India to Release a Draft AI Regulation Paper Within 30 Days, and Targets Indigenous GPUs in Three to Four Years

India will publish a draft consultation paper on regulating artificial intelligence within 30 days, Electronics and IT Minister Ashwini Vaishnaw said on 8 October 2026. He also said the country is working towards indigenous GPUs, particularly for AI inference, over the next three to four years.

Vaishnaw spoke in New Delhi at the India launch of the World Bank Group's World Development Report 2026, which this year is devoted to artificial intelligence. The launch was organised with the IndiaAI Mission and the Ministry of Electronics and Information Technology (MeitY).

What the paper is expected to cover

According to reports of the minister's remarks, the consultation paper will focus on:

• AI safety and human-centric development
• Deepfakes and synthetic media
• Psychological harms from AI systems
• Skilling and inclusive access, alongside emerging risks such as cybersecurity

Vaishnaw said industry will carry the bulk of responsibility for managing these risks. He described the government's approach as "techno-legal": using technical tools and standards together with law, instead of relying only on new legislation.

The government has not yet said what specific obligations companies may face. Until the draft is published, details such as risk tiers, compliance timelines and penalties are unknown. Some reports have speculated about them, but those claims have not been confirmed.

Compute: GPUs today, Indian GPUs tomorrow

The second half of Vaishnaw's message was about hardware. Reports of the event say the government estimates demand for about 10,000 GPUs, with an initial tender for 5,000. To make supplying them commercially attractive, the government will guarantee GPU usage for up to five years, in phases, so bidders can be sure the hardware they bring onto the IndiaAI compute platform will be used.

In the longer term, India wants its own chips. Vaishnaw said work is under way towards domestic GPU capability, with inference as the first focus, over the next three to four years. Inference is where a trained model is run to answer queries. It is a more realistic first target than training-class accelerators, which need cutting-edge manufacturing processes and huge memory bandwidth.

Why it matters

India has so far avoided a single, comprehensive AI law of the kind the European Union passed with its AI Act. It has relied on existing IT rules, advisories and sector-specific guidance. A formal consultation paper is the first clear sign of a dedicated framework, and the choice of deepfakes and psychological harm as headline concerns reflects problems Indian users already face, from election-season synthetic videos to AI-driven scams.

The "techno-legal" framing also suggests India will lean on tools it can build itself, such as watermarking, detection systems, and the AI safety work being done under the IndiaAI Mission, rather than copying a rulebook from elsewhere.

On compute, the stakes are just as high. The IndiaAI Mission has been subsidising GPU access for startups and researchers building foundation models. At the India AI Impact Summit in February, Sarvam AI released 30-billion and 105-billion-parameter models, and the government-backed BharatGen released Param2, a 17-billion-parameter model supporting 22 Indian languages. Each of these needs sustained access to accelerators, almost all of which come from a handful of foreign suppliers. An Indian inference chip would not end that dependence, but it would cover a growing share of everyday AI workloads.

The World Bank backdrop

That the announcements came at the launch of the World Bank's flagship report on AI was fitting. The report looks at how AI can drive development in emerging economies, an argument India has made repeatedly as it positions its approach — open digital public infrastructure, multilingual models, subsidised compute — as a template for the Global South.

What to watch

The real test is the draft itself. Developers, startups and civil-society groups will look at how risk is defined, whether open-source models get separate treatment, how responsibility is split between model builders and app developers, and how the rules fit with the Digital Personal Data Protection Act. If MeitY keeps to its timeline, the paper should be public by early November.

Sources

• https://www.business-standard.com/technology/artificial-intelligence/govt-to-release-draft-ai-regulation-paper-within-30-days-ashwini-vaishnaw-126100801183_1.html
• https://www.outlookbusiness.com/deeptech/govt-to-release-ai-regulation-consultation-paper-in-a-month-eyes-indigenous-gpus-in-3-4-years-vaishnaw
• https://www.dqindia.com/news/india-to-release-ai-regulation-consultation-paper-in-30-days-vaishnaw-12660030
• https://www.storyboard18.com/digital/india-to-release-ai-regulation-consultation-paper-within-a-month-union-minister-ashwini-vaishnaw-112199.htm
• https://www.adgully.com/post/21972/centre-to-release-ai-regulation-consultation-paper-within-a-month-confirms-ashwini-vaishnaw
• https://pragativadi.com/wdr-2026-artificial-intelligence-india-launch-insights/
• https://techobserver.in/news/egov/indiaai-mission-sovereign-ai-models-indian-languages-321425/