China’s GPU Ecosystem in 2026 – Beyond NVIDIA, Who Are the Real Players?

If you follow the global semiconductor industry, you already know the headline: NVIDIA is the undisputed king of GPUs. In the first half of 2026, the global GPU computing chip market reached approximately ¥1.02 trillion, and NVIDIA alone captured 89.3% of that — worth ¥957.3 billion. AMD, the distant second, holds just 4.5%. Intel trails at 0.5%. But beneath this seemingly unassailable dominance, a new ecosystem is quietly taking shape in China — one that is growing at triple-digit rates and attracting the attention of hyperscalers, investors, and supply chain professionals alike.

1. The Global GPU Market — NVIDIA’s Fortress

To understand China’s GPU ecosystem, you first need to appreciate the scale of what Chinese players are up against. The global GPU market is not just large — it is the largest segment of the entire computing chip market, accounting for more than 73% of all computing chip revenue. And it is still growing: from ¥2 trillion in 2026 to an estimated ¥4.4 trillion by 2030, at a CAGR of 22.5%.

The numbers are stark. NVIDIA’s GPU revenue alone exceeds the entire computing chip revenue of all other categories — CPU, ASIC, and FPGA — combined. This level of concentration is “extremely rare in semiconductor industry history”.

But here is where it gets interesting for overseas businesses watching China. While Chinese GPU vendors collectively hold less than 2% of the global market, their domestic share is rising rapidly. In 2025, Chinese AI accelerator shipments reached approximately 1.65 million units out of a total 4 million — a 41% domestic share. And the momentum is accelerating.

2. The Chinese GPU Landscape — Seven Players to Watch

The Chinese GPU ecosystem is diverse, with players pursuing different technical approaches, target markets, and commercialization strategies. Here is a breakdown of the seven most significant players, based on H1 2026 market share data.

While the global shares appear modest, the growth rates are anything but. Several of these companies are growing at 100%+ year-over-year, and the domestic GPU market is projected to exceed ¥80 billion in 2026, with annual growth exceeding 60%.

3. Company Profiles — Who Is Doing What

🔷 海光信息 (Hygon) — The 800-Pound Gorilla

海光信息 (688041.SH) is the largest Chinese GPU/DCU player by revenue. In H1 2026, the company reported ¥9.10 billion in revenue, up 66.52% year-over-year, with net profit of ¥1.80 billion. The company’s “CPU + DCU” dual-product strategy has proven highly effective: its CPUs, compatible with the x86 instruction set, have achieved large-scale deployment in finance, telecommunications, and other critical industries.

Its DCU (Deep Computing Unit) products, based on the GPGPU architecture, have been fully adapted to more than 400 mainstream large models, covering 99% of non-closed-source large models globally. The company’s software stack has achieved over 99% operator coverage, making it one of the most CUDA-compatible domestic alternatives. Hygon’s next-generation Deep Computing Unit is expected in H2 2026, with performance projected to double compared to the previous generation.

🟣 摩尔线程 (Moore Threads) — The Full-Function Challenger

摩尔线程, listed on the STAR Market in late 2025, reported H1 2026 revenue of ¥1.74 billion, up 147.42% year-over-year — exceeding its entire 2025 full-year revenue of ¥1.51 billion. The company pursues a full-function GPU strategy, covering AI computing, professional graphics, and consumer graphics.

Its MUSA software stack is compatible with NVIDIA’s CUDA ecosystem through a migration tool called MUSIFY, which helps customers port existing CUDA code. The company’s flagship training-inference card, the MTT S5000, is already in mass production. Moore Threads also announced plans to list on the Hong Kong Stock Exchange, with a market capitalization of approximately ¥269.8 billion as of August 2026.

🟢 天数智芯 (Iluvatar CoreX) — The Inference Specialist

天数智芯, which went public in Hong Kong in early 2026, is positioning itself as a leader in inference GPUs. The company’s revenue in 2025 was approximately ¥1.03 billion, with GPU sales accounting for about 90%. Huatai Securities projects 2026 revenue of ¥3.04 billion, up 194% year-over-year.

The company’s product lineup includes the 天垓 (Tiāngāi) training series and the 智铠 (Zhìkǎi) inference series. ByteDance is reportedly in discussions to purchase 50,000 AI chips from Iluvatar for its Dou+ AI chatbot. If the deal closes, it would nearly equal the company’s total 2025 shipments of 52,000 units.

🟡 沐曦股份 (MetaX) — The Training-Inference Integrator

沐曦股份 (688802.SH) focuses on training-inference integrated GPUs. The company’s revenue grew from ¥743 million in 2024 to ¥1.64 billion in 2025, a 121% increase. Analysts project 2026 revenue of ¥3.10 billion, with the company expected to reach profitability in 2026.

Its product lineup includes the 曦云 C-series (training-inference), 曦思 N-series (inference), and the upcoming 曦彩 G-series (graphics rendering). The company’s core team has deep AMD backgrounds, with the CTO and other key executives having previously served as AMD Fellows. MetaX holds approximately 1.7% of the Chinese market.

🔴 燧原科技 (Enflame) — The DSA Maverick

燧原科技 recently passed its STAR Market IPO review. Unlike most Chinese GPU vendors, Enflame has chosen a DSA (Domain-Specific Architecture) approach and does not attempt to be CUDA-compatible. This is a bold bet: it means higher customer adoption costs, but potentially better performance for specific workloads.

The company’s revenue has grown from ¥300 million in 2023 to ¥990 million in 2025, and H1 2026 revenue is projected at ¥1.06-1.15 billion. Notably, Tencent accounted for 74.9% of Enflame’s 2025 revenue — a concentration risk that the company will need to address. Enflame’s AI accelerator cards and modules accounted for approximately 1.7% of the Chinese AI accelerator card market in 2025.

🔵 壁仞科技 (Biren) — The High-Performance Contender

壁仞科技 (06082.HK), listed in Hong Kong, is the “GPGPU” specialist. The company expects H1 2026 revenue of ¥1.15-1.30 billion, up from just ¥58.9 million in H1 2025 — a 1,850-2,100% increase. Biren’s 2025 full-year revenue was ¥1.04 billion.

Biren’s market share in China’s GPGPU market was just 0.20% in 2024, but analysts project rapid growth: 2026 revenue estimates range from ¥2.38 billion to ¥2.52 billion, with 2027 projections as high as ¥9.63 billion. The company is benefiting from the shift toward inference workloads and is seen as a scarce GPGPU asset in the domestic market.

🟠 景嘉微 (Jingjia Micro) — The Graphics Veteran

景嘉微 (300474.SZ) is the oldest player in this group, with deep roots in graphics display and defense applications. In H1 2026, the company reported ¥334 million in revenue, up 72.96% year-over-year, though it remained unprofitable with a net loss of ¥108 million.

The company’s graphics display products (JM series) remain its core revenue driver, contributing ¥239 million in H1 2026, up 182%. Jingjia Micro is a classic example of a defense-to-commercial transition — its products primarily serve military and trusted computing markets where domestic substitution is mandatory.

4. Four Technical Approaches — How Chinese GPU Vendors Compete

Chinese GPU vendors have adopted four distinct technical approaches, each with different trade-offs in terms of ecosystem compatibility, performance, and customer adoption costs.

📌 Full-Function GPU — Graphics + Compute

摩尔线程 and 景嘉微 pursue this approach. Their chips handle both graphics rendering and AI compute, making them versatile but also more complex to design. Moore Threads has completed five generations of architecture iteration and is planning a new “Huagang” architecture for 2026.

📌 GPGPU — CUDA-Compatible General-Purpose GPU

海光信息, 壁仞科技, 天数智芯, and 沐曦股份 all follow the GPGPU route, emphasizing CUDA compatibility to reduce customer migration costs. Hygon’s DCU achieves over 99% operator coverage, while Biren and MetaX are building on the same foundational approach.

📌 Full-Stack Self-Development — End-to-End Control

燧原科技 takes the most distinctive path: DSA architecture with a completely self-developed software stack, not compatible with CUDA. This approach offers potentially better performance for specific workloads but faces higher customer adoption costs because developers must learn a new ecosystem from scratch.

📌 Inference-Focused — Specialised for Reasoning

Several vendors, including 天数智芯 and 寒武纪, have placed particular emphasis on inference workloads — the fastest-growing segment of AI compute. Inference chips require less advanced manufacturing and face weaker ecosystem lock-in than training chips, making them the most realistic entry point for domestic vendors.

5. Opportunities and Challenges — What Lies Ahead

📈 Opportunities

• Domestic substitution is accelerating. Chinese GPU substitution is projected to grow from ¥200 billion in 2026 to ¥750 billion by 2030, with the substitution rate rising from 48% to 68%. This represents a cumulative替代 market of approximately ¥2.2 trillion over five years.
• Inference is the entry point. Inference workloads require less advanced process technology and face weaker ecosystem lock-in than training. This is where Chinese vendors — particularly 天数智芯, 燧原科技, and 沐曦股份 — are gaining traction.
• Hyperscaler adoption is growing. ByteDance, Alibaba, and Tencent are all actively deploying domestic GPUs for non-critical workloads. This creates a real-world testing ground for domestic chips.
• Policy support is strong. From the ¥344 billion National Big Fund III to mandatory domestic procurement quotas, the policy environment is unambiguously supportive of domestic GPU adoption.

⚠️ Challenges

• The CUDA ecosystem is a fortress. NVIDIA has accumulated more than 600,000 registered developers, with 100% operator coverage. Migrating away from CUDA is expensive and time-consuming — a barrier that GPGPU vendors are trying to lower through compatibility tools.
• Advanced manufacturing is constrained. Chinese GPU vendors rely on domestic foundries like SMIC, which are 2-3 generations behind TSMC in advanced process technology. This limits peak performance and power efficiency.
• Profitability is elusive. Most Chinese GPU vendors remain unprofitable. 摩尔线程, despite ¥1.74 billion in H1 revenue, still reported a net loss of ¥11.56 million (though adjusted profit was positive). 沐曦股份 is expected to reach profitability in 2026, but others are further from that milestone.
• Customer concentration is high. 燧原科技’s reliance on Tencent (74.9% of revenue) is the most extreme example, but several vendors depend heavily on a small number of large customers — a risk if those customers switch suppliers.

6. What This Means for Overseas Businesses

If your company has supply chain relationships, investments, or partnerships in China’s tech sector, the evolving GPU landscape has direct implications for you.

• Your Chinese partners are likely transitioning to domestic GPUs. With export controls restricting access to NVIDIA’s most advanced chips and domestic procurement mandates kicking in, many Chinese firms are actively replacing NVIDIA GPUs with domestic alternatives for inference and some training workloads.
• New vendors are entering the supply chain. Companies like 海光信息, 摩尔线程, and 天数智芯 are becoming legitimate suppliers for major Chinese enterprises. If you are doing business with Chinese tech companies, you need to know who their GPU suppliers are — and whether those suppliers are compliant with both US and Chinese regulations.
• Due diligence must include GPU supply chain verification. The chip source, technology licensing, and export control exposure of your Chinese partners are now critical compliance considerations. A partner that relies on NVIDIA chips may face supply disruptions; a partner that has switched to domestic chips may face different risks.
• Regulatory risk is asymmetric. US export controls and Chinese domestic substitution mandates are pulling in opposite directions. Companies that understand both regulatory regimes — and their partners’ positions within them — will be better positioned to manage risk.

7. How ChinaBizInsight Can Help You Navigate the GPU Landscape

At ChinaBizInsight, we specialise in providing overseas businesses with reliable, verifiable information about Chinese companies. As China’s GPU ecosystem undergoes its most dramatic transformation, the need for accurate due diligence has never been greater.

We help overseas businesses:

• Verify Chinese company credentials — including business licences, shareholder structures, and director information — through official government sources.
• Access comprehensive credit reports that go beyond basic registration to include legal risks, financial health, and operational history.
• Obtain notarisation and apostille services for Chinese corporate documents, ensuring they are recognised in your home jurisdiction.
• Conduct specialised due diligence on companies in the semiconductor and AI infrastructure sectors, including GPU design, manufacturing, and supply chain relationships.

Final Take — A New Ecosystem Is Taking Shape

NVIDIA’s dominance of the global GPU market is unchallenged at the global level — 89.3% market share, ¥957 billion in H1 2026 revenue, and a CUDA ecosystem that has taken nearly two decades to build. But in China, a different story is unfolding.

Seven domestic vendors — 海光信息, 摩尔线程, 天数智芯, 沐曦股份, 燧原科技, 壁仞科技, and 景嘉微 — are growing at triple-digit rates, capturing an expanding share of the Chinese market. The domestic GPU market is projected to exceed ¥80 billion in 2026, with substitution reaching ¥750 billion by 2030.

For overseas businesses, this transformation brings both risks and opportunities. The key to navigating this new landscape is reliable, up-to-date information about your Chinese partners and their GPU supply chains. Know who you are doing business with — because the landscape has changed, and it will never be the same.