US AI export curbs spark concerns in India, experts flag long-term challenges

The controls are set to take effect in 120 days. Donald Trump takes over as the president on January 20.

For industry veterans such as iSPIRT Foundation co-founder Sharad Sharma and HCL co-founder and EPIC Foundation chairman Ajai Chowdhury the move is a throwback to days of licensing, when restrictions were imposed on key technologies such as cryogenic engines and nuclear technology.

Sunil Gupta, the CEO of Yotta Data Services, a firm which sets up Tier 2 and 3 data centres, said," In the long term, India must diversify suppliers and avoid dependency on a single country."

A couple of months ago, Yotta became the first company in India to acquire Nvidia's H100 GPUs.

It is also one of the companies bidding for the procurement of 10,000 GPUs under the Rs 10,738 crore IndiaAI Mission. The industry does not anticipate any impact on the procurement process because of this policy.

'A clear design to control AI'

"This move by the US government clearly indicates their desire to control the creation of AI models. The US will be happy for others to use their foundational models. This has significant economic and national security implications," Sharad Sharma told Moneycontrol.

Sharma, also a member of a United Nations advisory body on AI, said India has to deal with this situation much the way it dealt with controls on cryogenic engines, nuclear technology and so on. If the incoming Trump administration does not revisit the decision, AI will join this list of controlled technologies on which restrictions have been imposed by the US, he said.

No short-term impact on IndiAI Mission 

Multiple industry experts said there is likely to be no short-term impact on India's AI ambitions.

The government's ambitious Rs 10,738-crore IndiaAI Mission is well underway with several companies in the fray for the procurement of 10,000 GPUs.

Ashok Chandak, president of the India Electronics and Semiconductor Association (IESA) said, The IndiaAI Mission aims to develop infrastructure with over 10,000 GPUs through public-private partnerships, supported by a Rs 10,000 crore investment over five years.

Also read: In farewell address, Biden warns an 'oligarchy' of ultra-wealthy is 'taking shape in America'

“In the short term, the new export controls may not significantly impact India. However, the uncertainty of securing licences and trade negotiations could challenge India's ambitions for large-scale AI hardware deployment," he said

Gupta said the next two years may not see a major impact since the market is still developing. However, any form of control will hamper development in the long term.

"India was not even at 1,000 GPUs a year ago. Today, we are talking about 4,000-5,000 GPUs, but this is still far from what India aspires to achieve... US' move, in the short term, may not have much impact, but long term, India will need lakhs of GPUs. Once commercial production use cases are scaled to millions of citizens, the pinch will be felt," he said.

Long-term implications

The industry foresees several hurdles for India, which include restrictions in accessing advanced AI chips. The licensing requirements could also raise costs and lead to delays. Indian companies may also have to heavily rely on global corporations for AI infrastructure.

"Large-scale AI data centres, requiring several hundred thousand GPUs, may be delayed or scaled down, putting global companies at a competitive advantage over Indian enterprises. However, small-scale setups could still enable experimentation, innovation and restricted model development," Chandak of IESA said.

India may not produce multiple large language models and instead focus on fine-tuning and inferencing purposes if this policy starts biting, Yotta's Gupta said.

The way forward

Experts want the government to bet big on its datasets and leverage them with the US to get better access to GPUs. India should also design its chips with indigenous technologies such as RISCV.

Reduced instruction set computer five (RISCV) is an open source architecture developed by IIT-Madras for chip operations.

Also read: Biden is packing up to leave White House. But what about any classified documents?

Diversify: Gupta said, "The government is quietly pushing for technological sovereignty with initiatives like Make in India and semiconductor subsidies. Self-dependence starts from the chip level — designing and manufacturing chips — and moves up to data centres. However, in the short term, every country is dependent on someone.

"The US policy seems targeted at high-performance chips like NVIDIA’s GB200 series, which are crucial for training large language models. Inferencing use cases may rely on mid-level or lower-end chips, reducing immediate impact.

“In the long term, India must diversify suppliers and avoid dependency on a single country."

While things may change in the Trump administration, Neil Shah, Counterpoint Research if the policy were to remain like this, India would have to prove herself an ally to the US to ensure access to this critical technology.

"From the outset, it looks like the time it takes for a company to go to market would extend as a result of this. For instance, data centres would be dependent on US-produced compute chips. However, things may look different with the Trump administration coming in post-20th. But, if it were to remain like this, India would have to prove herself an ally to the US to ensure access to this critical technology," he added.

Continental datasets: iSPIRT's Sharma said, "Going forward, India needs to marshal its continental size datasets and leverage them to negotiate better access to GPUs. Just like all emails flow through a protocol (SMTP), India is rolling out a techno-legal system where all sharing of personal data happens through a DPI called DEPA (Data Empowerment and Protection Architecture). This has to be accelerated.

"If we don't take these steps, India will remain vulnerable and will not achieve strategic autonomy on AI in critical sectors like defence and healthcare. Much of the gains in avoiding digital colonization through India Stack will be lost."

Open up DLI: Chowdhry said, "In semiconductor chips, we have created our own RISCV technology in IITM for which we don’t need to keep paying licence fees. Using this we should design our chips in RISCV to secure ourselves from future sanctions.

"This should be taken up in a mission mode with funds provided under ISM’s (India Semiconductor Mission) new DLI (Design Linked Incentive) policy. And open up DLI for all Indian companies, including corporates/ER&D exporters, MSMEs and startups, by increasing the size of the funding from the small Rs 30 crore to a generous Rs 50 crore to 150 crore. Maybe we can create our own NVIDIA and AMD in the next ten years."