China's AI Chip Market Darling Cambricon Faces Its Ultimate Test

摘要： 中国的AI芯片市场明星企业寒武纪面临最终考验，其表现将决定其在竞争激烈的AI芯片市场中的地位，面对这一挑战，寒武纪需展示其技术实力和市场适应能力，以应对日益增长的AI应用需求。...

中国的AI芯片市场明星企业寒武纪面临最终考验，其表现将决定其在竞争激烈的AI芯片市场中的地位，面对这一挑战，寒武纪需展示其技术实力和市场适应能力，以应对日益增长的AI应用需求。

AI-generated image

TMTPOST -- In the global race for artificial intelligence, the stakes are high—and so are the costs. Even deep-pocketed tech giants like OpenAI are reconsidering how much they can afford to burn on AI chips.

According to the Financial Times, OpenAI is reportedly planning to collaborate with U.S. semiconductor powerhouse Broadcom to mass-produce its own AI chips, aiming to create customized alternatives to Nvidia’s offerings. The first of these chips are expected to hit the market next year, signaling the intensifying competition in the AI hardware arena.

Meanwhile, in China, the domestic AI chip industry has been red-hot. Cambricon, one of the country’s leading AI chipmakers, has seen its stock price soar by an astonishing 468% over the past year, even surpassing Moutai at one point to become the most valuable company on China’s A-share market. Yet this meteoric rise has sparked a heated debate: is Cambricon merely an overhyped bubble, or is it truly the crown jewel of China’s AI computing ecosystem?

Stock Price Soaring Nonstop

“Cambricon represents our hope for an era of explosive growth, and the ability to survive any potential extinction events to become the most adaptable species,” said founder Chen Tianshi, explaining the meaning behind the company’s unusual name. The analogy is apt: the company has ridden the wave of China’s chip industry boom, fueled in part by investor enthusiasm and capital market speculation.

On August 28, 2025, Cambricon’s market capitalization peaked at 663.7 billion yuan, temporarily surpassing Moutai to claim the top spot in A-shares. Yet, the company itself issued a stock trading risk alert the same day, warning investors that the current valuation might be far removed from underlying fundamentals.

Financially, the warning is warranted. Cambricon projects annual revenue of only 5 to 7 billion yuan for 2025 and has no new product launches planned this year. Its latest AI training card, the MLU370-X8, was released in March 2022. For comparison, global competitors like Nvidia often release new chips every one to two years, reflecting the rapid pace of innovation in the sector.

Despite this, the stock has continued to soar, up more than 468% over the past year and more than 108% over the past eight months alone. As Shanghai Securities News commented, “While Cambricon’s explosive performance is certainly impressive, a triple-digit price-to-earnings ratio means its valuation has already priced in much of its future growth potential.”

From a market perspective, however, the rationale behind Cambricon’s rising stock price becomes clearer. According to IDC, China’s AI chip supply in 2024 is expected to reach 2.8 million units, but domestic chips will account for only 820,000 units, leaving the majority of the market dominated by foreign suppliers such as Nvidia.

Against this backdrop, the government’s push for domestic AI chip production has created massive growth opportunities. Investment banks have taken notice. Goldman Sachs, for instance, has raised Cambricon’s target price three times within a year, from 607.8 yuan in April to 1,223 yuan, then to 1,835 yuan in August, and most recently to 2,104 yuan on September 1. Despite public concerns about overvaluation, Wall Street appears confident that Cambricon’s stock still has room to grow.

China’s AI chip landscape is diverse, including startups such as Biren Technology and Moore Threads, as well as projects backed by giants like Huawei’s HiSilicon and Baidu’s Kunlun chip. But Cambricon occupies a unique strategic niche that enhances its scarcity value.

The AI chip market in China is projected to reach $30 billion by 2025, growing at a compound annual rate of over 40%. Within this market, cloud-based training and inference chips—used in intelligent computing centers—represent the primary growth driver. According to official data, more than 393 project bids for domestic intelligent computing centers were published in the first half of 2025. Beijing aims for 100% self-reliant technology in intelligent computing centers by 2027, while Shenzhen targets 25 EFLOPS of computing power by the end of 2025.

Policies further support the sector. In late 2023, the Ministry of Industry and Information Technology issued the “Action Plan for High-Quality Development of Computing Power Infrastructure,” which set a goal of achieving 35% of national computing power via intelligent computing by 2025. TrendForce analyst Gong Mingde told Times Weekly, “The adoption rate of domestic AI chips is expected to exceed 60% in both AI training and AI inference applications. Localization of next-generation computing power chips is accelerating.”

This makes 2025 a milestone year for computing power chips in China, with demand from intelligent computing centers expected to surge. Yet among domestic manufacturers, few focus on cloud-based training and inference. Many concentrate on edge chips or specialized applications: Horizon Robotics and Black Sesame for autonomous driving, Moore Threads and Biren Technology for GPU-centric tasks such as graphics rendering and video acceleration.

Cambricon, along with Huawei’s Ascend series and Baidu’s Kunlun chips, is one of the few domestic players specializing in cloud-based computing power chips. Since Ascend and Kunlun are backed by major tech giants and are not under immediate pressure to go public, Cambricon stands out as a rare domestic AI chip target capable of commanding a valuation premium.

The Challenges Ahead

While Cambricon enjoys a privileged position, its path forward is far from easy. The domestic market for cloud-based computing power chips is entering a three-way race among Cambricon, Huawei Ascend, and Baidu Kunlun. Each player is aggressively developing next-generation chips designed to rival Nvidia’s dominant offerings.

For example, Huawei’s Ascend 910B competes closely with Nvidia’s A100, while the 910C is expected to rival Nvidia’s H100. Baidu’s third-generation Kunlun chip and Cambricon’s anticipated Siyuan 590 chip aim to provide alternatives to Nvidia’s H20. Yet while Ascend and Kunlun chips are already mass-produced, Cambricon’s latest publicly available products remain the Siyuan 370 and 270 series, with no confirmed launch date for the Siyuan 590.

Production capacity further complicates matters. In 2024, shipments for Huawei Ascend, Baidu Kunlun, and Cambricon reached 640,000, 69,000, and 26,000 units, respectively. Cambricon’s capacity is only 4% and 38% of the first two, constraining revenue growth and limiting participation in major computing center projects.

For instance, in China Mobile’s 2025-2026 procurement for AI inference servers, Kunlun-based products dominated, capturing 70% to 100% of three bid packages, with total contract values reaching billions of yuan. Cambricon, meanwhile, was absent from major contracts.

Cambricon also faces challenges in technology iteration and ecosystem development. NVIDIA dominates the global AI training market with its CUDA ecosystem, while Huawei Ascend builds its MindSpore framework. Cambricon’s own platform, although functional, still lags in developer community, toolchain maturity, and compatibility.

A Stock Market Darling

Despite these challenges, Cambricon’s stock price has soared over the past month, reflecting both the broader push for domestic AI computing power and the high expectations of investors. Goldman Sachs and other foreign institutions see Cambricon as a core player in cloud training and inference, with its technological accumulation and first-mover advantage positioning it for growth under supportive policies and robust market demand.

However, market optimism does not eliminate reality. Financial performance, production capacity, and ecosystem development remain critical hurdles. Cambricon’s ultimate test will be whether it can evolve from a “capital market darling” into a true industry leader—successfully launching next-generation chips and overcoming production bottlenecks. Only then can it validate the hype and prove its long-term value in China’s burgeoning AI sector.

Cambricon’s meteoric rise highlights both the promise and perils of China’s domestic AI chip market. On one hand, the company has leveraged strategic positioning, first-mover advantages, and supportive government policies to capture investor attention and capitalize on a booming AI ecosystem. On the other hand, production limitations, a lack of new products, and stiff competition from tech giants pose real risks to sustaining its growth.

For investors, the lesson is clear: while Cambricon may be the “king” of domestic AI computing power in the eyes of the stock market, the road from market darling to technological powerhouse is fraught with challenges. In the high-stakes world of AI chips, only those who can deliver both innovation and execution will survive—and thrive.