In a move aimed at propelling India and helping local startups in AI innovation, Union Minister Rajeev Chandrasekhar unveiled a plan to establish a cluster of 25,000 Gpus. 

This initiative, set to be realised through a public-private partnership (PPP), is in discussion at the highest levels of the Ministry of Electronics and IT. Chandrasekhar announced this ambitious endeavour in September, shedding light on a commitment to fostering real-world AI applications. 

The minister stated that the ongoing discussion about AI is almost always about applications like ChatGPT. “Our mission is real-world AI use cases. We are looking at health, governance, education and creating AI-specific integrated circuits for those applications,” he said in a statement.

Abundant Data, Need GPUs

At its core, this initiative is not just about enhancing India’s AI prowess, it’s also about safeguarding the nation’s data sovereignty. The scarcity of GPUs within the country has driven many businesses to rely on overseas cloud-based solutions. Recognising the urgency of addressing this issue, one of the seven AI working groups established by MeitY strongly recommended the creation of a 25,000 GPU cluster.

Experts contend that India’s abundant data resources and human capital necessitate supercomputing power to compete effectively in the global AI arena. This massive cluster of GPUs is a pivotal step toward achieving this objective. For context, India’s current fastest supercomputer, ‘Airawat‘, boasts a mere 640 GPUs, ranking 75th globally. In contrast, the world’s top supercomputers feature over 30,000 GPUs.

Once the proposal is finalised, the Government will initiate a standard tendering process to invite private companies to participate in establishing the GPU cluster. Notably, discussions with tech giants like NVIDIA, including a meeting between its founder and CEO Jensen Huang and Prime Minister Modi, have underscored the potential for collaboration. 

Huang told AIM that India will get about 10s of thousands of GPUs in order to build infrastructure – i.e. about 1,00,000 GPUs. “We are going to bring out the fastest computers in the world. These computers are not even in production [so far]. India will be one of the first countries in the world [to get them],” Huang said, confirming that these would be faster than anything the world has ever seen.

Going by the figures that the NVIDIA founder was hinting at, the 25k GPU cluster could very well be part of a larger shipment containing 1,00,000 GPUs.

However, while this initial initiative is undeniably significant, it merely marks India’s initial foray into the arena of countries promoting AI research and development capabilities. Compared to companies like OpenAI, which possess over 20,000 GPUs, and a $10 billion investment from Microsoft, the government will require private-sector partnerships to fully harness the potential of this computing power.

The estimated cost of this ambitious project falls in the range of INR 8,000-10,000 crore and is currently under deliberation at the highest echelons of MeitY. Indian AI startups, industry players, and prominent CEOs have persistently advocated for such investments in computing capacity to address the scarcity and prohibitive cost of GPUs.

Issues Facing Local AI Advancement

National initiatives to construct supercomputers and projects aimed at training LLMs in multiple Indian languages are already in progress. However, there are many issues facing it.

When companies seek access to GPUs from cloud service providers or GPU manufacturers, they often face extended wait times, sometimes spanning months. To address this bottleneck, companies are urging the government to invest in essential computing infrastructure for AI systems and applications. Without this support, India risks lagging in the global AI race, which encompasses applications from banking to space stations, all powered by algorithmic intelligence.

“Leading Indian startups are grappling with the challenge of obtaining access to 1,000-GPU clusters, often diverting valuable funds from their fundraising efforts,” MD of PeakXV Partners (previously Sequoia India) Rajan Anandan, said; emphasising the need for affordable access to such clusters, suggesting a pyramid approach: free access for academic institutions, subsidised access for startups, and commercial access for larger companies.

IBM CEO Arvind Krishna, recently reiterated the same, saying, “In many nascent technologies, you often need the government to step in first before others follow.”

“The government should set up a national Al computing centre,” Krishna stressed.

India is home to over 60 active genAI startups as of May 2023, having received approximately $475 million in funding between 2021 and 2023. While India’s AI ecosystem is thriving, it lags behind countries like the United States and Israel in foundational AI models and funding. 

What About Other Countries?

In contrast, governments in other nations have committed substantial funds to secure GPU access for research purposes. The UK, Saudi Arabia, the UAE, and Chinese tech giants have all invested significantly in acquiring GPUs to bolster their AI capabilities. Even the United States has offered a 50% discount to researchers working on supercomputing projects.

Lack of local GPU access may drive Indian companies to opt for foreign cloud providers, leading to data localisation issues. Sudipta Ghosh, partner and leader of data and analytics at PwC India, emphasised the importance of regulatory frameworks for responsible and ethical AI. Such frameworks would not only bolster public trust but also ensure transparency and accountability.

The scarcity of GPUs has also made them more expensive in India, leading providers to be cautious about shipping them to the country where demand, payment capabilities, and ticket sizes are comparatively smaller.

Addressing this challenge may require domestic GPU manufacturing, potentially supported by government incentives. Prashant Garg, Partner at EY, suggests following the model of attracting global automakers to set up shop in India.

Efforts are already underway to provide controlled GPU access to AI startups through collaborations between Nasscom and the Centre for Development of Advanced Computing (CDAC). However, experts agree that India needs to invest in fundamental research and attract top AI scientists to drive innovation.

Simultaneously, collaborations between American GPU manufacturer NVIDIA and Indian giants Reliance Jio and Tata Sons hold promise for providing computing infrastructure to emerging businesses.

Looking ahead, a lot of interest from other semiconductor giants like AMD, Micron, SOLIS-IDC, Foxconn, STMicroelocronics could snowball into a lot of GPU manufacturers coming in as well. With smooth regulations and favourable business conditions, India could have GPUs being manufactured locally.

The post India’s 25K GPUs for AI: Is that Enough? appeared first on Analytics India Magazine.