Morgan Stanley is shifting the calculus for success in China's AI chip market, prioritizing economic efficiency and software ecosystems over raw chip specifications.
Morgan Stanley initiated coverage on three of China’s domestic AI chip suppliers on April 26, forecasting the market will reach $67 billion by 2030 as the industry’s focus shifts from policy-driven growth to economic returns and software maturity.
The bank’s analysts argue that winning is no longer just about chip specifications, but "depends on the economics of unit token cost, the maturity of the software ecosystem, and deep customer-binding relationships."
The report assigned "Overweight" ratings to Cambricon and MetaX, while giving Moore Threads an "Equal-weight" rating. Cambricon is recognized for its large-scale deployment with cloud providers like ByteDance, while MetaX has secured large pre-orders and offers a differentiated, diversified supply chain.
This new framework suggests the race for China's AI dominance is entering a new phase where business strategy and ecosystem control are more critical than technical specs alone, a shift that could reshape investment flows toward companies with proven customer traction and a clear path to profitability.
Cambricon Leads With 79% Revenue Concentration
Morgan Stanley views Cambricon as the domestic leader in AI inference chips, primarily due to the successful large-scale deployment of its MLU590 chip in the search, advertising, and recommendation systems of major cloud service providers, most notably ByteDance. This close, multi-year partnership has created a significant customer anchor and allowed for continuous hardware and software co-optimization. The report highlights that Cambricon became the first domestic AI chip company to achieve sustainable profitability in 2025, with revenues of 6.50 billion yuan. The bank forecasts Cambricon’s revenue will grow 222% to 20.94 billion yuan in 2026. However, the heavy reliance on a single customer is a key risk, with ByteDance contributing 79.15% of revenue in 2024.
MetaX De-Risks With Supply Chain and Large Pre-Orders
MetaX earns its "Overweight" rating from a differentiated strategy focused on supply chain diversification. Unlike peers who rely on a single foundry, MetaX utilizes multiple compliant production lines, enhancing its capacity and reducing the risk of supply disruptions. According to the report's channel checks, the company has already secured large pre-orders from major Chinese cloud providers, with shipments expected to begin in the second quarter of 2026 and ramp up significantly in the second half. These orders are projected to contribute over 4 billion yuan in revenue between 2026 and 2027. The bank projects MetaX’s 2026 revenue will reach 3.06 billion yuan, growing to 7.5 billion in 2027, with the company expected to hit profitability in the second half of 2026.
Moore Threads' Software Edge Not Yet Enough
Moore Threads receives a more cautious "Equal-weight" rating. The company's primary advantage is its MXMACA software platform, which supports over 6,000 CUDA APIs, offering a relatively seamless migration path for users of Nvidia's dominant platform. While the company has achieved mass production with stable yields, it lags significantly behind Cambricon and MetaX in penetrating the major cloud service provider market. The report notes that large-scale commercial deployment still requires validation. Morgan Stanley forecasts 2026 revenue of 4.03 billion yuan, but at a target price-to-sales ratio of 75 times, its valuation appears less attractive than its "Overweight"-rated peers.
This article is for informational purposes only and does not constitute investment advice.
