The IndiaAI Mission was approved by the Union Cabinet back in March 2024, with a ₹10,371.92 crore outlay over five years across seven pillars: compute, foundation models, datasets, application development, AI safety, startup support and skills. 2026 is the year several of those pillars produced something concrete rather than just funding announcements.
Compute: the GPU build-out
The Mission's compute pillar targets 100,000 publicly accessible GPUs by December 2026, up from roughly 34,000 at the start of the programme, backed by a $1.25 billion government commitment to sovereign AI infrastructure. That's the unglamorous but genuinely load-bearing part of "sovereign AI" — without in-country compute, the rest of the mission's ambitions (India-trained foundation models, an AI safety institute) don't have anywhere to run.
The India AI Impact Summit and the New Delhi Declaration
The India AI Impact Summit ran 16-21 February 2026 at Bharat Mandapam, New Delhi — the first major global AI summit hosted in the Global South. It produced the New Delhi Declaration on AI Impact, endorsed by 88 countries and international organisations, centred on strengthening AI infrastructure, investing in local talent, and protecting citizens from AI-related risks. Separately, the summit was also where India's AI Governance Guidelines were unveiled (see our companion comparison post).
Three indigenous AI models were unveiled at the summit under the sovereign-AI umbrella — the clearest public evidence yet that the Mission's foundation-model pillar is producing shipped models, not just funding rounds. This is also the space where catalogue entries like Sarvam AI and Krutrim already sit: India-hosted, India-tuned models built specifically for the country's language and data-residency needs.
What's still just a target
The 100,000-GPU figure is a December 2026 target, not a delivered number as of this post — worth checking the Ministry of Electronics and IT's own updates for the current count rather than assuming the target has already been hit. Similarly, "seven pillars" is the Mission's structure on paper; how much each pillar has actually disbursed and shipped varies pillar to pillar, and skills/dataset progress is far less publicly visible than the compute and summit headlines.