The expanding demand for AI computing power is creating structural growth opportunities for China's AI chip market. In a report released on April 26, Morgan Stanley initiated coverage on three Chinese AI chip suppliers, noting a shift in the industry's primary driver from policy support to economic viability. Success in the sector is no longer determined solely by chip specifications but increasingly by the cost per token, the maturity of the software ecosystem, and the depth of customer relationships.
Applying this framework, Morgan Stanley assigned Overweight ratings to Cambricon and MetaX, and an Equal-weight rating to Moore Threads. Each company has a distinct strategic position. Cambricon has emerged as a leader in cloud inference, driven by the large-scale deployment of its MLU590 chips in cloud service providers like ByteDance. MetaX has carved out a differentiated advantage through its diversified supply chain strategy and has secured significant pre-orders for its AI chips, indicating high order visibility. Moore Threads is distinguished by its software compatibility features, though its customer penetration lags behind peers, and its current valuation appears less attractive.
Channel checks indicate that inference demand remains robust, with both token prices and GPU rental rates on the rise. However, price competition has emerged earlier than anticipated, with some vendors already reducing prices to gain market share, suggesting the industry may be entering a market share battle phase sooner than expected. Morgan Stanley maintains a positive long-term outlook on China's AI chip market, projecting it will grow to $67 billion by 2030.
Cambricon: Cloud Inference Leader with Strong Customer Lock-in The report positions Cambricon as a leading domestic AI inference chip play, supported by three key factors. First, its MLU590 chips are already deployed at scale within the search, advertising, and recommendation systems of major cloud service providers like ByteDance, creating significant customer lock-in. Second, its multi-year collaboration with ByteDance has enabled continuous hardware-software co-optimization. Third, Cambricon achieved full-year profitability in 2025, becoming the first sustainable profitable AI chip company in China, with revenue reaching RMB 6.497 billion.
The report forecasts Cambricon's 2026 revenue will reach RMB 20.944 billion, representing approximately 222% year-on-year growth, surpassing the market consensus of RMB 14.057 billion. The company anticipates a revenue compound annual growth rate of about 90% from 2026 to 2028 and projects a 2026 net profit of RMB 6.156 billion. Key risks include high customer concentration, with ByteDance accounting for 79.15% of its 2024 revenue, and intensifying competition from leading peers.
MetaX: Diversified Supply Chain and Large Orders Underpin Growth The Overweight rating on MetaX is based on the differentiated advantage offered by its diversified supply chain strategy. Unlike peers reliant on a single wafer fab, MetaX employs a multi-foundry approach, allowing production across various compliant production lines. This enhances production capacity certainty and helps mitigate supply disruption risks.
Channel checks indicate that major domestic cloud service providers have placed substantial pre-orders for MetaX's AI chips, with shipments expected to commence in Q2 2026 and scale significantly in the second half of the year. These orders are projected to contribute over RMB 4 billion in revenue for MetaX from 2026 to 2027. On the software front, MetaX's GPGPU architecture is highly compatible with CUDA, having already assisted customers in migrating large models from Nvidia platforms to its own, effectively reducing migration friction.
Forecasts project MetaX's 2026 revenue at RMB 3.06 billion, up 196% year-on-year, rising to RMB 7.5 billion in 2027. The revenue CAGR from 2025 to 2028 is estimated at approximately 122%. The company is expected to achieve quarterly profitability in the second half of 2026 and full-year profitability in 2027. Key downside risks include potential delays in chip shipment schedules, obstacles in production ramp-up, and a postponement of the profitability inflection point due to heightened competition.
Moore Threads: Software Compatibility a Highlight, Customer Traction Needs Validation The Equal-weight rating for Moore Threads reflects a balanced risk-reward profile. The company's core differentiation lies in its MXMACA software platform, which supports over 6,000 CUDA APIs, providing a relatively low-friction migration path for Nvidia users. Furthermore, the company has achieved mass production, with better supply chain visibility and yield stability than some peers.
However, Moore Threads lags behind Cambricon and MetaX in penetrating cloud service provider customers, and its large-scale commercial deployment still requires time to validate. The company reported 2025 revenue of RMB 1.644 billion but has not yet reached profitability. Revenue for 2026 is forecast at RMB 4.03 billion, representing approximately 145% growth, with a projected CAGR of about 66% from 2026 to 2028. The company is expected to achieve its first profit in 2026, with EPS around RMB 0.05. A target price of RMB 758 implies a 2026 price-to-sales ratio of 75x, making its current valuation appear less attractive compared to Cambricon and MetaX.
Zooming out to the industry level, Morgan Stanley reiterates its long-term view on China's AI chip market. It projects the total addressable market will reach $67 billion by 2030, implying a CAGR of approximately 23% from 2024 to 2030. The core driver of this growth is the rapid commercialization of AI inference demand. Channel checks suggest that AI-related capital expenditure by major Chinese tech platforms is expected to increase 38% year-on-year to RMB 597 billion in 2026. The monthly token processing volume on platforms like ByteDance has surged significantly, confirming robust downstream demand. As the market landscape continues to evolve, determining the ultimate winners will require more time.