LowV Technology Geopolitics & AI3 November 2026, Tuesday
India releases its first homegrown foundation models, sets 24-month GPU target
The IndiaAI Mission released 3 homegrown foundation models in New Delhi on 3 November, the largest being Soket Labs' 26.3-billion-parameter EKA-1. On the same day, S. Krishnan, secretary at the Ministry of Electronics and IT, said India aims to develop a domestic GPU within 24 months.
According to Analytics India Magazine, EKA-1 was trained on about 7.5 trillion tokens and supports 22 official Indian languages plus English. A foundation model is a large-scale language model that can be adapted to dozens of different tasks. Soket CEO Abhishek Upperwal said on 3 November that the model was designed for banking, finance and cybersecurity. ZenteiQ's Dhīr 2B and Dhīr 5B models work in English and 10 Indian languages and can run on a phone without a cloud connection.
The 3 models were developed on subsidised national compute infrastructure and will be made available on the AIKosh platform. The size of the subsidy, the number of GPUs used and the training cost had not been disclosed as of 4 November; these items could not be verified.
According to Crypto Briefing and TokenPost, Krishnan tied the domestic GPU goal to Semiconductor Mission 2.0, approved in July 2026 with about 1.27 lakh crore rupees (roughly 15 billion dollars) spread over 12 years. Of the 12 projects approved under the first mission, 5 have moved into commercial production. The government says its 2026 compute procurement will remain technology-neutral, meaning a domestic chip will have to compete with imported GPUs in tenders.
Talay assessment
Bottom line
India has delivered its first concrete output at the model layer, but the real bottleneck is hardware. No roadmap or interim milestones have been published for the 24-month domestic GPU target, and the technology-neutral procurement policy will force a domestic chip to compete with imported GPUs on price and performance. In the near term, India's AI compute will therefore remain dependent on imported hardware.
Likely effects
- Finance and cyber usePositive1–6 months
EKA-1, designed for banking and cybersecurity, offers a domestic option to public and financial institutions that require data localisation.
- GPU dependenceNegative6 months+
With a domestic GPU at least 24 months away, compute infrastructure stays reliant on imported accelerators and on supplier countries' export regimes.
- Semiconductor investmentPositive1–6 months
Mission 2.0's coverage of accelerator design could lift incentive applications from chip design firms.
Possibilities, ranked
- 1Target slips55%
Design work on a domestic GPU advances, but a commercial-scale product takes more than 24 months.
Watch: Approvals of accelerator design projects under Semiconductor Mission 2.0
- 2Prototype arrives on time30%
A public or private team demonstrates a working accelerator prototype within 24 months.
Watch: The IT ministry's GPU roadmap and announcement of interim targets
- 3Focus shifts to procurement15%
The government prioritises buying more imported GPUs over a domestic chip.
Watch: The number of GPUs in IndiaAI compute tenders
Probabilities are calibrated judgement based on the sources, not measurement, and are revised as new information arrives. Not investment advice.
Market reaction
Indicators affected
- EKA-1 parameters▲ 26.3 billion
- Domestic GPU target▲ 24 months
- ISM 1.0 in production▲ 5/12
Sources
- Analytics India Magazine — IndiaAI Mission Releases 3 Indigenous AI Models Developed by Soket, ZenteiQ (4 November 2026)
- Goodreturns — IndiaAI Mission releases EKA-1 and Dhīr models (3 November 2026)
- Crypto Briefing — India plans indigenous GPU chips within 24 months, says MeitY Secretary (3 November 2026)
- TokenPost — India Sets 24-Month Target to Develop Homegrown GPU Chips (4 November 2026)