US AI Chip Export Controls: Implications for India's AI Ambitions

AI Chips and Geopolitics: India's Concerns Over US Curbs

Recently, the United States imposed restrictions on AI chip exports, and that has created a wave of concerns all over the globe, not an exception for India. Advanced semiconductors, popularly known as AI chips, are critical for processing complex algorithms and large data sets. This is needed to significantly advance in machine learning, , and natural language processing. Without the best AI chips, countries will fall behind in many technology-focused industries, like health care and defense. These chips matter so much in politics because they could change the global balance of power. Control over access to AI chips has become a strategic tool for countries wishing to maintain their technological edge. In this regard, the United States has tightened its export controls, especially against , to protect its technological hegemony.

Proposed US restriction on AI chip export threatens India's AI hardware plans: IESA

The US administration's new framework restricts the import of artificial intelligence chips citing national security concerns about the technology and economic interests of producers and other countries. The 18 key allies of the US, however, has no restriction for which they have been clubbed under Group 1, but there are restrictions on the quantum that can be exported to other countries under Group 2, including India.The proposed restriction on artificial intelligence chip export to non-key allies nations, including India, will challenge the country's plan for AI hardware, crucial for the local development of emerging technologies, semiconductor industry body IESA said on Wednesday. India's National AI Mission aims to develop infrastructure with over 10,000 GPUs (graphics processing units) through public-private partnerships, supported by a Rs 10,000 crore investment over five years. The US administration has proposed a new framework that restricts the import of artificial intelligence chips due to national security concerns about the technology and economic interests of producers and other countries. The proposal has no restriction for 18 key allies of the US, which have been clubbed under Group 1, but there are restrictions on the quantum that can be exported to other countries under Group 2, including India. "In the short term, the new export controls may not significantly impact India. However, the uncertainty of securing licenses and trade negotiations could challenge India's ambitions for large-scale AI hardware deployment," India Electronics and Semiconductor Association (IESA) President Ashok Chandak said in a statement. The proposal restricts export to less than 1,700 GPUs per company per year in Group 2 countries. "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 said. India could potentially secure General National Validated End User (NVEU) authorisations due to its status as neither a re-exporter of Compute ICs (integrated circuits) nor an advanced compute manufacturing hub, he added. "This distinction, combined with the presence of significant Indian design centres for GPU makers like NVIDIA and AMD and their management commitment of support to India, could position India favourably for license approvals," Chandak said. US-based providers like Amazon, Microsoft, Meta, etc. are expected to receive global authorisations but will be limited to deploying only 50 per cent of their AI computing power outside the US under the proposed framework. Chandak said the export controls are set to take effect in 120 days, allowing the incoming administration under President-elect Donald Trump to potentially amend these rules. "Thus, it is a bit uncertain whether the Trump administration will make it easy or pass the rule as is. Amid growing concerns from technology industries, the global AI landscape may witness a shift, impacting both US technological leadership and India's growth trajectory in AI infrastructure. "Though, in the short term, India may not have a major impact, but in the long run, scaling up by any Indian conglomerate could face the hurdles of quantity cap," he said. ARTPARK Co-founder and AI Foundry Chairman Umakant Soni, however, said India stands to benefit greatly from the geopolitical realignment driven by these restrictions, as countries and corporations like Apple and Samsung look to diversify their supply chains away from China. "India already has a strong base in chip design and engineering talent (companies like Intel, NVIDIA, and Qualcomm have R&D centres in India). The restrictions could also boost India's chip design and research ecosystem. However, India will need to ramp up investments in training its workforce in semiconductor manufacturing and design to fully capitalise on the opportunity," Soni said. The proposed framework completely bars exports of AI Chips to countries in group 3 comprising China, Russia, Iran, Iraq, Cambodia, Belarus etc.

India to take up Biden's AI chip export curbs with Trump govt

India will raise the issue of the US imposing a country-specific cap on exports of AI compute and foundational models with the incoming Donald Trump administration, a senior official told ET. The Ministry of Electronics and Information Technology (MeitY) has held "initial discussions" with various arms of the government, including national security advisor (NSA) Ajit Doval's office, and has shared analysis and concerns pertaining to the move by the outgoing Joe Biden US administration. "We really have to take it up with the new government once it comes in," the official said. The cap could be detrimental to India's tech industry in the long term even if it's not an immediate crisis for the country, the person added. The government will raise the issues under the India-US Initiative on Critical and Emerging Technologies (iCET) framework, the official said. The framework, launched by Prime Minister Narendra Modi and President Biden on the sidelines of the Quad Summit in Tokyo in May 2022, focuses on collaboration in critical and emerging technologies. Earlier this month, US national security adviser Jake Sullivan was in India and also met Doval. The two sides regularly discuss cooperation on areas including artificial intelligence, quantum computing, semiconductors, defence and space. "We have an iCET framework; it is led by the NSA. So, we'll take it up through that channel," the official cited above said. "MeitY has done an assessment, NSA has done an assessment...everyone is assessing what the implications are, what it implies, and so on. It's not an immediate crisis, we should be able to manage. But, as we go along, it's more important to (raise this)." As per the new US directive, India won't be able to import more than 50,000 GPUs from the US in a year and even at the country level India will need a licence. The government is currently sourcing close to 10,000 GPUs under the India AI Mission. The number may go up in the future. ET had reported on January 15 that the rule may increase compliance complexities for companies like Reliance Industries, which recently placed an order for Nvidia GPUs. It could put data centre providers such as Tata Communications, Yotta Data Services, E2E Networks and Ctrl S-which are expanding to offer AI cloud computing-at a disadvantage relative to US rivals. Uncertainty over licences and trade negotiations could challenge India's ambitions for large-scale AI hardware deployment, the official said. The regulations divide the world into three groups-the first comprising 20 countries including Australia, the UK, France, Germany, Sweden, Switzerland have unrestricted access to AI chips, while 20 others including China, Russia, North Korea and Iran will have no access.

Don't lose sleep over US chip rule

US export controls on chips are unlikely to impact India's AI hardware growth. India operates within existing limits set by the Biden administration. Restrictions on China could benefit India by redirecting AI development work. India has significant potential as a semiconductor manufacturing hub, aligning with US technological goals without challenging them.US export controls on chips under the Framework for Artificial Intelligence Diffusion rule, announced last week, are unlikely to derail India's trajectory in AI hardware development, even as President Donald Trump begins his term today. The country is operating well within limits imposed by the previous Joe Biden administration, and it will be up to Trump to maintain these controls in their current form or make alterations. US companies are allowed to operate abroad under different quotas, which should address some of India's rising chip demand. India provides critical research inputs to US chipmakers that could prove valuable when seeking import-limit enhancements. Finally, restrictions on China will redirect some AI development work toward India, presenting a significant opportunity for the country. India's AI contribution is complementary to, not competitive with, US technological progress. Silicon Valley relies on Indian developers and consumers for faster AI diffusion. Hardware capacity is interconnected, and US export controls must remain dynamic, accounting for co-development and India's indigenous growth. This will also need to incorporate India's ambitions to become a base for semiconductor manufacturing. The country is only at the beginning of this journey and should have minimal impact on US concerns about re-export of advanced parallel-processing GNU chips to countries on its ban list. That New Delhi has not joined a Beijing-led Asia-Pacific free-trade bloc further strengthens its anti-proliferation credentials. From its earliest stages, AI development has been driven by commercial interests, unlike other technologies such as nuclear energy or space travel, which evolved from strategic use. Proliferation must be addressed differently, with a greater focus on encouraging economic benefits while discouraging an AI arms race. Private companies and agencies will play a larger role, just as they do in the commercial exploitation of space. With Trump inviting Silicon Valley to Washington, India will have influential allies to lobby for its AI programme.

Navigating AI Export Controls: US Sidelines India From Benefits To Closest Partners

U.S. providers such as Microsoft, Amazon, and Meta are anticipated to secure global authorizations, yet they will be restricted to deploying no more than 50% of their AI computing capacity outside the United States. Ashok Chandak, President of IESA (India Electronics and Semiconductor Association said, "The Export Control Framework for Artificial Intelligence (AI) diffusion, introduced by the Biden administration, is expected to significantly affect global AI development and deployment. These restrictions extend beyond U.S.-based companies, influencing nations like India, which now faces licensing requirements for importing advanced AI chips."

AI Mission Control: We Have a GPU Problem

The outgoing US administration's proposed rule to restrict import of graphics processing units (GPUs) -- needed for running heavy artificial intelligence models and applications -- may hit India's ambitions in this area in the long term, officials and experts told ET.The outgoing US administration's proposed rule to restrict import of graphics processing units (GPUs) -- needed for running heavy artificial intelligence models and applications -- may hit India's ambitions in this area in the long term, officials and experts told ET. Restricted access to advanced AI chips may slow innovation and development, while licensing needs could raise costs and introduce delays, they added. The rule released January 13 may increase compliance complexities for companies like RIL, which recently placed an order for Nvidia GPUs. It could put data centre providers such as Tata Communications, Yotta Data Services, E2E Networks and CtrlS, which are expanding to offer AI cloud computing, at a disadvantage relative to US rivals. The framework became effective from the date of release while still inviting public comments for potential adjustments, experts said. Government officials have called the move harsh since the India AI Mission will now come under the total country compute power applicable for India, requiring the country to apply for a licence. "It may cause some near-term hiccups on the compliance side," a senior government official told ET, adding the AI consumption is well within the cap as of now. The country cap imposed under the new rule is 50,000 GPUs or equivalent capacity. The government is currently sourcing close to 10,000 GPUs under the India AI Mission. The number may go up in the future. "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," said Ashok Chandak, president, India Electronics and Semiconductor Association. "However, small-scale setups could still enable experimentation, innovation, and restricted model development." Uncertainty over licences and trade negotiations could challenge India's ambitions for large-scale AI hardware deployment, he said. Mukesh Ambani-led RIL aims to build India's largest AI data centre, powered by Nvidia's most advanced GB200 GPUs, in a record 24 months. The GB200 GPU is six times more powerful than its predecessor, the H100. "Even if (Reliance-owned) Jio can secure GPUs, the quantity may be limited, potentially delaying or hindering their ability to scale their project," said Sajai Singh, partner, JSA Advocates & Solicitors. "The mandatory licensing process would also likely introduce additional bureaucratic hurdles and costs, which could further slow the development of their GPU clusters." There is also uncertainty over how countries will oversee equitable distribution and competition among large and small native companies, experts said. Reliance, Tata Communications, Yotta and Ctrl S didn't respond to queries. The regulations divide the world into three groups: the first have unrestricted access to AI chips, while 23 others including China, Russia, North Korea and Iran will have no access. For the remaining 140 nations, including India, there will be a country-specific cap on computing power equivalent to 50,000 advanced GPUs (worth about $1 billion).

India-US spark stands defused?

The outgoing Biden administration's Framework for Artificial Intelligence Diffusion categorizes countries based on trust with the US, impacting India's AI industry by imposing restrictions on importing GPUs. This classification may hinder Indian AI companies' global competitiveness and complicate diplomatic relations. The policy highlights the growing intersection of technology and geopolitics.Last week, outgoing president Joe Biden's administration released 'Framework for Artificial Intelligence Diffusion', which aims to establish export and security regulations for the global AI market. It divides countries in three groups based on the trust they enjoy with the US. While this is an outgoing US president's rule-making, it reflects behavioural elements of the new Donald Trump administration's transactional foreign policy approach. India might have expected a more favourable classification - perhaps alongside Britain and Australia (tier-1). While US companies face no restrictions on deploying computing power in tier-1 countries, tier-2 nations will encounter limits on how much they can import unless the computing power is hosted in secure environments. For India, this could mean restrictions on importing GPUs, potentially affecting its plans to procure 10,000 GPUs to bolster domestic computing capacity. Meanwhile, tier-3 countries like Russia, China, Libya and North Korea face an almost total ban on US tech exports. India's tier-2 status suggests two possibilities: India is seen as a potential competitor to the US in AI. Or, the classification serves as leverage in trade and diplomatic negotiations, pressuring India to grant greater access to US firms. Now, Indian firms will have to navigate General Validated End User (VEU) programme, a system that explicitly ties tech access to foreign policy. This is disconcerting since the US-India partnership, particularly through the iCET initiative, has wide expectations for deepening technological ties. India has been a strong proponent of a free and open internet, and has been working to strengthen its position in the global AI race. US restrictions, however, may stifle collaboration with international players, particularly in regions like Russia, Southeast Asia and Africa. The restrictive measures could further divide the global AI landscape into isolated silos, with a select few countries holding the reins of the future of AI. This rule could have a significant impact on multiple fronts: Global hurdle The new rule will hinder Indian AI companies' global competitiveness. A cumbersome case-by-case process for acquiring US tech will cause delays, create uncertainty and reduce India's appeal for AI-related FDI. It may also push Indian firms toward alternative suppliers, risking a fragmented and less efficient global AI ecosystem. Additionally, US cloud providers might be required to disclose customer details and AI usage, complicating operations for Indian businesses and threatening the country's digital sovereignty. Access denied Access to advanced AI models and chips is crucial for nations to sustain technological leadership and foster innovation. Falling behind in acquiring these technologies risks eroding competitive advantage, slowing economic growth and weakening their position in GVCs. Geopolitical effect Though not an explicit adversary, India is increasingly seen by the US through a lens of strategic competition. As Washington strengthens Indo-Pacific alliances, it aims to manage ties with non-aligned states like India to maximise leverage over their technological and military capabilities. The challenge for India lies in balancing its deepening partnership with the US while preserving its independent stance on global issues that may conflict with Washington's agenda. Diplomatic coercion The AI diffusion regime risks becoming a tool of diplomatic coercion. By tying tech access to political alignment, the US signals that it views AI as a strategic asset in the global balance of power. For India, this is a reminder that even as it marches forward with its ambitious digital transformation, it must also confront the reality that tech is increasingly intertwined with geopolitics in today's world. In this context, India must reassess its position in the global technological race. India's technological future will depend not only on its ability to innovate internally but also on how it navigates the increasingly complex web of global alliances and restrictions on emerging technologies. In the race for technological supremacy, the true battle lies not in the systems we create but in the alliances we build. The writer is a corporate adviser

Biden admin's cap on GPU exports may hit India's AI ambitions

The outgoing US administration's proposed rule to restrict import of graphics processing units (GPUs) -- needed for running heavy artificial intelligence models and applications -- may hit India's ambitions in this area in the long term, officials and experts told ET. Restricted access to advanced AI chips may slow innovation and development, while licensing needs could raise costs and introduce delays, they added. The rule released January 13 may increase compliance complexities for companies like RIL, which recently placed an order for Nvidia GPUs. It could put data centre providers such as Tata Communications, Yotta Data Services, E2E Networks and Ctrl S, which are expanding to offer AI cloud computing, at a disadvantage relative to US rivals. The framework became effective from the date of release while still inviting public comments for potential adjustments, experts said. Government officials have called the move harsh since the India AI Mission will now come under the total country compute power applicable for India, requiring the country to apply for a licence. "It may cause some near-term hiccups on the compliance side," a senior government official told ET. The AI consumption is well within the cap as of now, the person said. The country cap imposed under the new rule is 50,000 GPUs or equivalent capacity. The government is currently sourcing close to 10,000 GPUs under the India AI Mission. The number may go up in the future. Also Read: Nvidia dominates GPU offerings in government's Rs 10,000-crore AI mission "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," said Ashok Chandak, president, India Electronics and Semiconductor Association. "However, small-scale setups could still enable experimentation, innovation, and restricted model development." Uncertainty over licences and trade negotiations could challenge India's ambitions for large-scale AI hardware deployment, he said. Mukesh Ambani-led Reliance aims to build India's largest AI data centre, powered by Nvidia's most advanced GB200 GPUs, in a record 24 months. The GB200 GPU is six times more powerful than its predecessor, the H100. "Even if (Reliance-owned) Jio can secure GPUs, the quantity may be limited, potentially delaying or hindering their ability to scale their project," said Sajai Singh, partner, JSA Advocates & Solicitors. "The mandatory licensing process would also likely introduce additional bureaucratic hurdles and costs, which could further slow the development of their GPU clusters." There is also uncertainty over how countries will oversee equitable distribution and competition among large and small native companies, experts said. Reliance, Tata Communications, Yotta and Ctrl S didn't respond to queries. Also Read: Vendors offer top GPUs as MeitY decides on Rs 10,000-crore AI mission tender The regulations divide the world into three groups -- the first comprising 20 countries such as Australia, UK, France, Germany, Sweden, Switzerland have unrestricted access to AI chips, while 20 others including China, Russia, North Korea and Iran will have no access. For the remaining 140 nations, including US allies such as India, Saudi Arabia, and the UAE, there will be a country-specific cap on computing power equivalent to 50,000 advanced GPUs (worth about $1 billion). This cap can be doubled if these countries sign a pact with the US to uphold strict security standards, the rule said. "India's GPU consumption is well within the 50,000 cap as of now," said the chief executive of an Indian data centre company. "Currently, the combined compute power in India would be equivalent to 25,000 advanced GPUs. However, India being a vast market for AI use will require much more capacity in future, especially if the US continues to restrict model usage." He explained that this may create a sovereign requirement for India to build its own AI models. Global tech bodies and companies like Nvidia and Oracle have criticised US President Joe Biden's directive as the "most destructive to ever hit the US technology industry". The world's largest cloud providers, data centres, model developers and even nations will now need a licence from the US to use their AI chips as per the rule, they said. The move, which comes days before Donald Trump takes over the US presidency, is set to revise the existing export controls on advanced computing hardware. "Indian companies will face new licensing requirements to obtain advanced computing ICs and AI model weights from the US," said the spokesperson of a US-based tech policy body. "They will be subject to country-specific Total Processing Performance (TPP) allocations for imports of advanced ICs, limiting the total compute power available. Indian firms will not be eligible for the more permissive licence exceptions." Even the Indian government's AI Mission to procure large GPU clusters through public-private partnerships could get disrupted. "This may limit the scale or increase the costs," the person said. Moreover, if the country cap of 50,000 GPUs is reached, American hyperscalers such as Microsoft, Amazon and Google would still be able to deploy 7% of their global capacity in India even as their local counterparts won't be able to. "This could result in supply shortages, delays, and potentially higher costs for Indian firms that rely on cutting-edge GPU technology to remain competitive in the AI space," said Singh of JSA Advocates & Solicitors. "Companies like Nvidia, which dominate the AI GPU market, would be constrained in their ability to supply these products to Indian companies, thereby hindering their capacity to innovate and scale AI solutions."

United States Enforces Licensing Requirements For AI Chip Export

The United States has announced a new AI licensing regime that places restrictions on the exports of Artificial Intelligence chips to certain countries of concern, while vastly expediting the same process for its trusted allies. Secondly, the rules also add greater restrictions on the exports for weights of models trained with 10^26 computational operations or more. Thirdly, the Bureau of Industry and Security (BIS) will also impose security conditions on the storage of advanced AI models. A statement from the White House clarified that the policy, named Interim Final Rule on Artificial Intelligence Diffusion, was meant to prevent the smuggling of AI chips to adversaries of the United States and maintain American dominance over the industry. It stated that adversaries of the USA were using AI chips for offensive cyber operations, and aiding human rights abuses, such as mass surveillance. "To enhance U.S. national security and economic strength, it is essential that we do not offshore this critical technology and that the world's AI runs on American rails," said the statement. The Bureau of Industry and Security (BIS) will implement a global licensing system for the export, reexport, and transfer of advanced computing integrated circuits (ICs) and top-tier AI model weights. This system applies to all destinations and end users worldwide. The licensing decisions consider the destination's sensitivity, computing power quantity, AI model performance, and recipient security commitments. BIS will generally deny licenses for large quantities of advanced computing ICs intended for AI model training. The rules place countries under three categories: However, the policy also offers an exception for the export, reexport, or transfer (in-country) of eligible advanced computing ICs, related software and technologies to the first group of countries. It also provides an exception for the export of low amounts of compute, up to 26,900,000 Total Processing Performance (TPP) per year to a single consignee, as long as they are not members of adversarial nations. However, the exporters must provide information about the ultimate consignee of the goods to the BIS, alongside their certification. They must also ensure that the equipment is not accessed by entities within adversarial nations, without the prior approval of the BIS. The BIS established two categories of Data Center Validated End-Users (DC VEUs) in October 2024: Universal VEUs (UVEUs) and National VEUs (NVEUs). Companies can apply for UVEU status if they or their ultimate parent have their headquarters in the first group of nations. NVEU status is available to companies worldwide, except those in Macau or Country Group D:5 destinations, for acquiring large quantities of advanced computing integrated circuits. UVEUs may build data centres in any country in the world, except in Macau or Country Group D:5 destinations. They also face geographic restrictions on AI computing power allocation. Those headquartered in friendly countries cannot transfer more than 25% of their total computing power outside those designated countries, with a 7% limit per individual country. US-headquartered UVEUs cannot transfer more than 50% of their computing power outside the United States. NVEUs operate under a quarterly cumulative Total Processing Power (TPP) allocation system from 2025 through 2027. The allocations begin at 633,000,000 TPP in Q1 2025 and increase progressively to 5,064,000,000 TPP by 2027. The policy also implements new controls on advanced AI model weights effective starting January 13, 2025. For reference, AI model weights are numerical parameters that define the internal structure and decision-making logic of a machine learning model, and allow an AI system to learn patterns from data and make predictions or decisions. The controls focus on closed-weight models trained with more than 10^26 computational operations, as these represent the frontier of AI capabilities where security risks become significant. Companies that want to export model weights to other countries can only do so to foreign entities that have received a license from the BIS. However, an exception exists for specified friendly nations. The regulation also excludes open-weight models from control requirements, as they currently operate below the controlled threshold and offer significant research and development benefits. Small-scale additional training operations on open-weight models remain unrestricted. Reaction to the new export policy from the tech industry was not necessarily positive. Ned Finkle, vice president of government affairs at NVIDIA criticised the move in a blog post and accused the Biden administration of seeking to "undermine America's leadership." He termed the policy to be a "sweeping overreach," that would stifle competition and establish bureaucratic control over the industry. He also stated that the policy covered technology that was widely available in gaming PCs. India is mentioned once in the policy, alongside China in a two member list of countries eligible for General VEU Authorisations. These authorisations allow for the export, reexport, and transfer of eligible items to validated end-users of any in specific eligible destinations. Pranay Kotasthane, Deputy Director and Chairperson of the High Tech Geopolitics Program at the Takshashila Institution said that the new policies could increase friction in international trade for India. He stated that small orders made by Indian startups could be fulfilled easily, but any attempts at creating big data centres would need authorisation. "From a geopolitical point of view, it places countries into different clubs. India is not in the club with unrestricted access currently," he said. Kotasthane drew comparisons with the geopolitics of the 1960s and 70s, when countries existed in nuclear and non-nuclear camps. "While there is no immediate threat to Indian companies, the trend is certainly worrying. The USA could easily choose to decrease the allowed export thresholds. This creates uncertainty," he said. Kotasthane also weighed in on the Biden administration's decision to establish controls on the exports of model weights. "The rationale is that essentially they think that chip controls are not enough to stop the progress of China and Russia. The reason is because once you get the model weights, you don't need the GPUs," he said, explaining that model weights are arrived at after developers train an extensive amount of data on certain GPUs. "But if I have access to the model weights, then I don't even need those many GPUs," he said. He also pointed out that even current state of the art AI models were all below the threshold above which export restrictions applied. "There's no restriction on current models. So it is like a future-facing policy action," he said.