India is racing to 100,000 public GPUs by December, its bet on sovereign AI compute

The scarce resource in modern AI is not ideas but compute — the specialised chips needed to train and run models. India is trying to make sure it has enough of its own, and to hand it to the people building things.

Under the IndiaAI Mission, the government has been expanding a pool of graphics processing units, or GPUs, offered at subsidised rates to startups, researchers and academics. Having deployed roughly 34,000 GPUs by mid-2026, it is adding tens of thousands more toward a target of 100,000 publicly available GPUs by December.

What The Plan Involves

The compute is the centre of it, but not the whole of it.

The mission, funded at over 10,000 crore rupees, treats AI compute as one of several pillars alongside foundation models, datasets, safety and skills. On the skills side, a fellowship programme has been expanded to support 13,500 scholars — undergraduates, postgraduates and PhD researchers — the human capacity meant to use the hardware being installed.

Why 'Sovereign' Is The Point

The word attached to this effort is sovereignty, and it has a concrete meaning.

Most cutting-edge AI compute worldwide is rented from a small number of large cloud providers, nearly all American. For a country that wants its own AI industry, depending entirely on foreign infrastructure is a strategic vulnerability: prices, access and terms are set elsewhere, and can change. A subsidised domestic pool is an attempt to give Indian developers a base of compute that does not run through another country's companies.

The Case It Is The Right Bet

There is a strong argument for this approach, particularly for India.

India has the developers and the researchers but has lacked affordable access to serious compute, which has pushed its AI talent toward foreign platforms or foreign employers. Subsidised GPUs lower that barrier directly, and doing it through public infrastructure rather than leaving it to the market is a legitimate choice for a country that wants the capability broadly available, not concentrated in whoever can pay. The fellowship funding shows awareness that hardware without trained people is wasted.

The Hard Questions

Fairness requires the doubts too, and they are practical rather than ideological.

A GPU count is an input, not an outcome. What matters is whether the compute is well utilised — whether the chips are the right kind, kept busy, and actually reachable by the startups they are meant for, rather than tied up in bureaucracy. India also does not yet make these chips; the hardware is imported, so 'sovereign compute' still rests on foreign silicon, even if the pool is domestically run. And 100,000 GPUs, while substantial, is modest against the clusters the largest US and Chinese labs command.

The target is a means, in other words, and the measure of success is what gets built on it, not the number itself.

What To Watch

Whether the 100,000-GPU target is actually hit by December, and how much of the capacity is genuinely used. Whether Indian startups and researchers report that the compute is accessible in practice, not just announced. And whether this feeds real Indian foundation models and products, which is the outcome that would justify the spending.