India buys accelerators as a public service
On 7 March 2024 India's Union Cabinet approved the national IndiaAI mission with an outlay of Rs 10,371.92 crore. Its load-bearing part is a pool of 10,000 or more graphics processing units built through public-private partnership, access to which the state subsidises for startups, researchers and universities.
Why it matters
States had bought compute for their own research centres before. This is different: the pool is built not for an institution but as a market service, admitting other people's companies at a reduced price. In a country where an accelerator costs more than an engineer, that is the one obstacle a budget can actually remove.
The mission is implemented by the IndiaAI Independent Business Division within Digital India Corporation. There are seven pillars: compute capacity; an innovation centre meant to build indigenous large multimodal and domain-specific foundation models; a platform of non-personal datasets; application development against problems sourced from central ministries and state departments; FutureSkills, with data and AI labs in tier 2 and tier 3 cities; startup financing; and safe and trusted AI. What the record does not claim. The report's claim of models for India's 22 official languages is not on the release page, which speaks of large multimodal and domain-specific foundation models without listing languages. The record gives no dollar equivalent because the source gives none. This is a decision about a budget rather than about a result: how many accelerators were actually bought, and what was trained on them, is not established here. The report cited Press Information Bureau release 2012234. That is a release of the same day about a terrorist designation; the one needed is 2012357.