The Narendra Modi government has become known for its focus on building roads, highways, railways and airports, the infrastructure that speeds up economic growth. Now, it is planning to build another kind of infrastructure, the infrastructure that boosts research, innovation and development in artificial intelligence, machine learning and data processing and management. As India is getting more and more digital, it needs more computing infrastructure, and for that it needs cutting-edge computing hardware, the GPUs, or graphic processing units.
Since GPUs are scarce and very expensive, the Indian government has two options: let businesses and industry solve its own GPU problem or spend public money to make GPUs more accessible to startups, research bodies and key industries, just as it has done with the Production-linked Incentives (PLI) scheme to promote key industries for its Make in India mission. In the first case, India will remain a laggard in AI and other emerging technologies; in the second, it can emerge as a new-age tech hub. Large domestic computing capacity also ensures a country's data sovereignty.
GPU access: Govt may chip in with Nvidia deal
The government has realised the importance of GPUs for India and is working to find a solution. It may strike a deal with Nvidia, the world's biggest GPU producer, to source graphics processing units (GPUs) from it, and offer them to local startups, researchers, academic institutions and other users at a subsidized rate under its Rs 10,000 crore Artificial Intelligence Mission, ET has reported based on information form sources.
What exactly are GPUs, and what is their use?
GPU, as the name suggests, is basically a technology that can render graphics better. A more familiar related technology is the CPU, or central processing unit, which is the brain of a computer. A GPU is often called a computer's heart or soul. GPUs are an evolved form of CPUs for their capacity of parallel computing. While the CPU mostly tackles tasks one by one, the GPU can break a huge and complex task into thousands or millions of small parts and execute them at once. That's why a GPU has such humungous computing power.
Graphics card, which helped your computer display images better, was an early avatar of GPU. Originally, like graphics cards, GPUs were used for better and faster rendering of 3D graphics and, therefore, were mainly of importance for gaming and 3D modelling. However, now their huge computing power is used for artificial intelligence, machine learning and supercomputing. With the sudden explosion of AI in the past few years, GPUs have become the most wanted, and expensive, hardware for computing.They are also needed in data centres where data in very large amounts has to be processed and managed.
Why India needs GPUs so much
India has turned digital at a mass level in the past decade. Due to explosive growth in digital payments due to the India stack, data revolution in telecom after Reliance Jio brought down tariffs, and government schemes that promote digital technologies, deeper penetration of internet, the launch of 5G services, growth of e-commerce, etc. India needs a bigger and more powerful digital hardware infrastructure, the GPUs, that is.
With so much of business, economic and social activities going digital, more and more data centres are coming up in India, especially when the government also wants companies to store Indian data in India itself. These data centres need GPUs.
But even more importantly, India needs GPUs to spur innovation in emerging technologies such as AI. Startups, small innovators and businesses and academics can't afford to buy GPUs. India's computing infrastructure is less than 2 percent of global capacity which is a limiting factor in its contribution to research that remains in the range of 2 per cent, as per NVIDIA Asia South MD Vishal Dhupar. While speaking at StartUp Mahakumbh last month, Dhupar said, "India today is approximately sub-2 per cent as compared to the US and China combined which is closer to 58-59 per cent." He said India's contribution to research globally has a direct correlation with its low computing capacity. Indians abroad, Dhupar said, are contributing 12 percent because of the computing infrastructure available there.
India now has over 100 generative AI startups. But the investment into this sector has been comparatively small. The US saw nearly $250 billion private investments into AI startups between 2013 and 2022, while investments in India stood at just $8 billion. Over the same period, China saw $95 billion of investments while for the UK, the number stood at $18 billion, as per data from the AI Index 2023 Annual Report.
If India needs to be among the leaders in emerging technologies, it must acquire more computing power.
What government can do
The problem with the GPU access in India is that GPUs are expensive and scarce. According to industry estimates, NVIDIA dominates the GPU market with about 88 per cent share and there is a lag of 12-18 months in getting GPUs from the company due to its high demand across the world.
In such a scenario, the government will have to spend money to make GPUs accessible to industry, innovators and academia. The governments often give incentives to industries and businesses which are vital for economic growth. For instance, the government has spent billions of dollars on the PLI scheme to subsidise several kinds of industries to make India a manufacturing power and achieve self-reliance. A similar initiative is required in case of GPU access too.
Already, the government has approved the India AI Mission with an outlay of Rs 10,372 crore for five years to encourage AI development in the country. Under this mission, supercomputing capacity, comprising over 10,000 GPUs, will be made available to various stakeholders for creating an AI ecosystem “The government’s AI mission is incredibly exciting for India and its vibrant startup ecosystem. Investing in 10,000 GPUs and making them available to researchers and innovators will make a huge difference as we aim to build (India) the AI application capital of the world,” Rajan Anandan, managing director, Peak XV Partners, has told ET recently, adding that it will provide critical building blocks that AI startups can leverage to “build in India, for India and for the world”.
Globally, countries like the UK, Saudi Arabia and the UAE have been shelling big money to acquire AI chips to boost their countries' companies. For instance, the UK is building a national AI resource, as a part of which it would acquire 5,000 Nvidia GPUs. Saudi Arabia, through the King Abdullah University of Science and Technology, reportedly bought 3,000 Nvidia GPUs worth $40,000 each. This year, the UAE announced a $500 million investment to Falcon Foundation to develop open-source generative AI models and provide technology access to emerging economies.
Pointing out that the “market of GPUs is very dynamic with new technologies coming in and rendering existing technologies outdated,” Tanuj Bhojwani, head, people+ai, an initiative by Infosys co-founder Nandan Nilekani’s EkStep Foundation, has told ET recently that “the government should play the role of investor and not (of) a purchaser or a customer. “Merely going out into the market and purchasing 10,000 GPUs may not do the trick, the government should think like an investor,” he added.
India is considering two possible ways to provide AI compute infrastructure to its companies, as GPUs have become a very expensive and scarce resource, ET has reported.While one is "rent-and-sublet", where the ministry of electronics and information technology will acquire the GPUs from Nvidia, the other is a marketplace model where the government will encourage companies themselves to strike a renting or subletting deal with the supplier and then provide them incentives, as under production-linked incentive (PLI) schemes. "These incentives will be based on incremental productivity achieved through these GPUs," a government official has told ET.
GPUs are a critical but scarce resource and that's why an effective government intervention is required for affordable and accessible GPUs.
No comments:
Post a Comment