Between 3 and 8 November 2026, MeitY released three indigenous foundation models, its secretary described a five-layer AI stack with a 24-month target for an Indian GPU, and the minister called India the world's future AI application capital. In the same week, Uttar Pradesh approved an AI Mission with five lakh AI-fluent youth and a procurement scheme that lets startups sell to government. This post argues that sovereignty is not a chip announcement but a first customer, and proposes a national 2-5% startup procurement set-aside with published, machine-readable demand lists.
On 3 November MeitY released three indigenous foundation models: Soket Labs' EKA-1, a 26.3-billion-parameter model covering all 22 scheduled Indian languages, and two open Dhir models from ZenteiQ that run offline in English and 10 Indian languages (The Researchers, 7 November 2026). Most readers filed that under "models" and moved on.
Two days later, MeitY Secretary S. Krishnan put a roughly 24-month development target on an indigenous GPU and described the stack in five layers: power and energy, chips and accelerators, servers and compute, models, applications (CIO, 4 November 2026) — all of it sitting under Semicon 2.0 and its Rs 1,27,500 crore outlay. On 6 November the minister opened applications under three more pillars of that programme and said India is "very focused on making India the application capital of the world" (Mint, 6 November 2026).
Then the item I would circle. Writing in the Times of India on 8 November, UP's IT principal secretary Alok Kumar and NITI Aayog's Debjani Ghosh set out the state's AI Mission: five lakh young people made AI fluent, 50,000 "forward-deployed" AI professionals, deep-tech U-Hubs in Noida and Lucknow, and — more consequentially — a Startup Procurement Scheme 2026 that opens "pathways for eligible startups to sell innovative solutions directly to government, including higher procurement limits for sunrise sectors such as AI and deep tech" (Times of India, 8 November 2026).
They named the real constraint in the same paragraph: finding the first customer.
In Fuel India's AI mission (26 June 2025) I welcomed the other half of this — ten firms supplying 14,000 GPUs, with up to 40% of compute costs underwritten. Generous supply-side policy. But a subsidised GPU hour and a paid first order are different instruments, and India has so far built only one of them.
Here is what I would carry from Lucknow to Delhi. This is my suggestion, not existing policy:
- A default 2-5% set-aside for startups in every AI and IT procurement by central ministries, states and public sector units, each with a published problem statement attached.
- A published, machine-readable demand list, refreshed quarterly: department, problem, budget band, deadline, output format. A builder deciding where to spend his next six months of GPU time should be able to read it and choose.
- An interoperability clause — anything bought with public money exposes open APIs, so a district office can change models without rebuilding its workflow.
The third point is the one I would fight for. In No Claims : Just Comparison (26 April 2025) I set IndiaAGI's multi-model round table against the single-LLM answer, and argued that sovereignty rests on locally curated data rather than on one vendor. I still believe it. India needs a federation of small, cooperative, Indian-language models — which is roughly the direction the offline Dhir releases point to — and not a national monolith that turns every government buyer into a captive customer.
One actor is being left out of the count. AKTU in Lucknow has just been approved a Rs 20 crore AI Centre of Excellence — Rs 8 crore from MeitY, Rs 8 crore from the state, Rs 4 crore from IBM — reaching three lakh students across 750-plus affiliated colleges (Roori, 7 November 2026). Those students are the vendors of 2028. They need a buyer list in 2026.
Sovereignty is not a chip announcement. It is a purchase order. A chip we build and nobody at home buys stays a laboratory curiosity; a startup with a government order in hand becomes a company. Implementation will take years, and I may not see the end of it — but the first customer can be found next quarter.
With regards,
Hemen Parekh
www.Iam-Immortal.ai / www.IndiaAGI.ai / www.YourContentCreator.in
8 November 2026
No comments:
Post a Comment