Homegrown AI chips are rapidly reshaping China’s GPU cloud market. As AI demand surges, domestic providers are gaining control. At the same time, geopolitical pressure has reduced access to foreign hardware. As a result, local chipmakers have moved into the spotlight. By early 2025, Baidu and Huawei together controlled more than 70% of the market. This shift signals a broader push toward technological self-reliance.
Moreover, this transformation extends beyond two firms. Emerging players, IPO-driven funding, and ecosystem maturity are reinforcing China’s position. Consequently, the country is reducing long-term dependence on overseas GPU suppliers.
Dominant Market Positions for Homegrown AI Chips
Homegrown AI chips now anchor China’s GPU cloud leadership. Baidu leads the market with a 40.4% share in the first half of 2025. The company relies on Kunlunxin processors integrated into its Baige platform. These chips support scalable AI training and inference.
Meanwhile, Huawei follows closely with a 30.1% share. It deploys Ascend chips in dense clusters built for large models. Together, both companies control the full stack. This control spans chip design, system integration, and cloud delivery. As a result, adoption of domestic silicon has accelerated far faster than rival platforms.
In addition, both firms pool thousands of proprietary GPUs into unified virtual resources. This structure supports large models without foreign hardware. Consequently, the market has shifted from isolated chip design toward integrated ecosystems. In many ways, this mirrors Nvidia’s CUDA-driven strategy.
Catalysts Accelerating Homegrown AI Chips
Several factors explain the surge in homegrown AI chips. First, U.S. export controls since 2022 restricted access to Nvidia’s A100 and H100 GPUs. As a result, demand shifted toward local alternatives. Huawei’s Ascend 910C and Baidu’s M100 and M300 benefited directly.
Second, AI workloads expanded quickly. Large language models, multimodal systems, and enterprise inference all drove demand. In 2024 alone, Baidu shipped nearly 70,000 Kunlunxin units. This scale reflects production readiness rather than experimentation.
At the same time, policy support reinforced momentum. Beijing pushed for self-sufficiency in advanced computing. State-backed funding flowed into interconnects, packaging, and power efficiency. Consequently, development cycles shortened.
IPO activity further boosted expansion. Baidu filed confidentially for a Kunlunxin listing in Hong Kong. Biren Technology raised major funding. Moore Threads surged after its Shanghai debut. Enflame and MetaX followed with capital raises. Together, these listings fueled fabrication and software investment.
Meanwhile, export pressure intensified. Regulators reportedly urged domestic firms to pause Nvidia H200 orders. This step further accelerated domestic adoption.
Broader Ecosystem Growth Around Homegrown AI Chips
Beyond leaders, homegrown AI chips are reshaping the wider ecosystem. Alibaba deployed its Hanguang chips in clustered environments. As a result, Nvidia reliance dropped sharply in inference workloads. Tencent adopted hybrid clusters using Ascend, Cambricon, Biren, and Moore Threads chips.
At the same time, Cambricon gained momentum from rising edge-to-cloud inference demand. Startups also played a role. Moore Threads introduced Huagang architectures aimed at dense compute. These designs target 100,000-GPU clusters with higher efficiency.
Moreover, policymakers backed several firms to strengthen competition. This support boosted investment and projected market share growth through 2028. In parallel, China’s cloud services market expanded rapidly in early 2025. AI-driven demand became a primary growth engine. Consequently, GPU cloud deployments accelerated further.
Persistent Hurdles for Homegrown AI Chips
Despite progress, homegrown AI chips still face limits. In raw performance, leading domestic GPUs trail Nvidia’s H200 by several multiples. That gap may widen as Nvidia advances its architectures. Software also remains a challenge. Many platforms still require CUDA porting and optimization.
In addition, integration complexity restricts broader adoption. Most deployments remain concentrated among large hyperscalers. However, targeted optimizations are closing gaps. Baidu’s PaddlePaddle-Kunlun stack and Huawei’s CANN framework improve cost efficiency. These gains matter most for Chinese-language and localized workloads.
From Strategic Substitute to Industry Pillar
Over the next three to five years, homegrown AI chips are likely to shift roles. They may move from strategic backups to core AI infrastructure. IPO liquidity and policy support will drive this transition. As interconnects improve and process nodes advance, performance gaps should narrow.
At the same time, software stacks will mature. Optimization frameworks will reduce friction for developers. Continued demand from hyperscalers and enterprises will reinforce scale. Together, these forces position China’s GPU cloud as a durable pillar in the global AI compute landscape.
