India has the highest AI skill penetration rate of any country in the world, the most enterprise AI adoption, and a government that has deployed 38,000 graphics processing units toward sovereign compute infrastructure NVIDIA Blog. What it does not have, according to the director of IBM Research India, is an answer to whether any of that adds up to the companies that will own the next wave of AI.
"Whether we can make the next three trillion dollar startups, that's a whole other question," Amit Singhee said in an interview with the Eye on AI podcast published this month Eye on AI Podcast. His view: India can apply AI for its own benefit. Whether it can build the AI companies that will sit at the top of the global technology stack is a different problem, and the country's track record on research talent suggests the gap is real.
The Stanford HAI AI Index 2026, released this month, provides the dataset behind Singhee's caution Indian Express / Stanford HAI AI Index 2026. India scored 3.0 on the AI skill penetration measure, the highest of any country measured, ahead of the United States at 2.0 and Germany at 1.8. Enterprise AI adoption reached 88 percent last year, up from 77 percent the prior year. Private investment in Indian AI companies totaled $4.09 billion in 2025.
The same dataset shows the countervailing pressure Indian Express / Stanford HAI AI Index 2026. India recorded the world's largest net outflow of AI research talent, at negative 16.9 on Stanford's flow index. It also recorded the sharpest rise in AI anxiety of any country surveyed, up 14 percentage points in one year, against a 2-point rise in excitement. The government's own estimate is that India will need one million AI professionals by 2026. The people are leaving faster than the infrastructure is arriving.
Singhee returned to India from the United States roughly a decade ago, he said, for family reasons rather than because India presented the best research opportunity Eye on AI Podcast. IBM has maintained a research presence in India for 27 years, embedded initially inside IIT Delhi. The company runs four to five funded research collaborations with IIT Bombay, IIT Delhi, and IISc at any given time, with co-principal investigators from the institutions and IBM directing graduate student work. Singhee said IBM finds its talent mostly in the local market rather than through repatriation programs. "If there is good talent, there's a high chance that they'd want to go to a research lab like ours," he said. "It's definitely not the case at large that there's a lot of local talent at that level available at scale in the way it would be, say, in the US."
The IndiaAI Mission is the government's answer to the compute side NVIDIA Blog. The program has deployed more than 38,000 graphics processing units through the IndiaAI Compute Portal, backed by over $1 billion in government partnership commitments with NVIDIA, toward sovereign compute capacity built outside foreign cloud providers. A 17-billion parameter mixture-of-experts model called BharatGen is also in development, with IBM contributing to the broader mission ecosystem IBM Blog / Business Today.
Singhee said the government's focus and ambition on compute is real. He pointed to China as the instructive case Eye on AI Podcast. Beijing repatriated talent through competitive salaries, but that alone did not close the gap. "They suddenly didn't leapfrog the whole supervised learning optimization phase," Singhee said. "It was the sudden transformation to generative AI, the transformer algorithm, that hit. And now they're competitors." The hardware lesson is that compute investment is necessary but not sufficient.
IBM's India team works on the Granite open model series, agentic middleware for enterprise AI deployment, and Watson X code assistant products Eye on AI Podcast. The company also participates in the AI Alliance, an open-source consortium that includes IIT Bombay, IIT Madras, and IIT Kharagpur alongside global partners.
What Singhee was describing is the gap between two questions the Stanford data conflates: whether India can absorb AI, and whether it can build AI. The nervousness number suggests the country's workforce is uncertain of the answer. The brain drain number suggests the researchers who might build are not staying to find out. The government is spending on hardware. The question of who will run it is one the GPU count does not answer.
What to watch: whether the IndiaAI Mission's infrastructure buildout produces measurable talent retention signals in the next Stanford index cycle, or whether the building arrives before the people meant to fill it.