US restrictions on exports of AI chips likely to hurt India by 2027

The Joe Biden administration’s interim final rule to impose restrictions on the export of GPUs (graphics processing units) critical for powering AI could hinder India’s ambitious plans to scale up its AI programme by 2027.

 

“While the short-term impact might be minimal, restricted access to GPUs will have an adverse impact from 2027 as India plans to scale up its AI play. We think it is a more or less done deal,” remarked Ashok Chandak, president of the India Electronics and Semiconductors Association (IESA).

 

Echoing this sentiment, Bhaskar Chakravorti, dean of the Fletcher School of Business, noted: “In the initial stages, India will need around 25,000 GPUs (the government is currently empanelling companies to offer 10,000 GPUs of computing power needed for AI). For now, India’s GPU requirements will be met, but the demand will grow exponentially, and any kind of constraint is problematic.”

 

The Ministry of Electronics and Information Technology (Meity) has stated that it is examining the potential impact of the proposed order while awaiting further clarity on policies under Donald Trump, who is set to assume the presidency in the US next week. A top ministry official conceded: “There might be some impact, surely.”

 

Experts in the US attribute India’s exclusion from the “friendly” list to its perceived closeness to Russia and doubts about its ability to control and manage technology as robustly as the 18 countries on the list. However, many anticipate that India’s bilateral strategic partnership with the US in critical and emerging technologies will help it negotiate the removal of the curbs.

 

“Trump might use this as a negotiating tool to extract tariff concessions or ease qualitative restrictions like the PC import regime,” said a senior executive at an Indian IT company.

 

Under the proposed rules, 18 countries -- including close US allies such as the UK, Canada, Germany, Japan, and South Korea -- will enjoy unrestricted access to chips for verified companies. Meanwhile, embargoed countries such as China and Russia will be entirely banned from procuring GPUs. India is in a third category of “non-embargoed” countries, alongside Israel and Singapore, which will be subject to licensing requirements. These countries will be allowed to import up to 1,700 GPUs annually without restrictions. Beyond this, companies must secure a “national validated end-user authorization” from the US government, with annual caps set at 100,000 units in 2025, rising to 270,000 in 2026 and 320,000 in 2027. The computing power of these chips is benchmarked at Nvidia’s H-100 GPU equivalent across all years.

 

Another contentious issue is the establishment of AI-focused data centers. Under the rules, US companies like Google, Microsoft, and Meta will receive global authorisation for GPU usage but must deploy 50 per cent of their capacity in the US, 25 per cent in the 18 allied countries, and the remaining 25 per cent in “non-embargoed” countries like India. However, a cap limits deployment to no more than 7 per cent of their capacity in any single country.

 

Chakravorti warned that these restrictions could deter US tech companies from establishing AI-led data centers in India. “Microsoft and Google’s needs will grow exponentially, so they will think twice before investing in data centers in India if they worry about hitting a GPU constraint.”

 

Chandok added: “US companies will have an advantage in setting up AI data infrastructure compared to Indian players, creating a non-level playing field. After all, Indian players will face restrictions on the number of GPUs, while US companies won’t.”