China’s AI Stack May Support Expansion into Cost-Sensitive Markets
Chinese AI ecosystem is broadening beyond models and chips, supporting wider domestic deployment and potentially strengthening its longer-term position in cost-sensitive overseas markets, Fitch Ratings says.
This could reduce reliance on foreign technologies and improve viability of domestic AI systems, but stronger usage may not translate quickly into stronger profitability.
China’s next phase of AI development is likely to depend less on benchmark gains and more on deployment efficiency across the wider stack.
Recent developments involving DeepSeek, Huawei and data-centre networking suggest China is making progress across multiple layers of the AI stack as it focuses on software and deployment infrastructure rather than model capability alone.
DeepSeek said it was open-sourcing programming infrastructure for Huawei’s Ascend platform, including computing and communication libraries, while the two companies jointly advanced a 128-chip Ascend 950 supernode.
The development suggests continued efforts to improve software support around domestic hardware and strengthen the broader AI ecosystem.
China’s efforts to achieve self-sufficiency are extending beyond models and processors. Chinese authorities are reportedly reviewing reliance on Broadcom switches in state-backed data centres, according to the Financial Times.
Networking and interconnect systems are critical components of large AI clusters, suggesting domestic substitution initiatives are broadening into additional infrastructure layers.
Taken together, these latest developments indicate that China’s domestic AI ecosystem is developing across software, processors, networking and infrastructure.
Significant constraints remain in advanced semiconductors, memory, software maturity and cluster efficiency, but progress across the wider stack could support adoption in some emerging markets where affordability matters more.
China’s leading model developers are also expanding the supply of lower-cost, open-weight models that can be customised and deployed by enterprises, governments and cloud providers on their own infrastructure.
These models may become increasingly relevant for applications such as coding assistance, document processing, workflow automation, customer service and enterprise knowledge management, where deployment costs, data control and implementation flexibility can be as important as frontier model performance.
Leading US providers continue to hold advantages in advanced chips, hyperscale infrastructure, software ecosystems and cloud platforms.
Cost-sensitive Global South markets may provide a longer-term route for Chinese providers’ overseas expansion, but the opportunity remains prospective.
Web traffic and model downloads indicate international interest, but do not establish Chinese providers’ share of the overall market, enterprise adoption or revenue.
Commercial relevance would require measurable ex-China usage and investment in overseas infrastructure or local partnerships to support service performance, distribution and customer requirements.
Lower-cost open-weight models already contribute to lower deployment costs and wider ecosystem reach, including through international developer interest.
Their more immediate commercial relevance, however, is likely to remain within China, where continued development across software, processors, networking and infrastructure could support broader enterprise adoption.
The commercial implications remain uncertain. Broader adoption may support demand for cloud services, infrastructure and software integration, but wider usage does not necessarily imply stronger profitability.
Evidence suggests adoption is expanding faster than monetisation, with many AI deployments still focused on productivity and efficiency gains rather than direct revenue generation.
Anthropic’s $2trn IPO Puts AI’s Circular Money Machine on Trial
