Context: Indian startups are increasingly adopting Chinese large language models (LLMs) such as Qwen, DeepSeek and Kimi to reduce Artificial Intelligence (AI) costs.
- These models offer significant cost advantages while approaching the capabilities of leading U.S. frontier models; however, the article argues that China’s present open-weight strategy may gradually become more restrictive.
Why China is Promoting Open-Weight AI
1. Cost Advantage
- DeepSeek-R1 reportedly cost about $294,000 to train, far below the expenditure of leading U.S. AI laboratories.
- Knowledge distillation and architectural efficiencies have helped Chinese firms reduce research and development costs.
2. Geopolitical Prestige and Diplomacy
- Open-weight AI has become an instrument of technological diplomacy and international influence.
- Examples include China’s promotion of AI cooperation, the proposed World Artificial Intelligence Cooperation Organization (WAICO) bloc and training opportunities for developing countries.
- The January 2025 release of DeepSeek-R1 also demonstrated the potential global market impact of Chinese AI models.
3. Commoditisation of AI
- U.S. firms largely monetise proprietary model weights through commercial products and services.
- Freely available models that are sufficiently capable for most applications can weaken the pricing power and commercial advantage of proprietary frontier models.
- China therefore gains strategically even without consistently producing the world's most commercially successful AI products.
4. State-Directed Capital
- China's financial system channels substantial capital towards strategic sectors, supporting AI firms and potentially encouraging investment and excess capacity.
- By early 2026, around 820 LLMs had reportedly been registered with China's cyberspace regulator, indicating intense domestic competition.
5. Infrastructure Ecosystem
- Open models encourage wider AI adoption, increasing demand for complementary sectors such as cloud computing, energy and physical infrastructure, where Chinese companies have significant capabilities.
- For instance, Alibaba Cloud revenue reportedly grew 34% year-on-year while its Qwen models were being offered openly.
What Could Make China Restrict Open Access?
1. Consolidation
- Managing a small number of dominant AI companies is easier than coordinating hundreds of competing firms.
- China's reported shift from the “Hundred Model War” towards a “Top Five Basic Models” could facilitate greater regulatory control.
- U.S. chip export controls may indirectly accelerate consolidation by increasing costs for Chinese AI firms.
2. Global Lock-in
- China would have greater incentive to restrict models only after foreign developers become sufficiently dependent on Chinese AI models and cloud infrastructure.
- Restricting access too early could encourage users to migrate to alternative U.S. or other open-weight ecosystems.
3. Market Saturation
- Restrictions become more viable once Chinese models have substantially weakened the pricing power of U.S. frontier AI companies and alternative open-weight models can independently maintain competitive pressure.
- Models from companies such as Meta, Mistral and Nvidia could reduce dependence on Chinese open-weight systems.
Possible Form of Future Restrictions
- The transition is more likely to involve graduated restrictions rather than an abrupt closure of access.
- Possible mechanisms include:
- Frontier models available through Application Programming Interfaces (APIs) before their weights are released after a delay.
- Commercial licensing for models above specified capability thresholds.
- Continued free availability of smaller distilled models.
- Open model weights but restricted tool-use and agentic capabilities.
- Preferential access for countries participating in Chinese-led AI cooperation frameworks suchas WAICO.
Implications for India
1. Build Model-Agnostic AI Systems
- Government departments and regulated sectors should avoid excessive dependence on a single AI provider.
- Abstraction layers and interoperable AI architectures can allow systems to switch between different models when availability, cost or geopolitical conditions change.
2. Prioritise Selective AI Capabilities
- India can focus on areas where it has stronger potential advantages: AI applications, industrial and language datasets, edge-inference chip design and domain-specific fine-tuning.
- This approach reflects Atmashakti—building capabilities selectively—rather than pursuing complete technological self-sufficiency.
3. Develop Public-Sector AI Infrastructure
- Instead of focusing exclusively on subsidies for individual Graphics Processing Unit (GPU) resources, India could explore a public-sector equivalent of an AI model-routing platform, allowing government users to access multiple models according to cost, capability and security requirements.
4. Use the Current Open-Access Window
- India can participate actively in multilateral discussions on open-weight AI norms while the ecosystem remains relatively open.
- It can simultaneously expand domestic capabilities to reduce vulnerability to future restrictions.
Strategic Significance for India
- India's immediate priority can be rapid and broad AI diffusion across sectors rather than attempting to replicate the entire frontier-AI ecosystem.
- However, dependence on freely available foreign models creates switching and supply-chain risks if geopolitical competition eventually restricts access.
- A balanced strategy therefore requires rapid adoption today alongside technological diversification and selective domestic capability-building for tomorrow.
