Discuss how reliance on foreign open-source AI models can create both opportunities and strategic vulnerabilities for developing economies like India.

Open-weight models — where parameters are released but training data and code are not — from Chinese labs such as DeepSeek, Qwen and Kimi have become the cheapest route to advanced AI for capital-scarce economies. For India they are a real opportunity, but one held on another state's terms.

Opportunities

• Cost compression: lower training costs achieved through distillation and architectural efficiencies [1] pass downstream, letting Indian startups build and serve AI applications at a fraction of proprietary-model costs.
• Bypassing the compute floor: India's IndiaAI common compute pool crossed 34,000 GPUs, offered at subsidised rates [2] — adequate for fine-tuning, not for frontier pre-training. Open weights let firms skip pre-training and compete at the application layer.
• Avoiding single-vendor lock-in: a multi-ecosystem approach keeps India from being captured by either the American or the Chinese stack [1].
• Localisation: released weights can be adapted to Indian languages and sectors, complementing IndiaAI's sovereign foundation model effort [3].

Strategic vulnerabilities

• Access is revocable, not guaranteed: openness is statecraft — aimed at undercutting American labs' proprietary revenue and driving global adoption of Chinese full-stack cloud infrastructure [1]. What is given for adoption can later be graduated or withheld.
• Savings are not capability transfer: weights arrive without training corpora or curation code, so the gain accrues at inference, building no domestic pre-training capacity that would survive a withdrawal.
• Model-integrity risk: alignment and framing behaviour are encoded in the parameters; local hosting relocates data, not the value judgements baked in at training time.
• Fragile dependencies: the Economic Survey cautions that technology choices must reinforce long-term growth rather than create fragile dependencies [4].

Open weights are best treated as a window, not a foundation. India should use the cost relief to fund efficiency research, Indian-language datasets and portable domestic models, while archiving current checkpoints as insurance. Pursued through the Economic Survey's decentralised, application-driven path [4] and IndiaAI's mandate [3], a borrowed advantage can mature into durable technological self-reliance.