The government is recalibrating the IndiaAI Mission as GPU shortages and rising costs delay the project. Since the March 2024 launch, prices have doubled, leaving the mission with 30,000 of the promised 45,000 GPUs. A senior official said, "We are constantly talking to them, and we are trying to increase."

The government is planning to recalibrate the IndiaAI Mission as the rollout of critical compute components, particularly GPUs, faces delays and mounting cost pressures. That follows a sharp rise in memory and GPU prices since the mission was launched in March 2024. In most instances, prices have more than doubled over the past year.

The IndiaAI Mission currently has access to about 30,000 GPUs against a total commitment of 45,000. Several firms that had committed capacity have been unable to deliver due to market pressures.

The IT ministry is now planning to issue fresh bids while also pressing companies to meet their pledges. "Compute across the world is constrained. We are pushing hard to see what we can give to our companies," a senior government official told ET. "We are constantly talking to them, and we are trying to increase the overall level of supply." The ministry also plans to buy 3,000 GPUs.

The IT ministry will do this through the Centre for Development of Advanced Computing (C-DAC). The capacity will be used by the government and be made available to startups and researchers, and is a shift from the earlier strategy of not owning compute.

Companies said that the sharp increase in prices has disrupted procurement plans under the Rs 10,372 crore mission. For instance, Nvidia's H200 has risen from about $20,000 per GPU to more than $40,000 now. Similarly, a server that cost about Rs 2 crore a year ago now costs around Rs 4 crore.

"We are speeding up what we need to do and absorb the funds which we have been allocated under the IndiaAI Mission phase one," a senior government official told ET.

The official said the government was working on multiple tracks, including building its own compute capacity, contracting capacity for the longer term and organising a framework to procure AI services to ensure both the government as well as India's startups and research organisations have access to compute.

"We have some agreed rates, and those agreed rates are up to a point," the official said. "Now we are actually trying to build our own compute, and also on a longer-term basis, contract compute. So for that, we are doing some more bids."

The official said the government had access to "closer to about 30,000 GPUs." CDAC has floated a bid for about 3,000 GPUs, which will be installed as new capacity. Some of it will be allocated for government use and the rest for startups and research.

IndiaAI Mission did not respond to queries.

Cyfuture chief executive Anuj Bairathi, NxtGen Datacenter & Cloud Technologies chief executive AS Rajagopal and an executive of the Appsquadz Software-Vensysco consortium-participants in the mission-said there has been a sharp increase in the cost of GPUs, memory, storage, networking equipment and other components. Currency movements, shortages of critical server parts and limited visibility on future demand have added to the pressure.

Cyfuture's Bairathi said the cost of a complete server had nearly doubled in the past year. A server typically includes GPUs besides memory and storage, he said. Cyfuture had offered around 1,100 GPUs under an earlier empanelment round and delivered about 160 Nvidia H100 GPUs, Bairathi said. In a subsequent round, it committed to 1,024 Nvidia B200 GPUs but has struggled to procure them amid supply constraints.

(This story has not been edited by economictimes.com and is auto-generated from a syndicated feed we subscribe to.)