How India should view China’s ‘open’ AI pitch

Why in the News

China has offered to lead the creation of a BRICS open source artificial intelligence (AI) community, along with a BRICS digital ecosystem cloud platform, support for cooperation on large language models and a programme of AI training. The offer was made by the Chinese President at the BRICS Summit in New Delhi. The New Delhi Declaration issued after the summit mentions neither the community nor the cloud platform, and commits members instead to broader cooperation on improving access to AI resources. The pitch positions Chinese AI technology as an alternative to proprietary systems controlled largely by companies in the United States. For India the question is whether a grouping wide platform led by Beijing widens access to AI for developing countries or routes that access through a single supplier.

What is the proposed BRICS AI open source community?

1. China in the lead: China would take the lead in setting up the community.
2. Model cooperation: It would support cooperation among members on developing and deploying large language models (LLMs), systems trained on very large text collections to generate and interpret language.
3. Training and seminars: It would run specialised AI seminars and training courses, described as building an open AI ecosystem.
4. Cloud platform and adjacent areas: A BRICS digital ecosystem cloud platform was proposed alongside it, with expanded cooperation on digital skills, technology exchanges and intelligent manufacturing.

Why is China making this pitch to developing countries now?

1. An alternative to proprietary systems: The initiative widens Beijing’s effort to position its AI technology against systems controlled largely by companies in the United States.
2. Commitments already made: At the World Artificial Intelligence Conference in Shanghai in July, 5,000 AI training and seminar opportunities for developing countries over five years were announced.
3. Cooperation centres: AI application cooperation centres were proposed with groupings including BRICS, ASEAN and the African Union.
4. A contest for the Global South: Both India and China aspire to be the leading voice of the Global South, and Beijing holds a clear edge in AI capabilities.

What is open source artificial intelligence?

1. Open weights and code: A model released under a licence that lets others run, modify and redistribute it.
2. Contrast with a proprietary system: A proprietary model’s weights stay with the vendor and are reached only through an interface the vendor controls and prices.
3. Why it bears on access: A released model can be run on a user’s own hardware, which removes the need to buy access from the developer for every use.
4. Limits of the label: Openness of weights does not always extend to the training data or to the terms on which the model may be used commercially.

Why was the proposal not adopted by the grouping?

1. The declaration is silent: The New Delhi Declaration does not mention the proposed open source community or the cloud platform.
2. What it commits to instead: Members are committed more broadly to cooperation on improving access to AI resources, with a focus on safety, security, reliability and inclusiveness.
3. Existing text carried forward: The declaration refers to an earlier BRICS statement on global AI governance and records that members will continue cooperation in the area.
4. The proposal can return: China takes over the BRICS chairship in 2027 and could place the proposals before the grouping again.

What is India’s own position on access to AI?

1. The access demand: At the AI Impact Summit earlier this year India pushed for broader access to compute, datasets, models and other AI infrastructure, particularly for developing countries.
2. Domestic capacity: The IndiaAI Mission funds subsidised compute infrastructure and supports Indian foundation models and datasets.
3. The two run alongside each other: Any eventual BRICS programme on models or cloud infrastructure would sit next to India’s own effort to expand access without relying entirely on foreign providers.

Challenges to a BRICS platform for open source AI

1. Compute is the binding constraint, not model access: Releasing model weights does not give a developing country the accelerators or the electricity to train or serve them at scale. Eg. Advanced AI accelerators are subject to United States export controls that reach third countries.
   The Fix: Pair any model sharing commitment with pooled access to compute capacity physically located in member countries.
2. Dependence on one member’s technology stack: A cloud platform built and operated by a single member leaves participants dependent on that member’s chips, software and terms of service. Eg. Huawei’s Ascend accelerators and their accompanying software stack underpin much of China’s domestic AI infrastructure.
   The Fix: Require any BRICS platform to expose hardware neutral interfaces, so a workload can be moved to another member’s infrastructure.
3. Divergent data governance among members: Members differ on cross border data transfer and on state access to data, which blocks a shared dataset pool. Eg. India’s Digital Personal Data Protection Act, 2023 sets its own regime for transfers outside the country.
   The Fix: Begin with model and training cooperation and leave datasets to bilateral arrangements until a common transfer standard exists.
4. Language and content coverage: A model released by any one member carries that member’s language priorities, so coverage of other members’ languages stays thin. Eg. Indian language performance in globally released models lags their performance in English.
   The Fix: Make a language corpus contribution from each member a condition of participation in the community.
5. Safety obligations left unattached to release: An open release removes the developer’s ability to withdraw a model later found unsafe, because copies already exist. Eg. Once weights are downloaded and mirrored, a subsequent restriction cannot reach the copies in circulation.
   The Fix: Attach an evaluation and disclosure requirement at the point of release rather than relying on a recall mechanism afterwards.

Conclusion

Access to AI is being contested as a question of who supplies it, not of whether it should be shared. An offer to open the models while owning the platform beneath them widens use without widening capability, and that is the distinction India has to hold on to. What to watch is whether the grouping’s next chair converts the access language already agreed into a commitment on compute, or leaves it as a statement of intent.

Back2Basics: IndiaAI Mission

1. A national mission under the Ministry of Electronics and Information Technology, approved in 2024.
2. Built around seven pillars, including IndiaAI Compute Capacity, the IndiaAI Innovation Centre and the IndiaAI Datasets Platform.
3. Its compute pillar subsidises access to graphics processing units for startups, researchers and public institutions.
4. Its remaining pillars cover application development, skilling, startup financing and safe and trusted AI.

Matching Previous Year Question

“[2026, GS2, 10 marks] “BRICS acts as a powerful counterweight in global governance, actively amplifying the voice and influence of the Global South.” Explain the role of BRICS in projecting itself as an alternative to other groupings.”