Indian cloud companies and data centre providers are on edge as the United States' new rules restricting export of advanced AI chips—including graphics processing units (GPUs)—take effect next week.
The Artificial Intelligence (AI) Diffusion Framework may severely impact India’s ability to import high-end chips like Nvidia’s B200, with estimates suggesting a limit of just 20,000 units under the proposed total processing performance (TPP) limit, industry insiders said.
They want the government to take up the issue with the US authorities.
“We need more processing power,” Darshan Hiranandani, chairman and cofounder of data centre service provider Yotta, told ET.
India—with more than a billion population and one of the largest user bases for the likes of OpenAI and Microsoft—has the same quota of processing power as a country with three million population, he pointed out.
“Though the fact that India is within the 50,000 GPUs threshold is comforting, we are a large data market. There is a lot of uncertainty and that is concerning,” Hiranandani said.
Others are also finding the compliance process cumbersome.
The US policy—notified by the previous Joe Biden government—seeks to protect the country’s AI leadership and national security by limiting global supply of advanced AI chips, which are controlled by US firms like Nvidia, AMD and Intel.
As per the policy, India is among Tier-2 countries where besides volume caps, transactions that could contribute to development of advanced AI models would require validated end user (VEU) authorisation.
Under the proposed TPP cap, if India opts for Nvidia B200, the country can buy only up to 33,000 GPUs while the number would drop to 27,000 GPUs in the case of B300, explained Piyush Somani, founding chief executive of Cloud company ESDS Software Solution.
If it is GB200, then India can buy only up to 25,000 GPUs, and if it is H200, then the country can buy up to 50,000 GPUs, said Somani, who is also the president of industry body Cloud Computing Innovation Council of India.
Through our foreign ministry, we need to ask for special exemption for India, he said.
This limitation could significantly impact Indian AI developers who rely on these powerful GPUs for training complex AI models. They might be forced to depend on overseas cloud providers, incurring higher costs compared to utilising locally hosted infrastructure, experts said.
India should either be in a Tier-1 category with no GPU import limits, or the country’s quota within Tier-2 should be adjusted based on factors like population size, digital adoption, and the strong, friendly relationship India has with the US, Sunil Gupta, cofounder and chief executive of Yotta Data Services, said.
Hiranandani is hopeful that the US would ease the restrictions. “Given that this is a notification by the previous administration in the US, I am hopeful that US companies like Nvidia, Intel and AMD—along with global cloud providers like Amazon, Microsoft and Google—will work with the current administration and have this streamlined as the stakes are much higher for them across the world," he said.
Uncertainty over process
An executive from an Indian cloud company highlighted the uncertainty surrounding import processes post May 15 when the new rules come into effect.
“Even if you want to buy it now it has become ridiculously complex, and from May 15, when the US' AI Diffusion Framework comes into effect, we don't know what the process is and how to import it,” the executive said. “Despite the narrative that India is within the 50,000 GPUs threshold, there is a massive process. None of the companies know how to import.”
The overarching concern remains the TPP limit, which could stifle India's ability to access the most advanced AI hardware necessary for future growth and innovation.
“The restriction that we have right now is very onerous because we cannot have the flexibility for the future without this restriction going up,” said Prateek Jhawar, managing director and head of infrastructure and real assets investment banking at Avendus Capital.
The company is advising bidders in the IndiaAI Mission GPU tender like NxtGen, and is closely monitoring these developments and their potential impact on the market.
“Fifty thousand GPUs can be consumed very quickly in very few developments essentially, and then you are locked out as a country completely,” Jhawar said.
As May 15 approaches, the Indian AI community awaits clarity on the final form of the US AI Diffusion Framework and its long-term implications for the nation's technological aspirations.
#Pahalgam Terrorist Attack
Live Updates| India strikes Pakistan two weeks after terrorist attack
Pakistan a safe haven for terrorists, further attacks on India were planned, says MEA
Watch how Operation Sindoor destroyed Pakistan terror camps
They want the government to take up the issue with the US authorities.
“We need more processing power,” Darshan Hiranandani, chairman and cofounder of data centre service provider Yotta, told ET.
India—with more than a billion population and one of the largest user bases for the likes of OpenAI and Microsoft—has the same quota of processing power as a country with three million population, he pointed out.
“Though the fact that India is within the 50,000 GPUs threshold is comforting, we are a large data market. There is a lot of uncertainty and that is concerning,” Hiranandani said.
Others are also finding the compliance process cumbersome.
The US policy—notified by the previous Joe Biden government—seeks to protect the country’s AI leadership and national security by limiting global supply of advanced AI chips, which are controlled by US firms like Nvidia, AMD and Intel.
As per the policy, India is among Tier-2 countries where besides volume caps, transactions that could contribute to development of advanced AI models would require validated end user (VEU) authorisation.
Under the proposed TPP cap, if India opts for Nvidia B200, the country can buy only up to 33,000 GPUs while the number would drop to 27,000 GPUs in the case of B300, explained Piyush Somani, founding chief executive of Cloud company ESDS Software Solution.
If it is GB200, then India can buy only up to 25,000 GPUs, and if it is H200, then the country can buy up to 50,000 GPUs, said Somani, who is also the president of industry body Cloud Computing Innovation Council of India.
Through our foreign ministry, we need to ask for special exemption for India, he said.
This limitation could significantly impact Indian AI developers who rely on these powerful GPUs for training complex AI models. They might be forced to depend on overseas cloud providers, incurring higher costs compared to utilising locally hosted infrastructure, experts said.
India should either be in a Tier-1 category with no GPU import limits, or the country’s quota within Tier-2 should be adjusted based on factors like population size, digital adoption, and the strong, friendly relationship India has with the US, Sunil Gupta, cofounder and chief executive of Yotta Data Services, said.
Hiranandani is hopeful that the US would ease the restrictions. “Given that this is a notification by the previous administration in the US, I am hopeful that US companies like Nvidia, Intel and AMD—along with global cloud providers like Amazon, Microsoft and Google—will work with the current administration and have this streamlined as the stakes are much higher for them across the world," he said.
Uncertainty over process
An executive from an Indian cloud company highlighted the uncertainty surrounding import processes post May 15 when the new rules come into effect.
“Even if you want to buy it now it has become ridiculously complex, and from May 15, when the US' AI Diffusion Framework comes into effect, we don't know what the process is and how to import it,” the executive said. “Despite the narrative that India is within the 50,000 GPUs threshold, there is a massive process. None of the companies know how to import.”
The overarching concern remains the TPP limit, which could stifle India's ability to access the most advanced AI hardware necessary for future growth and innovation.
“The restriction that we have right now is very onerous because we cannot have the flexibility for the future without this restriction going up,” said Prateek Jhawar, managing director and head of infrastructure and real assets investment banking at Avendus Capital.
The company is advising bidders in the IndiaAI Mission GPU tender like NxtGen, and is closely monitoring these developments and their potential impact on the market.
“Fifty thousand GPUs can be consumed very quickly in very few developments essentially, and then you are locked out as a country completely,” Jhawar said.
As May 15 approaches, the Indian AI community awaits clarity on the final form of the US AI Diffusion Framework and its long-term implications for the nation's technological aspirations.
