When Western analysts talk about China's AI chip industry, they usually talk about one company: Huawei. That focus is understandable. Huawei's Ascend series has become the dominant domestic accelerator, and its market share gains have been dramatic. But Huawei is not the whole map. It is the largest peak on a mountain range that now includes at least a dozen companies, four of which have completed IPOs in the past ten months, and several more preparing to raise capital.
The full picture matters because the competitive dynamics inside China's AI chip market are shifting. The question is no longer whether domestic chips can replace Nvidia. The more relevant questions are which domestic chips, built on which architectures, with which software ecosystems, and at which price points.
The Market Structure: One Superpower, Many Strong Players
The Chinese phrase used to describe the current market is "one superpower, many strong players." Huawei dominates, and below it sits a dense layer of smaller competitors.
Huawei is the superpower. According to Bernstein Research, Huawei is projected to reach 50 percent of China's AI chip market in 2026, up from roughly 40 percent in 2025. AMD is expected to hold 12 percent, followed by Cambricon at 9 percent, Hygon Information at 8 percent, T-Head (Alibaba's chip unit) at 5 percent, and Kunlunxin (Baidu's chip unit) at 3 percent. Nvidia's share, which was approximately 95 percent in 2022, is projected to fall to 8 percent.
The domestic share of China's AI accelerator market crossed 50 percent in 2025, up from less than 30 percent in 2024, according to Bank of America and JPMorgan. By 2028, domestic suppliers are expected to approach 80 percent.
The shift is driven by a combination of supply constraints on Nvidia's China-specific products, the maturation of domestic alternatives, and a pricing advantage that domestic chips hold in inference workloads. Bank of America noted that the core advantage of Chinese models lies in inference cost efficiency, not frontier training capability.
The Four That Went Public
The most visible change in China's AI chip industry in 2026 has been the wave of IPOs. Four domestic GPU companies — Moore Threads, MetaX, Biren Technology, and Enflame Technology — have all listed on public markets. They are collectively known as the "Four Little Dragons" of Chinese AI chips.
Biren Technology was the first. It listed on the Hong Kong Stock Exchange on January 2, 2026, raising HK$5.58 billion ($717 million) at HK$19.60 per share. The stock closed its first day up 76 percent, and institutional demand was nearly 26 times the shares on offer. Biren, founded in 2019, develops general-purpose GPUs for AI and high-performance computing. In July 2026, it raised an additional HK$7.07 billion through a share placement, and in September it was reportedly considering another $1 billion raise.
Moore Threads listed on Shanghai's STAR Market in December 2025 at 114.28 yuan per share, raising approximately 8 billion yuan. It was the first domestic GPU company to go public. In the first half of 2026, Moore Threads reported revenue of 1.736 billion yuan, a year-on-year increase of 147.42 percent. In September 2026, it announced plans for a Hong Kong listing, aiming to raise at least $1 billion.
MetaX listed alongside Moore Threads on the STAR Market and reported first-half 2026 revenue of 1.324 billion yuan, up 44.67 percent. Its net loss narrowed significantly, and in the second quarter it achieved a single-quarter profit on a non-GAAP basis.
Enflame Technology listed on the Shanghai Stock Exchange on September 11, 2026, raising $912 million by selling 43 million new shares at 142.18 yuan each. The stock surged nearly 200 percent on its debut. Enflame specializes in AI inference chips. Its first-half 2026 revenue reached 1.12 billion yuan, a year-on-year increase of 279 percent. Tencent is its largest shareholder, holding approximately 18 percent after the IPO. Enflame expects to reach break-even in late 2026 or 2027.
The combined market capitalization of the three already-listed companies — Moore Threads, MetaX, and Biren — stood at approximately 171.1 billion yuan, 197.5 billion yuan, and HK$97.1 billion respectively as of September 10, 2026.
Beyond the Four: The Full Roster
The "Four Little Dragons" get the IPO headlines, but they are not the only companies building AI chips in China.
Huawei remains the dominant player by a wide margin. Its Ascend 910C, manufactured by SMIC on a 7-nanometer-class process, delivers approximately 60 percent of Nvidia H100 performance in real-world testing. Huawei's answer to the single-chip gap is architectural: the CloudMatrix 384 supernode packs 384 chips into a single logical unit, delivering 1.7 times the BF16 performance of Nvidia's GB200 NVL72 system. Huawei expects AI chip revenue of approximately $12 billion in 2026.
Cambricon is the second-largest domestic player by market share. Its SiYuan 590 chip is benchmarked against Nvidia's A100. Cambricon plans to more than triple its production in 2026, aiming to take share from Huawei and fill the void left by Nvidia. Third-party estimates place its 2025 cloud AI chip shipments at approximately 100,000 units.
Hygon Information uses a "CUDA-compatible" ecosystem through its Deep Computing Unit (DCU) series, allowing customers to migrate workloads with minimal code changes. It holds approximately 8 percent of the domestic market.
T-Head, Alibaba's chip unit, has shipped hundreds of thousands of its Zhenwu PPU (Parallel Processing Unit) chips, surpassing Cambricon in cumulative volume. The PPU features 96 GB of HBM2e memory and 700 GB/s inter-chip interconnect bandwidth. Alibaba uses the PPU extensively for training and inference on its Qwen models.
Kunlunxin, Baidu's chip subsidiary, has deployed its P800 chips in clusters exceeding 10,000 cards as of February 2025, with plans to scale to 30,000 cards. Its products serve Baidu and enterprise customers in finance, energy, and manufacturing.
Enflame and Biren are covered above. MetaX and Moore Threads round out the publicly traded group.
Beyond these, there is a second tier of emerging companies. Yuanli Semiconductor raised over 500 million yuan in a Pre-A round in April 2026 for edge inference chips. Fangqing Technology completed a 1 billion yuan Pre-A round in March 2026. XiWang raised over 1 billion yuan in April 2026, becoming the first pure-inference GPU unicorn valued at over 10 billion yuan. Guangyu Xinchen raised nearly 2 billion yuan over six months, closing an A round in September 2026.
The Architecture Divide
The Chinese AI chip industry is split between two architectural philosophies.
GPGPU architecture, used by Moore Threads, MetaX, and Biren, is compatible with Nvidia's CUDA ecosystem. This reduces migration costs for developers, who can port existing code with relatively minor modifications. Moore Threads has developed its own MUSA architecture, and its developer community has grown to over 800,000.
DSA architecture — domain-specific architecture — is used by Huawei's Ascend, Cambricon, and Google's TPU. These chips are designed for specific workloads and can achieve higher efficiency in those workloads, but they require developers to learn new programming models. Huawei open-sourced its CANN software stack in December 2025, a move designed to accelerate ecosystem adoption.
The two architectures are expected to coexist. GPGPU chips are easier to adopt for teams migrating from Nvidia. DSA chips offer better performance-per-watt for teams willing to invest in a new software stack.
The Software Ecosystem Challenge
The hardest problem for Chinese AI chip companies is not silicon. It is software.
Nvidia's CUDA has been built over nearly two decades and supports millions of developers. Chinese chipmakers are approaching the problem from two directions. Some, like Moore Threads and Hygon, build CUDA-compatible layers that allow existing code to run with minimal changes. Others, like Huawei with CANN and Cambricon with NeuWare, are building proprietary software stacks from scratch.
The results are mixed. A 2026 industry analysis noted that domestic chips have "generally recorded revenue growth with narrowing losses" and are "approaching their profitability inflection point." But the analysis also warned that competition has moved from research and development to market shakeout — and that the companies that survive will be those that can integrate "R&D investment, product performance, software ecosystem, and customer orders into a sustainable commercial chain."
What the Map Reveals
The Chinese AI chip industry is no longer a single-company story. It is a portfolio of companies at different stages of maturity, pursuing different architectures, targeting different segments of the market.
Huawei dominates. But below Huawei, there is a dense layer of companies that have raised capital, listed on public markets, and are generating revenue. Their combined share of the domestic market is growing. Their software ecosystems are maturing. Their losses are narrowing.
The question is not whether China can build AI chips without Nvidia. It already can. The question is whether the companies building them can build sustainable businesses — and whether the market is large enough to support a dozen competitors when the price of inference is falling.
The map is no longer a single peak. It is a range.
Sources: Bernstein Research via EET China (January 2026); Bank of America via ChinaAET (September 2026); JPMorgan via Reuters (September 2026); CCTV+ (September 2026); Securities Times via CNFin (September 2026); Bloomberg (September 2026); SCMP (July 2026); Forbes (September 2026); Financial Times Chinese (September 2026); Baidu Baike (September 2026); AskCI Consulting (July 2026); DigitalToday (June 2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Bernstein Research via EET China (January 2026)
- Bank of America via ChinaAET (September 2026)
- JPMorgan via Reuters (September 2026)
- CCTV+ (September 2026)
- Securities Times via CNFin (September 2026)
- Bloomberg (September 2026)
- SCMP (July 2026)
- Forbes (September 2026)
- Financial Times Chinese (September 2026)
- Baidu Baike (September 2026)
- AskCI Consulting (July 2026)
- DigitalToday (June 2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.