Nvidia has been talking about sovereign AI for years, but is finding that India’s cultural and economic diversity calls for a different approach.

Unlike in the US, truckloads of GPUs won’t drive the chipmaker’s expansion in India. Instead, the company plans to focus on software first, and deal with computing power  later. It’s betting on a data-first approach, with a localized AI plan customized for the diversity of demographics, cultures, and languages. 

Nvidia is also wooing India’s huge developer base by touting its open-source credentials and devising a localized AI plan that includes smaller language models adapted for the local energy and data center infrastructure. 

“The data that gets created here, when assembled together in an Indian stack, creates population scale stacks that citizens can enjoy and solve many of the mega challenges India faces, be it in education, healthcare, mobility,” said Vishal Dhupar, Nvidia’s managing director in South Asia.

The Indian government has specifically pointed out how it consumes AI differently than countries like the US. For example, farmers don’t need access to large language models (LLMs), and can solve problems with targeted, smaller models that run on low-power chips.

Nvidia’s plan for India’s sovereign IT efforts includes expanding open models such as Nemotrol and development tools to Indian developers. “Open models can supercharge sovereign AI, helping our developers reflect their own language, tradition and culture in everything we do,” Dhupar said.

Nvidia last year made it possible for developers to write CUDA programs using Python, which is popular in India. The code indirectly helps sell more Nvidia GPUs, because  developers writing CUDA code need its graphics chips to execute AI loads.

The company’s Blackwell-class GPUs are being installed in India data centers, although  the nation’s infrastructure isn’t as mature as in the US. But India has room for a post-GPU AI infrastructure environment, especially given power and resource constraints, which is why it is already preparing for a world of low-power chips and edge processing.

India’s IT minister, Ashwini Vaishnaw, at last month’s World Economic Forum proposed jumping straight to lower power consumption chips that can run small language models. 

Vaishnaw’s rationale is that the smaller models will solve 95% of Indian users’ problems at a fraction of the cost. Moreover, the nation doesn’t want to fall prey to the AI bubble, in which the collapse of one overvalued AI company could hurt the economy.

Nvidia’s open-source Nemotron models have played a role in the development of some localized models, including the 17-billion-parameter BharatGen, which powers applications in public services, agriculture, security and cultural preservation, the company said.

Nvidia has also contributed to AI in India’s central digital payment system called United Payments Interface (UPI), which has been praised for its speed and efficiency. NPCI, which runs UPI, “is exploring training FiMi, a financial model for India, using the Nvidia Nemotron 3 Nano model and its own datasets,” Nvidia said in a statement.

Nvidia’s AI technology also has ties to the 8-billion parameter model Chariot, a multilingual communications platform, and Sarvam.ai, a multimodal AI platform for India-specific applications.

Beyond that, the company is working closely with India to help train the next generation of developers and elevate emerging startups. For example, it is collaborating with the Indian government’s Anusandhan National Research Foundation (ANRF) to boost cutting-edge AI research across the nation’s leading academic institutions. 

The company will offer ANRF grantee institutions complimentary access to Nvidia AI Enterprise software and specialized technical mentorship through the Nvidia AI Technology Center. “The collaboration will also include AI bootcamps, workshops and hackathons to strengthen India’s AI research ecosystem,” an Nvidia spokeswoman said.

Yotta is putting 20,000 Nvidia Blackwell Ultra GPUs in its Shakti cloud, which is inline with India’s sovereign AI infrastructure plans. Larsen & Toubro and E2E Networks also announced plans to create new data centers with Nvidia GPUs. 

Nvidia has talked about its sovereign AI plans in Europe, which includes establishing GPU data centers and partnerships with telecom, software and industrial companies. Gartner recently said that European investments in sovereign IaaS could reach $12.6 billion in 2026, growing from $6.9 billion in 2025.