology
◦ Expected Increase: Net profit year-on-year +375.05%~440.26% (mid-year forecast)
◦ Core Driver: Wafer-level cyanide-free gold plating technology provides material support for ASIC chip wafer manufacturing, adapting to the advanced packaging needs of high-end chips. Chiplet technology drives the upgrade of packaging materials, and the company’s technology breaks overseas monopolies, accelerating domestic replacements.
◦ Technical Highlights: Product yield reaches 99.5%, with costs reduced by 20% compared to traditional gold plating processes, and has entered the supply chains of TSMC and SMIC.
5. Rockchip
◦ Expected Increase: Net profit year-on-year +190.61% (mid-year)
◦ Core Driver: Edge AI chips (such as RK3588) have shipped over 12 million units in industrial robots and smart cockpit scenarios. The AIoT market continues to grow rapidly, with the company’s products expanding in automotive electronics, machine vision, and various robotics fields.
◦ Technical Highlights: The RK3588 chip supports 8 TOPS INT8 computing power and has been applied in the in-vehicle vision systems of BYD and Xiaopeng, with the proportion of automotive-grade chip revenue increasing from 12% to 25% in the first half of 2025.
6. Haiguang Information
◦ Expected Increase: Net profit year-on-year +40.78% (mid-year)
◦ Core Driver: The DCU product “Deep Computing No. 1” supports full-precision AI computing, with performance reaching 60% of NVIDIA’s A100, capturing 18% market share in AI training. The demand for domestic high-end chips is rising, and after merging with Zhongke Shuguang, the company has formed a full-chain capability of “chip-machine-computing power service,” winning over 60% of orders for national computing power hub nodes.
◦ Technical Highlights: The Deep Computing No. 1 DCU improves performance by over 2.5 times in training large models with hundreds of billions of parameters compared to traditional clusters, surpassing international competitors by 1.2 times.
7. Lianqi Technology
◦ Expected Increase: Net profit year-on-year +85.5%~102.36% (mid-year forecast)
◦ Core Driver: Significant growth in shipments of DDR5 memory interface and module supporting chips, with a substantial increase in sales revenue from high-performance computing chips. The demand for AI servers has exploded, with the penetration rate of DDR5 memory interface chips in AI servers increasing from 35% in 2024 to 60% in 2025.
◦ Technical Highlights: Only three companies globally master the technology for the first generation of DDR5 memory interface chips, with the company holding a 45% market share, higher than Renesas Electronics (32%) and IDT (23%).
8. Juxin Technology
◦ Expected Increase: Net profit year-on-year +122.28% (mid-year forecast)
◦ Core Driver: The promotion of edge AI audio chips based on in-memory computing technology has been significant, with multiple leading brands initiating projects, and customer terminal products in the low-latency private wireless audio field achieving mass production first. The optimization of product and customer sales structure has steadily improved gross profit levels.
◦ Technical Highlights: The ATS2851 chip achieves an energy efficiency ratio of 100GOPS/W in wearable devices, three times better than traditional architectures, and has been applied in smart headphones from Huawei and Xiaomi.
9. Allwinner Technology
◦ Expected Increase: Net profit year-on-year +31.02%~43.46% (mid-year forecast)
◦ Core Driver: The T527 series high-performance octa-core dual-display AI chip has been mass-produced in multiple industries, covering smart vehicles, industrial control, and other scenarios. The intelligent upgrade of AIoT devices drives chip demand growth, with the company’s products showing significant cost-performance advantages.
◦ Technical Highlights: The T527 chip supports 4K video encoding and decoding and multi-sensor fusion, and has been applied in the smart cockpit systems of Xiaopeng and Weilai.
10. Cambricon
◦ Expected Increase: Q1 2025 net profit year-on-year +256.82% (complete mid-year data pending disclosure)
◦ Core Driver: Cloud AI chips (such as the Siyuan 590) have seen a surge in orders from government clouds and major internet companies, with performance reaching 80% of NVIDIA’s A100 and power consumption reduced by 35%. The company has secured 2.3 billion yuan in orders, with cooperation framework agreements with companies like ByteDance involving tens of billions in deliveries.
◦ Risk Warning: Customer concentration is as high as 94.6%, with the top five customers contributing over 79% of revenue, necessitating caution regarding reliance on a single customer.
2. Supporting Industry Chain Targets
1. Guangku Technology
◦ Expected Increase: Net profit year-on-year +60%~80% (mid-year forecast)
◦ Core Driver: Modulator chips adapt to GPU computing power node connection needs, developing high-speed optical engines compatible with CPO. The construction of AI computing power networks drives demand for optical modules, with the company supplying batches to internationally renowned AI enterprises.
◦ Technical Highlights: The high-speed optical engine achieves a transmission rate of 1.6Tbps, a 50% improvement over traditional solutions, and has been applied in NVIDIA’s DGX SuperPod clusters.
2. Wentai Technology
◦ Expected Increase: Net profit year-on-year +178%~317% (mid-year forecast)
◦ Core Driver: Automotive-grade power semiconductors and AI server foundry services cover computing power chip application scenarios. The explosive demand for new energy vehicles and AI servers has led to full order books for the company’s foundry business.
◦ Technical Highlights: AI server foundry share accounts for 15% of the domestic market, with clients including leading enterprises like Inspur and Shuguang.
3. Industry Trends and Investment Recommendations
1. Technological Route Differentiation:
◦ Cloud Training: Huawei Ascend 910B (320 TFLOPS FP16 computing power) and Cambricon Siyuan 590 (256 TFLOPS FP16 computing power) dominate the market, with a focus on the progress of computing power cluster construction.
◦ Edge Inference: Rockchip RK3588 (8 TOPS INT8 computing power) and Tailin Micro TL7000 (6 TOPS INT8 computing power) are more cost-effective in edge computing scenarios.
◦ In-Memory Computing: Juxin Technology ATS2851 (100 GOPS/W energy efficiency ratio) and Horizon Journey 6 (80 GOPS/W energy efficiency ratio) represent future directions.
2. Policy and Market Resonance:
◦ The State Council’s “New Generation Artificial Intelligence Development Plan” requires that by 2025, the market share of domestic AI chips exceeds 50%, significantly benefiting leading companies like Haiguang Information and Lianqi Technology.
◦ The resupply of NVIDIA’s H20 chip may bring short-term shocks, but domestic chips have advantages in energy efficiency (Ascend 910B 310W vs. H20 350W) and supply chain security.
3. Investment Strategy:
◦ Short-term: Focus on Rockchip and Tailin Micro, which have high performance certainty, as well as Haiguang Information, which is catalyzed by policies.
◦ Long-term: Invest in Juxin Technology (in-memory computing) and Cambricon (large model adaptation) with deep technological barriers, while being cautious of risks from technological iterations. Investors are advised to dynamically adjust their holdings based on the company’s product matrix, customer structure, and R&D investment, with a focus on Q3 order delivery and Q4 new product release schedules.
Currently, only a few companies have disclosed financial forecast data, and some have released mid-year reports for 2025. The above information is just a selection of the top-ranking companies in the AI chip concept sector. Please continue to pay attention to the latest financial disclosures from companies in this field.
*Reminder: Some companies have a relatively small proportion of revenue from related businesses, so please be cautious and beware of the risk of unsustainable performance growth.
Creating content daily is not easy, and we strive to bring you the latest concepts, hottest topics, and research reports on good companies across various industries. If this article is useful to you, please like or share it. Finally, I wish everyone successful investments and a prosperous account!
Special Statement: The content published by Research Report Point does not constitute investment advice and is for reference only!