IndiaAI Mission to be reset as GPU prices double and committed capacity goes undelivered

The news

New Delhi and Bengaluru. The IT ministry plans to recalibrate the IndiaAI Mission because prices of GPUs, memory, storage and networking have “more than doubled” in a year, The Economic Times reports. A GPU (graphics processing unit) is the chip that trains and runs AI models. The mission has access to about 30,000 GPUs against a commitment of 45,000, as several firms could not deliver. Nvidia’s H200 has risen from about $20,000 to over $40,000; a server that cost about ₹2 crore now costs about ₹4 crore, with the rupee at ₹96-97 to the dollar against ₹88 earlier. Cyfuture, one supplier, offered about 1,100 GPUs and delivered about 160. C-DAC has floated a bid for about 3,000 GPUs to own outright, “a shift from the earlier strategy of not owning compute”, alongside longer-term contracts. This is GS3 on science and technology and its effects on development.

Static syllabus linkage

1. The IndiaAI Mission has seven pillars, with compute as the first. The Union Cabinet approved the IndiaAI Mission in March 2024 with an outlay of ₹10,371.92 crore, implemented by the IndiaAI Independent Business Division under the Digital India Corporation of the Ministry of Electronics and IT (MeitY). Its pillars are compute capacity of 10,000 or more GPUs through public-private partnership, an Innovation Centre for foundation models, a Datasets Platform, application development, FutureSkills, startup financing and Safe and Trusted AI.
2. How subsidised compute works. The government empanels private data-centre firms that offer GPU hours at bid-discovered rates, and startups and researchers rent them at a subsidised price. The Centre for Development of Advanced Computing (C-DAC), a MeitY scientific society known for the PARAM supercomputers, is the agency now buying GPUs directly.

Why UPSC loves this

1. GS3 lists “indigenization of technology and developing new technology”. AI compute tests that line: the best policy stalls if the hardware cannot be sourced.

Analysis

1. Lens — Market and State: renting compute was cheap until the market turned. The original design let private firms carry the cost and risk of buying chips, with the government paying only for use. Once prices doubled, suppliers locked into old rates had every reason to delay, and the State had no machines of its own. A sound judgement is a mixed model: own a core for public research and strategic needs, and rent the rest so taxpayers do not carry hardware that ages fast.
2. The rupee has become an AI policy variable. GPUs are priced in dollars, so the rupee’s fall from ₹88 to ₹96-97 alone raises costs by about a tenth. A fixed rupee outlay buys less compute each year, which is why the official speaks of “speeding up” to absorb phase-one funds.
3. Voluntary safety pledges and missing chips are two sides of AI governance. The Hindu reports that U.S. President Donald Trump and six tech chiefs signed a voluntary “White House Accord on Super Intelligence”, pledging internal monitoring, independent external evaluations and board-level oversight; experts say its worth depends on firms acting on findings that delay a launch. Frontier powers debate guarding AI while India struggles to obtain it. Unless India plans safety rules and compute together, it will import both the chips and the rules.

Administrator's brainstorm