3 days ago
The MANGO paradox: Why India has been unable to harvest the fruits of the AI revolution
India finds itself in an ironic predicament. The country that produces over 20 million tonnes of mangoes annually — nearly half the world's supply — has failed to cultivate its own MANGO ecosystem. Not the sweet, succulent fruit that graces our summer tables, but the tech titans that rule the artificial intelligence landscape: Meta, Anthropic, Nvidia, Google, and OpenAI.
This is particularly puzzling given India's extraordinary contribution to these very companies. Indian engineers occupy critical development roles across all MANGO organisations. Sundar Pichai leads Google, Satya Nadella helms Microsoft (a key OpenAI partner), and countless Indian engineers power the innovation engines of Meta, Nvidia, and Anthropic. According to recent research, 65 per cent of leading US AI companies have at least one leader of Indian origin, while 70 per cent of full-time graduate students in AI-related fields are international — with Indians forming a substantial portion.
Yet, despite this talent goldmine and the ambitious India Techade initiative promising to make technology 20 per cent of the country's GDP by 2025–26, India remains conspicuously absent from the global AI leadership table.
The numbers paint a stark picture. While Meta and Microsoft train their models on supercomputers containing over 10,000 GPUs, India's fastest supercomputer, AIRAWAT, has merely 656 GPUs — ranking a humbling 75th globally. The government's recent announcement of establishing over 18,000 GPUs under the IndiaAI Mission sounds impressive until one realises OpenAI alone uses more than 100,000 GPUs for model training.
The computing capacity gap is staggering. Former IT Minister Rajeev Chandrasekhar claimed India needs about 24,500 high-performance GPUs to meet current demand. With AI operations scaling exponentially, even the planned 25,000-GPU cluster — if realised — would represent just a fraction of what leading AI companies deploy.
Export restrictions add another layer of complexity. While US-based tech giants can easily obtain National Verified End User licences to deploy GPUs at scale, Indian companies — even conglomerates like Reliance or Adani — face significant hurdles. This hardware accessibility challenge directly impacts India's ability to train competitive AI models from scratch.
Perhaps the most frustrating aspect is the brain drain. India's AI skill penetration rate of 2.8 is the highest globally, meaning Indian workers are 2.8 times more skilled in AI-related competencies than the global average. Yet this talent consistently flows westward. Recent data from job portals show an 11.4 per cent increase in international job listings targeting Indian talent, accompanied by a 59.4 per cent surge in applications by Indians seeking overseas opportunities. The reasons are well documented: Better remuneration, superior research infrastructure, and more conducive innovation environments.
India's struggles run deeper than infrastructure and talent retention. The country's R&D spending remains at around 0.7 per cent of GDP, compared to over 3 per cent in the US and 2.4 per cent in China. This underinvestment in research creates a vicious cycle where bright minds seek opportunities elsewhere.
The focus on IT services rather than deep tech innovation has also hindered progress. While companies like TCS, Infosys, and Wipro generate substantial revenues, they primarily offer services rather than developing cutting-edge AI products. The absence of large-scale AI companies comparable to OpenAI, DeepMind, or Anthropic reflects this service-oriented mindset.
Furthermore, the startup ecosystem, while growing, lacks the venture capital depth and risk appetite necessary for foundational AI research. Building a competitive language model requires not just talent but massive capital — estimates suggest between $7 and $15 million for basic foundational work, with training costs for advanced models reaching hundreds of millions.
The government's India Techade initiative represents recognition of these challenges. The IndiaAI Mission, with a budget of Rs 10,372 crore (approximately £1 billion), aims to create a scalable AI computing ecosystem. Recent partnerships with companies like Nvidia — where Indian IT giants including TCS, Wipro, and Infosys are building AI agents — show promising momentum.
China's DeepSeek offers an intriguing model. The company achieved competitive performance using just 2,000 Nvidia H800 GPUs compared to OpenAI's 100,000 GPUs, with training costs of $6 million versus OpenAI's $100 million. This efficiency-driven approach could provide a blueprint for Indian AI development, particularly given hardware access constraints.
Perplexity AI's founder, Aravind Srinivas — an Indian-origin entrepreneur — advocates for this path: Building indigenous models from scratch rather than fine-tuning existing ones. Drawing inspiration from ISRO's frugal innovation philosophy, he believes India can create globally competitive AI through strategic investment and ingenuity.
So here's the delicious irony: India, the land that perfected the art of growing mangoes — carefully selecting the right soil, climate, and cultivation techniques to produce the world's finest fruit — has somehow failed to apply the same principles to growing its MANGO ecosystem.
Perhaps we've been too busy exporting our best engineers to Silicon Valley's orchards, where they help others harvest the fruits of AI innovation. Meanwhile, back home, we're still debating which seeds to plant while our competitors have already built entire groves.
Maybe it's time India stopped just growing mangoes for export and started cultivating its own MANGO ecosystem. After all, who better than the land of mangoes to show the world how to grow the next generation of MANGOs? The irony would be complete when India finally builds its AI empire — and perhaps names its first unicorn AI company 'Mango' as a sweet reminder of what was always possible.
The ingredients are all there: The talent, the ambition, even the government support. Now India just needs to stop letting its best seeds grow in foreign soil and start planting its own orchard. The question isn't whether India can compete in AI — it's whether it will choose to do so on its own terms, in its own backyard.
Perhaps even the geopolitical noise from figures like Trump could prove a blessing in disguise — nudging India to define its AI future on its own terms.
The writer, a defence and cyber security analyst, is former country head of General Dynamics
