Alibaba’s Chip Breakthrough: The Trend of Domestic Substitution!

Alibaba’s subsidiary, Tsinghua Unigroup, has made significant progress in chip development, achieving a major breakthrough in 2025 with its new generation AI inference chip, PPU, which matches the performance of NVIDIA’s H20. The actual measured performance of ResNet50 reaches 125 TOPS with a power consumption of 350W, marking the official entry of domestic AI chips into the international first-class level. This advancement not only reshapes the domestic semiconductor industry chain but also forms a complete ecological closed loop from chip design, manufacturing, packaging, and testing to computing power services through deep binding with A-share listed companies.

1. Development Process: Dual-Drive of RISC-V Ecosystem and Self-Developed AI Chips

1. Complete Chip Product Matrix Tsinghua Unigroup has built an “end-to-cloud” chip product line: in the cloud inference field, it includes the Light 800 and the new generation Yitian 700+ chips; in the terminal and edge computing field, it has laid out the Xuantie series RISC-V processor IP and the Yuzhen series SoC chips. The Xuantie C930 processor, released in 2025, adopts a 15-stage out-of-order superscalar pipeline, supports PCIe 5.0 and DDR5 memory controllers, and its performance matches mainstream processors, breaking the limitation of “RISC-V = low power consumption” and beginning to penetrate data center scenarios.

2. Technical Breakthrough and Mass Production Capability The Tsinghua Unigroup PPU chip is compatible with NVIDIA’s CUDA ecosystem, supports multimodal inference, and is manufactured using SMIC’s 14nm process, with a monthly production capacity of 80,000 units (planned to expand to 150,000 units by the end of 2025) and a yield rate exceeding 95%. Its RISC-V ecosystem has covered over 150 customers, with more than 500 licenses granted, and the total revenue of ecosystem partners is expected to exceed 80 billion yuan in 2024.

2. Impact on the Industry Chain: Domestic Substitution and Ecological Reconstruction

1. Acceleration of Domestic Substitution in Core Links

• Manufacturing Link: SMIC undertakes the foundry of core chips such as PPU, with a 14nm process yield rate of 95%, completely eliminating dependence on TSMC.
• Packaging and Testing Link: Changdian Technology, as the exclusive packaging and testing service provider, uses SiP/3D packaging technology, achieving a 99.999% yield rate for 3D packaging, and undertakes over 70% of RISC-V chip packaging and testing orders, with a monthly production capacity exceeding 5 million units.
• Toolchain Link: Huada Empyrean’s analog circuit design tools have completed full-process adaptation with Xuantie processors, providing key support for chip design.

Expansion of RISC-V Ecosystem and Technological Upgrades The shipment of Xuantie RISC-V processors has exceeded 3 billion units, driving rapid growth in IP licensing and chip mass production businesses for partners such as Chipone and Allwinner Technology. For example, Allwinner Technology’s revenue from the RISC-V product line is expected to increase by 218% year-on-year in 2024, with its D1-H chip based on Xuantie C906 already in mass production. Tsinghua Unigroup plans to release the Xuantie C930 core in 2025, with a 50% performance improvement, further promoting the application of RISC-V in data centers and AI inference scenarios.

Upgrade of Computing Power Infrastructure Alibaba plans to invest 380 billion yuan in cloud and AI hardware infrastructure over the next three years, driving demand for liquid-cooled servers and IDC construction from companies like Inspur Information and Data Harbor. Inspur Information’s AI servers equipped with Light 800 won a 1 billion yuan procurement project from China Mobile, capturing 70% of the share; Data Harbor has built over 20 customized data centers for Alibaba Cloud, with revenue accounting for over 80%.

3. Benefiting Industry Chain and Core Listed Companies

(1) Chip Design and IP Licensing

1. Chipone (688521) The chairman unit of the China RISC-V Industry Alliance, providing RISC-V GPU IP licensing to Tsinghua Unigroup and jointly developing AIoT chip DSA expansion interfaces. The IP licensing revenue is expected to reach 736 million yuan in 2024, with significant contributions from Tsinghua Unigroup orders.

2. Allwinner Technology (300458) One of the first “Xuantie Preferred Partners”, jointly developing Xuantie C910 core chips. The D1-H chip based on Xuantie C906 has already been mass-produced, with revenue from the RISC-V product line expected to increase by 218% year-on-year in 2024.

3. Cambricon (688256) The core supplier of Alibaba Cloud AI inference chips, the Siyuan 590 chip has been adapted to Alibaba Cloud servers, with orders of 1 billion yuan expected in Q2 2025, increasing to 150,000 units in Q3, covering 60% of Alibaba Cloud’s inference demand.

(2) Manufacturing and Packaging

1. SMIC (688981) The leading domestic wafer foundry, undertaking the foundry of Tsinghua Unigroup’s PPU, Yitian 710, and other chips, with a 14nm process yield rate exceeding 95%. The monthly production capacity of PPU is expected to reach 80,000 units in 2025 (planned to expand to 150,000 units by the end of the year).

2. Changdian Technology (600584) The exclusive packaging and testing service provider, undertaking over 70% of RISC-V chip packaging and testing orders, using SiP/3D packaging technology, with over 1 million units of Light 800 packaged, and a single packaging price of about 200 yuan.

3. Liangyang Chip (688135) The core testing service provider for Xuantie chips, providing full-process testing solutions for PPU, covering high-complexity scenarios such as autonomous driving domain controller chips.

(3) Computing Power Infrastructure and Liquid Cooling Technology

1. Inspur Information (000977) The main supplier of Alibaba AI servers, with liquid cooling technology reducing PUE to below 1.1, expected to capture over 50% of Alibaba’s new cabinet orders. Servers equipped with Light 800 are expected to ship 5,000 units in 2025.

2. Data Harbor (603881) The core supplier of Alibaba Cloud customized IDC, having built over 20 data centers, with revenue accounting for over 80%, applying liquid cooling technology to reduce PUE to 1.15.

3. Invec (002837) The main supplier of high-power liquid cooling solutions for Alibaba data centers, with a market share exceeding 40%, and Alibaba’s procurement accounting for over 60%, supporting high-density computing clusters.

(4) Software Ecosystem and Applications

1. Runhe Software (300339) Launched the “Yiying 1520” development board based on Xuantie C910, with AI computing power reaching 4 TOPs, applied in intelligent driving. Its subsidiary, Runkai Hong, has over 30% market share in the edge computing module of Cainiao Logistics, with related revenue expected to grow by 80% in 2024.

2. Xuanji Information (300324) The exclusive agent for Xuantie C930 chips, planning to ship 50 million units in three years. Jointly developing a low-power IoT SoC platform, applied in fields such as NB-IoT.

3. Hang Seng Electronics (600570) Co-building financial AI solutions with Alibaba Cloud, with quantitative trading solutions based on Alibaba chips already implemented in multiple brokerages, with Ant Group holding a 20.79% stake.

4. Risks and Challenges

1. Technology Iteration Risk: The rapid technological updates in AI chips and the RISC-V field require continuous investment in R&D to maintain competitiveness.
2. Intensifying Market Competition: Companies like Huawei Ascend and Cambricon are also actively laying out AI chips, which may squeeze market share.
3. Uncertainty in Performance Realization: Some collaborations are still in the early stages, with limited short-term contributions to related companies’ performance, requiring attention to order fulfillment.

Conclusion: Alibaba Tsinghua Unigroup’s chip strategy has formed a strong ecological network effect through “chip + toolchain + community” three-dimensional empowerment. With the mass production of PPU chips and the expansion of the RISC-V ecosystem, the process of domestic substitution is accelerating, and significant opportunities are emerging in the design, manufacturing, packaging, and computing infrastructure segments of the industry chain. Investors should focus on companies with deep ecological ties and technological barriers, such as Chipone, SMIC, Changdian Technology, and Inspur Information, while also being wary of risks posed by technology iteration and market competition.