In-depth Analysis of Investment Value of China's Top Four GPU Players

NVIDIA went from near bankruptcy in the 1990s to becoming the world’s highest-market-cap company, and its success was not achieved overnight. Key turning points include:

• Paradigm Innovation of GPGPU
  : Shifted GPU from graphics rendering to general-purpose parallel computing
• CUDA Ecosystem Barrier
  : Built a developer ecosystem since 2006; currently, there are over 5 million CUDA developers worldwide, forming a deep moat
• Openness Overcomes Closure
  : Historically, open platforms (e.g., IBM PC-compatible machines) often outperform closed systems (e.g., early Apple); CUDA’s openness is key to its success

Current historical opportunities for domestic GPUs:

• Explosion of AI Computing Power Demand
  : Large models like ChatGPT drive exponential growth in training computing power demand
• Geopolitics Force Localization
  : U.S. embargo on high-end chips to China creates a protective market space for domestic GPUs
• Strong Policy Support
  : National strategies such as “New Quality Productivity” and “Eastern Data, Western Computing” provide policy dividends [1][2]

:

• Adopts “full-function GPU” strategy, covering both graphics rendering and AI computing
• Self-developed MUSA unified computing architecture, targeting CUDA
• Launched product lines like “Chunxiao” and “Suti”, supporting FP16/BF16 precision

:

• Consumer + enterprise dual-wheel drive
• Parallel product matrix of gaming GPUs and AI inference cards
• Builds brand through gaming market, gradually penetrates AI training market

:

• Most complete product line, highest market ceiling
• Higher brand awareness, largest investment in ecosystem building

:

• Dispersed resources; may lose to focused competitors in pure computing power competition for AI training
• Consumer market requires long-term user education

: ★★★☆☆ (Medium-high risk, high return potential)

:

• Focuses on government and enterprise private cloud market, highlighting security and controllability
• Adopts x86/ARM compatible architecture, reducing migration costs
• Deeply optimized for domestic CPUs (e.g., Loongson, Phytium)

:

• Binds party, government and military customers, provides integrated “chip + software + service” solutions
• Deeply adapted to domestic operating systems (UnionTech, Kylin)
• Rapid volume growth through government centralized procurement and central enterprise procurement

:

• Strongest customer stickiness, least affected by international situation
• Stable cash flow, strong survival ability

:

• Relatively limited market ceiling
• Difficult to enter global mainstream markets

: ★★★★☆ (Low risk, stable return)

:

• Targets high-end training cards like H100/A100
• Adopts advanced Chiplet packaging technology
• Pursues extreme FP8/FP16 computing power density

:

• Positioned for supercomputing centers and internet giants
• Conducts joint R&D with Alibaba, Baidu, etc.
• Strives to occupy a dominant position in the localized training market

:

• Clearest market positioning, directly addressing the biggest pain point
• High technical barriers; huge value once successful

:

• Highest technical risk, severe yield challenges
• Direct competition with NVIDIA; outcome uncertain

: ★★☆☆☆ (Extremely high risk, extremely high risk-return ratio)

:

• Deeply tied to Tencent Cloud, providing cloud-based AI computing power
• Adopts “cloud-native” design, optimizing virtualization performance
• Launched inference cards like T20/T21, highlighting high cost-performance ratio

:

• B2B2C model, reaching end-users through cloud services
• Provides multiple solutions like computing power leasing and private deployment
• Rapidly acquires customers relying on Tencent’s ecosystem

:

• Obvious channel advantages, low customer acquisition cost
• Healthy cash flow, backed by Tencent

:

• Risk of over-reliance on a single customer
• Difficult to grow independently

: ★★★☆☆ (Medium risk, medium return)

:

1. Muxi
   – Strong rigid demand in government and enterprise markets, highest certainty

:
2. – Cloud service binding provides stable income, relatively pragmatic technical route

:
3. – Most complete ecosystem building, largest long-term potential but requires patience

:
4. – High risk and high return, suitable for investors with extremely strong risk tolerance

:

• Main allocation: Muxi (60%) + Suiyuan (30%) + Cash (10%)
• Objective: Stable return, control drawdown

:

• Main allocation: Muxi (30%) + Suiyuan (25%) + Moore Thread (25%) + Birentech (20%)
• Objective: Obtain growth returns under risk control

:

• Main allocation: Moore Thread (40%) + Birentech (30%) + Muxi (20%) + Suiyuan (10%)
• Objective: Maximize long-term returns, bear large fluctuations

1. Technology Iteration Risk
   : AI chip architecture evolves rapidly; falling behind one generation may lead to elimination
2. Ecosystem Construction Lag
   : CUDA ecosystem barrier is extremely high; domestic replacement still takes time
3. Supply Chain Risk
   : Advanced process foundry is constrained by international situation
4. Valuation Bubble Risk
   : Valuations of some companies have reached hundreds of times; need to be alert to corrections

• Shipment Data
  : Monthly GPU shipment volume and market share changes
• Cooperation with Big Players
  : Progress of cooperation with internet giants like BAT and ByteDance
• Process Breakthrough
  : Whether to obtain 7nm and below advanced process capacity
• Ecosystem Migration
  : Migration cost and progress of CUDA code to domestic platforms
• Financial Performance
  : Revenue growth rate, gross profit margin, R&D investment ratio

Domestic GPUs are the “new oil” of the AI era and a must for the “All-in-AI” strategy. However, referring to NVIDIA’s 30-year development history, the final winner is often not the one with the highest starting point, but the one with the clearest route and most firm execution.

:

1. Short-term (1-2 years)
   : Muxi and Suiyuan have the highest certainty and cost-performance ratio
2. Medium-term (3-5 years)
   : Moore Thread has the greatest potential to become “China’s NVIDIA”
3. Long-term (5-10 years)
   : The industry will see integration, and finally 2-3 leading enterprises will remain

: Phased allocation; first layout deterministic targets, then gradually increase growth positions; avoid all-in on a single company or a single technical route. It is recommended to diversify investments through ETFs or index funds instead of betting on a single company (Note: There are no ETFs/index products directly tracking these four companies in the current public market).

[1] Yahoo Finance (Hong Kong Edition) - “Muxi Shares Surge Nearly 700% on First Day of Sci-Tech Innovation Board Listing; AMD Halo and Others Open Valuation Space for the Company” - https://hk.finance.yahoo.com/news/沐曦股份以近570-的漲幅亮相科創板-amd光環等為公司打開估值空間-020941795.html
[2] Yahoo Finance (Hong Kong Edition) - Same Source (Related Analysis and Chart Content) - https://hk.finance.yahoo.com/news/沐曦股份以近570-的漲幅亮相科創板-amd光環等為公司打開估值空間-020941795.html

Due to the 429 error returned by the network interface, more real-time information and price/financial data could not be obtained through news and search tools; the above conclusions are mainly based on limited search results [1][2] and public information for qualitative sorting. Missing key data includes but is not limited to: exact valuations and stock price performance of the four companies, latest shipment volume/revenue/profit, gross profit margin and R&D investment ratio, customer concentration and contract progress, process and yield details, ecosystem migration progress, etc. Without brokerage API and real-time news supplements, this answer should be regarded as a framework analysis and does not constitute investment advice; it is recommended to further verify and quantify with subsequent data updates.