-
CENTRES
Progammes & Centres
Location
Fuelled by talent but constrained by compute, can India build the engine of the AI age?
Image Source: Getty
Nations that secure abundant, low-cost, and scalable compute power will dominate tomorrow’s innovation economy. Much like the energy superpowers that shaped the 20th century, compute-rich nations will shape the 21st century. This is already evident in the techno-industrial contest between the US and China, with both racing to achieve self-sufficiency in semiconductor design and manufacturing, sovereignty in AI diffusion, and struggling for high-tech supply chain control. Recent US export controls on NVIDIA’s H100-class processors reflect a new era in which frontier hardware is treated as a securitised asset, withheld from rivals not just through denial, but also through delay and access tiers. The US leans on private-driven, government-supported models, which have made Silicon Valley the harbinger of tech-revolution, while China actively leverages industrial policy, a practice once dismissed by American policymakers but now grudgingly embraced in strategic industries like AI.
The US leans on private-driven, government-supported models, which have made Silicon Valley the harbinger of tech-revolution, while China actively leverages industrial policy.
This transformation parallels broader debates around resource optimisation. In India, high-technology thought leaders such as Nandan Nilekani propose tokenising land on a blockchain ledger to enhance liquidity and tradability, which can also generate higher property tax revenues and break entrenched inefficiencies in the Indian land recording system. Similarly, compute must be treated as a fundamental economic asset, one that nations strategically deploy rather than passively consume.
The world is not running out of data, it is running out of compute. Semiconductor scaling laws are slowing, just as AI demands are accelerating. The result? A massive bottleneck. Compute infrastructure, which includes advanced chips, data centre infrastructure, and AI accelerators, has become the upstream driver of downstream innovation.
Countries with fast, cost-efficient compute stacks (from chip design and packaging to cloud and fibre) will become global digital hubs. The downstream effects of this compute boom are already visible in startup ecosystems worldwide, where compute abundance unlocks use cases in defence tech and biotechnology.
This brings us to India.
During a recent event, India’s Minister of Commerce, Piyush Goyal, expressed concerns that the Indian startup culture has not been doing enough. India’s ambition to become a deep tech powerhouse is growing real, but it risks being compute-constrained. Government initiatives like the IndiaAI Compute Portal, which intends to democratise access to 18,600 Graphical Processing Units the base architecture of modern-day AI Accelerators deployed through data centres at heavily subsidised prices through public-private efforts, are commendable steps forward. However, they remain modest when placed next to trillion-dollar investments by the US and China. Elon Musk’s xAI has added 1,00,000 GPUs to develop its Colossus AI supercomputer, while Meta is believed to have deployed 6,00,000 H100 GPUs in the year 2024. While the ‘DeepSeek phenomenon’ has addressed the issue of efficient pre-training strategies through smarter data selection and model design, compute power remains a concern for a startup intending to develop the next Frontier AI Model.
Historical development in semiconductors created a substrate on which current-day digital economies are built, where compute is the currency.
India’s innovation landscape, exemplified by firms like Zomato’s Blinkit, is often portrayed as an internet-driven service sector addressing the immediate pain points of urban Indian households, such as hyperlocal grocery delivery, through agile execution and platform-driven network effects. However, to dismiss such firms as merely a “convenience layer” risks overlooking the strategic infrastructure that they have already embedded into the Indian economy. From the perspective of Jeffrey Ding’s theory of diffusion, deep tech sectors such as semiconductors preceded as well as created the modern consumer electronics market and the modern-day AI hardware accelerators market. Thus, historical development in semiconductors created a substrate on which current-day digital economies are built, where compute is the currency. India’s existing base-level industrial capacity will act as the means of AI diffusion, depending on the quality and quantity of computing available locally.
For India, the challenge is not a lack of human talent—Indian deep tech founders and engineers are globally competitive. The challenge is systemic, as stated by Saurabh Chandra, founder of Ati Motors, which builds autonomous bots to move material in factories and warehouses. India lacks the connective tissue to turn world-class innovation into industry. This includes the cases of capital misalignment. Venture Capital and Private Equity firms still prefer SaaS, e-commerce, media, real estate and fintech, where returns are quicker and risks are relatively lower. With domestic acquirers of deep tech startups’ products and solutions in sectors such as semiconductors, artificial intelligence, and quantum computing being few and far in between, a significant number of frontier-tech solutions providers are being acquired by foreign firms. For instance, BRIDGEi2i, a Bengaluru-based Artificial Intelligence startup which had Edelweiss Private Equity as a minority stakeholder, was acquired by Ireland-based Accenture Ltd in 2021. Similarly, Uncanny Vision, an AI-driven, cloud-based video surveillance company headquartered in Bengaluru, was acquired by the US-based Eagle Eye Networks. The absence of a robust and competitive network of domestic firms having a presence in high-value markets with a global network of clients actively seeking deep tech solutions means that India is missing a crucial commercial dynamic. Such a commercial dynamic ensures that the IP and value generated during the life cycle of the given startup stays within the national ecosystem of innovation and creates a self-sustaining domestic deep tech landscape.
With domestic acquirers of deep tech startups’ products and solutions in sectors such as semiconductors, artificial intelligence, and quantum computing being few and far in between, a significant number of frontier-tech solutions providers are being acquired by foreign firms.
The defence sector, which is a natural consumer of deep tech, not only has the requirement but also the inclination towards developing use cases where deep tech solutions can be deployed. A defence ecosystem that scouts and funds promising deep tech startups and signs preferential procurement contracts can shift the overall risk calculus of the civilian sector. For example, Fairchild Semiconductors (the company from which Intel spun out) secured its place in history when the National Aeronautics and Space Administration (NASA) chose Integrated Circuits (IC) for their Apollo Guidance Computer. While at the time, IC was a risky and unproven technology, NASA’s need for lightweight and reliable electronics drove its adoption. This procurement decision de-risked the IC technology, leading to the creation of the modern semiconductor industry. This blueprint can be adopted by India’s space and defence sectors for developing a long-term ecosystem of innovation in India.
The present-day deep tech developments, including AI, are compute-hungry. India has the potential to cater to the emerging use cases of AI at a scalable commercial scope. Without sustainable and democratised access to high-performance compute infrastructure, India’s human talent stays trapped in developing low-latency use cases with marginal innovation that prioritise product-market fit over foundational breakthroughs. This is because the current innovation ecosystem rewards scale and speed rather than originality and depth. Providing compute as a cold-start resource to startups working in the development of deep tech solutions, especially in AI, can unlock conditions for a homegrown innovation flywheel, which could assist with the development of the next General-Purpose Technology (GPT).
Without sustainable and democratised access to high-performance compute infrastructure, India’s human talent stays trapped in developing low-latency use cases with marginal innovation that prioritise product-market fit over foundational breakthroughs.
India needs a systemic intervention, not just further capital infusion. A robust deep tech ecosystem must be designed for downstream diffusion - creating additional demand, not just supply. This means capacity building in compute infrastructure through public-private partnerships. The finalisation of the IndiaAI Compute Grid’s L1 bids marks a key milestone in building the necessary compute infrastructure. Direct compute demand will cascade in three layers. Generative/Predictive AI startups will form the vanguard of compute demand since they require raw GPU power to train their models, specific to the sectoral requirements they are targeting to capture. SaaS platforms will form the second layer since their competition with global demand for embedding AI in existing software offerings will require cloud-native GPU clusters. A third layer of demand will be generated by traditional industries like manufacturing and banking, which will be shifting to cloud-based systems to fuse their IoT data with predictive analytics. These members of the industry will act as the consumer of members in layers 1 and 2.
Further, three key accelerators that will amplify this demand are broadband expansion to assess, create and deploy various AI use cases for 700 million plus internet users by 2030. Grid-stable states such as Maharashtra and Tamil Nadu that proactively develop their data centre ecosystems will also influence the quality of compute supply. The famous United States-led Framework for Artificial Intelligence Diffusion has regulated GPU supply for Tier-2 nations, capped at 49,901 till 2027 and a maximum of 1,699 advanced GPU per firm annually without requiring a National Validated End User license from the Bureau of Industry Standards. Further, the development of closed AI models with weights more than 10^26 FLOPS (Floating Points per Second) is also restricted for nations. Hence, commerce and trade diplomacy with the United States is a necessary variable to determine the quality as well as quantity of domestic compute supply. The interplay of these forces will create new markets where compute will determine the standards of deep tech diffusion in India.
Compute must be seen not as an infrastructure cost, but as an economic multiplier, the foundation upon which globally scalable and monetisable deep tech applications can be built.
Creating anchor markets through defence, health, and strategic public procurement, incentivising local VC participation in high-tech, long-horizon sectors and building acquirer networks to ensure value retention in India will play a significant role and inform the government and private sector decisions in the deep tech space. Compute must be seen not as an infrastructure cost, but as an economic multiplier, the foundation upon which globally scalable and monetisable deep tech applications can be built. In the 21st century, the most valuable resources are not mined. They have to be built by participating in the global value chain. Nations that recognise compute as strategic and align their policy, capital, and infrastructure accordingly will control the terms of tomorrow’s innovation economy. India has the talent. Now it needs the tools, the runway, and the will to take off.
Nisarg Jani is a Doctoral Candidate at Pandit Deendayal Energy University, Gujarat.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Nisarg Jani is a Doctoral Candidate at Pandit Deendayal Energy University situated in Gujarat, India. The author is also a Senior Research Staff Member at ...
Read More +
PREV
