Domestic GPU 'Four Dragons' In-depth Investment Value Analysis: Comparison of Technical Routes and Business Strategies

According to the latest market data, the domestic GPU industry witnessed an IPO boom in 2025, reflecting strong confidence from the capital market in the AI computing power sector:

• Moore Threads
  : Listed on the Sci-Tech Innovation Board (STAR Market), with a first-day gain of
  425%
  , demonstrating high market recognition of the full-function GPU route [1]
• MetaX Integration (MetaX)
  : Its first-day gain on the STAR Market reached an astonishing
  693%
  , setting a record [1]
• Biren Technology
  : Has submitted an IPO application in Hong Kong, planning to issue 247.7 million shares to raise up to
  USD 623 million
  , with an offer price range of HKD 17-19 [2][3]
• Suiyuan Technology
  : Plans to IPO on the STAR Market, targeting to raise approximately
  RMB 2 billion
  (about USD 280 million) [1]
• Cambricon
  : As a listed AI chip company, its stock price soared sharply in 2025, becoming a market focus [6]

The rapid development of China’s AI semiconductor industry is driven by multiple factors:

• National Strategic Support
  : Under the national goal of technological self-reliance and self-improvement, policy support is strong [1]
• Explosive Market Demand
  : Surge in demand for AI large model training and inference
• Huge Substitution Potential
  : Under the background of U.S. sanctions, the demand for domestic substitution is urgent
• Capital Favor
  : The market expects China to achieve chip technology breakthroughs in 2026-2027 [1]

• Strategic Positioning
  : Following NVIDIA’s full-function GPU route, pursuing integration of graphics rendering and AI computing
• Latest Progress
  : Launched a new generation GPU architecture
  “Huagang”
  in December 2025, with 50% higher computing density and 10x higher energy efficiency [4]
• Simultaneously released graphics chip “Lushan”
  , with 15x higher performance in AAA games [4]
• Technical Advantages
  : Unified architecture can cover multiple scenarios such as games, professional graphics, AI training and inference

• Market Positioning
  : Full coverage of consumer and enterprise levels
• Ecosystem Construction
  : Investing heavily in developer ecosystem to promote MUSA architecture adaptation
• Risk Factors
  : Huge R&D investment for full-function route; direct competition with NVIDIA with obvious technical gap; game compatibility issues still exist (DX11 games have rendering errors, texture loss, etc.) [4]

:

• High capital market recognition (425% first-day gain)
• Clear technical route, high ceiling by benchmarking NVIDIA
• Widest application scenarios

:

• Huge R&D investment, long profit cycle
• Facing strong competition pressure from NVIDIA
• Ecosystem construction requires long-term investment

• Strategic Positioning
  : Focus on dedicated AI chips for vertical fields such as government, enterprises, and finance
• Technical Route
  : Optimized for specific scenarios instead of pursuing full functionality
• Market Segmentation
  : Avoid direct competition with NVIDIA, focus on domestic substitution刚需 market

• Target Customers
  : Government agencies, financial enterprises, large state-owned enterprises
• Value Proposition
  :
  • Meet data security and independent controllability requirements
  • Optimized for specific workloads
  • Provide customized solutions

:

• Best listing performance (693% first-day gain), high market expectations
• Clear commercialization path in vertical fields
• Relatively controllable R&D investment, profitable in the near future
• Benefit from domestic substitution policies

:

• Relatively limited market space
• Customization leads to scaling difficulties
• Lower technical ceiling than full-function route

• Strategic Positioning
  : Focus on high-end AI training chips, benchmarking NVIDIA A100/H100
• Technical Route
  : Large computing power, high-performance training dedicated
• Market Positioning
  : Regarded as “one of China’s most likely competitors to NVIDIA” [1]

• Financing Capacity
  : IPO raises USD 623 million, showing strong capital appeal
• Customer Group
  : Large internet companies, AI laboratories, scientific research institutions
• Competition Strategy
  : Catch up with NVIDIA in performance indicators,争取 customers through cost advantages

:

• Positioned in high-end training market with highest single-chip value
• Attracted international media attention, high brand awareness
• Large financing provides sufficient funds for subsequent R&D

:

• Compete with NVIDIA in the highest-end market with obvious technical gap
• U.S. sanctions may affect access to advanced manufacturing processes
• High-end market customers have extremely high performance requirements and low fault tolerance

• Strategic Positioning
  : Bind with Tencent, deep plowing in cloud service inference scenarios
• Technical Route
  : Inference optimization, focusing on cost-effectiveness and energy efficiency ratio
• Application Scenarios
  : AI inference, cloud services, edge computing

• Core Advantage
  : Tencent-backed investment and business binding, ensuring initial orders
• Business Model
  : B2B2C, indirectly serving end customers through cloud service providers
• Partners
  : Tencent, National Semiconductor Fund, etc. [1]

:

• Has Tencent as a strategic customer
• Rapid growth in inference market demand (AI large model application landing)
• Avoid fierce competition in high-end training market

:

• High dependence on a few key customers
• Lower unit price of inference chips than training chips
• Market space limited by cloud service providers’ adoption意愿

: Moore Threads

1. Largest Market Space
   : Cover all scenarios of games, professional graphics, AI training and inference
2. High Technical Ceiling
   : Benchmarking NVIDIA, huge value if successful
3. Strong Ecosystem Effect
   : Unified platform can form network effect
4. Capital Market Recognition
   : Easier to get high valuation

1. Huge R&D Investment
   : Need continuous large-scale investment
2. Highest Technical Difficulty
   : Direct competition with NVIDIA, obvious gap
3. Long Ecosystem Construction
   : CUDA ecosystem barriers are difficult to break in the short term
4. Long Commercialization Cycle
   : Profit requires longer time

: MetaX, Suiyuan

1. Clear Commercialization Path
   : Focus on specific scenarios, easier to land
2. Controllable R&D Investment
   : Avoid huge investment in full-stack R&D
3. Short Profit Cycle
   : Can achieve break-even faster
4. Obvious Policy Support
   : In line with domestic substitution direction
5. Low Competition Pressure
   : Avoid direct competition with NVIDIA

1. Limited Market Space
   : Focus on fields with lower ceiling
2. Poor Scalability
   : Difficult to expand horizontally to other fields
3. Limited Valuation Potential
   : Lower valuation ceiling than full-function platform
4. High Customization Cost
   : Difficult to achieve scale

According to online search results and industry analysis,

, for the following reasons:

Under the background of US-China tech competition, the demand for independent controllability in key fields such as government, enterprises, and finance is most urgent, and vertical field chips can meet this demand fastest.

Large models are moving from training to application landing, and inference demand growth exceeds training. The inference scenario that Suiyuan is bound to fits this trend exactly. Amazon and other tech giants are also launching customized AI chips to focus on reducing inference costs [7].

Vertical route is easier to implement:

• Clearer customer portrait
• Shorter sales cycle
• Faster cash flow turnover
• Lower R&D risk

Although Moore Threads and MetaX have amazing first-day gains, the market is beginning to pay attention to:

• Sustained profitability
• Commercialization progress
• Cash flow status

Vertical route has more advantages in these aspects.

, if successful:

1. Can cut into from vertical market and gradually expand to full function
2. Can gradually evolve to full-function platform with technical accumulation
3. The final form may be similar to NVIDIA, starting from professional market and gradually covering all scenarios

But the premise is:

• Need long-term capital support
• Need technical breakthrough
• Need ecosystem construction
• Need continuous national strategic investment

:

1. MetaX
   : Vertical route + best listing performance + government and enterprise刚需
2. Suiyuan Technology
   : Tencent binding + inference track + cloud service scenario
3. Biren Technology
   : High-end training + large financing + international vision
4. Moore Threads
   : Full-function route + high ceiling + high risk

• Process Restriction
  : U.S. sanctions affect access to advanced processes
• Ecosystem Barriers
  : CUDA ecosystem is difficult to break
• Performance Gap
  : Obvious technical gap with international giants like NVIDIA

• Profit Pressure
  : High R&D investment leads to continuous losses
• Customer Concentration
  : Over-reliance on a few key customers
• Market Competition
  : Many domestic and foreign competitors

• Valuation Volatility
  : Large fluctuations in STAR Market and Hong Kong stocks
• Policy Changes
  : Uncertainty in international trade policies
• Technology Iteration
  : Rapid changes in AI technology may lead to product route adjustments

:

1. 2025-2027
   : Vertical field breakthrough route is dominant, focus on commercialization ability and profitability
2. After 2028
   : Full-function platform route value emerges, but needs technical and ecosystem breakthroughs
3. Final Winner
   : May be a hybrid mode starting from vertical fields and gradually expanding to full function

: MetaX has the most investment value

• Best listing performance (693% first-day gain)
• Clear commercialization path in vertical fields
• Government and finance demand is rigid
• Profitable in the near future

: Suiyuan Technology has the greatest potential

• Rapid growth of inference market
• Tencent ecosystem binding
• Mature cloud service scenario

: Moore Threads has the highest ceiling

• Full-function platform benchmarking NVIDIA
• Widest application scenarios
• But highest risk

: At the current stage, , reasons:

1. Higher Commercial Feasibility
2. Shorter Profit Cycle
3. In Line with Domestic Substitution Demand
4. Better Risk-Reward Ratio

But in the long run,

, similar to NVIDIA starting from professional graphics and gradually covering all AI computing scenarios.

:

• Establish technical and commercial barriers in vertical fields
• Accumulate sufficient cash flow and technical strength
• Expand to adjacent fields at the right time
• Finally form full-function platform capability

---

[1] Bloomberg - “China AI Semiconductor IPO Accelerates, Large Listings Succeed Continuously - Catching Up with the US” (December 22, 2025)

[2] Wall Street Journal - “Biren Technology, One of China’s GPU ‘Four Dragons’, Plans to Raise Up to USD 623 Million via Hong Kong IPO” (December 2025)

[3] Wall Street Journal - “Biren Technology Plans to Raise Up to USD 623 Million via Hong Kong IPO” (December 2025)

[4] Zhihu - “Moore Threads Launches New GPU Architecture ‘Huagang’, Computing Density Increases by 50%, What Impact Will It Have on Domestic G…” (December 20, 2025)

[5] Zhihu - “How About Buying Moore Threads’ Graphics Card in 2025?” (2025)

[6] Yahoo Finance - “Cambricon (688256.SS) Stock Price, News, Quotes and Records” (December 2025)

[7] Wall Street Journal - “Amazon Launches Custom AI Chip, Pose New Threat to NVIDIA” (2025)

[8] Bairong Cloud - “Deliver Results with AI – Bairong Cloud Launches RaaS Strategy and ‘Result Cloud’ Platform” (December 19, 2025)