The Applause Trap

Last week, Kiran Mazumdar-Shaw posted a prediction on X: “Mark my words: What Chinese tech & AI cos did to disrupt the digital tech world over the last decade, Indian tech & AI cos will do over the next decade to disrupt and deploy at scale.”¹

It got thousands of likes. It will change nothing.

Two months before Shaw’s tweet, India’s IT Minister Ashwini Vaishnaw stood at Davos 2026 and rejected the IMF’s assessment. The IMF had placed India outside the top tier of AI-ready nations. Vaishnaw insisted India “clearly” belongs in the first league. He cited the country’s 38,000-GPU national pool and progress across “all five layers” of the AI stack.² The Stanford AI Index 2026, published three months after Davos, would show something else. India had zero presence in notable model production, the world’s largest AI talent outflow, and national compute capacity below the report’s measurement threshold.³ The minister’s optimism at Davos got global media coverage. The Stanford data got footnotes.

I am from India. My career has been in Europe. I have seen how things get built in both, and how they get talked about. The gap is not closing. It is widening.

And the talking is making the building harder.

The pattern

A recurring sequence in Indian tech discourse keeps showing up with uncomfortable regularity. It goes like this:

Someone does real, incremental work. A paper gets published. A product ships. A small company solves a genuine problem. Then a prominent voice reframes that work as evidence of Indian genius, strips it of context. Skips past the hard remaining problems, and collects applause. The original workers get brief attention. The structural problems that constrain them remain untouched. And the public moves on to the next prediction.

I watched it happen in real time last week with a Bangalore startup called Dognosis. They published a Phase-2 study in the Journal of Clinical Oncology on using trained dogs to detect cancer from breath samples.⁴ Genuine science. Published in the world’s top cancer journal. A ~$2 per test aspiration for cancer screening in low-income settings. Worth celebrating.

Then the reframing arrived. “This kind of jugaad wouldn’t have been possible in any other country,” wrote one commentator with a verified account.⁵ His post got 10,000 views.

Except that the paper’s own author list tells a more complicated story. The “conception and design” credits include Claire Guest and Rob Harris from Medical Detection Dogs, a UK charity that has been pioneering canine cancer detection for over twenty years.⁶ They collaborated on this study. To be fair, Dognosis brought genuine innovations of its own: a Bayesian fusion framework that integrates readings from many dogs, and a camera-based behavioural analysis system. This is real engineering, not mere replication.

But the methodological foundations are British. Medical Detection Dogs has been publishing canine cancer detection research since 2004.⁶ And the corporate structure tells its own story. Dognosis is incorporated as a US entity (Dognosis, Inc.), with Indian operations running through a subsidiary.⁷ The patent on canine brain activity decoding is assigned to the US company. Delaware incorporation is standard practice for startups seeking venture capital, but that is the point. The equity value capture sits in the US. The foundational method comes from the UK. India provided the execution environment: six hospitals in Karnataka, a large patient pool, lower costs.

This is collaborative work with asymmetric value capture. Calling it uniquely Indian “jugaad that wouldn’t have been possible in any other country” erases the twenty years of British foundational science and the US corporate structure that holds the IP and captures the equity value.

I wrote a response pointing this out. It got 30 views.

The pattern beyond tech

The pattern extends beyond tech. Praveen Mohan, a YouTuber with over 2.4 million subscribers, travels to ancient Indian temples and claims they demonstrate advanced technology that predates modern invention. Telescopes carved in 900-year-old temples, evidence of extraterrestrial influence, machining precision that contemporary tools cannot replicate.¹⁸ His videos regularly attract millions of views.

Indian fact-checkers and historians have debunked his most viral claims. Mohan presented a “2,000-year-old bicycle” carved at Panchavarna Swamy temple as proof Indians had invented the bicycle millennia before Europe. Dr. R. Kalaikovan, a Tamil Nadu historian, thinks the carving was added during a temple renovation in the 1920s, when a bicycle was still a novelty in the city: perhaps the sculptor had seen one and recorded it in stone.¹⁹ While claiming to “decolonise” Indian history, Mohan projects Western technological concepts onto indigenous achievements rather than understanding them on their own terms. He strips genuine Indian architectural marvels of their actual context to feed a narrative of lost advanced civilisation.

The correction articles get a few thousand readers. Mohan’s claims get millions. The applause ratio is the same as in tech discourse.

The damage is structural. India built some of the world’s most sophisticated temple complexes using mathematical principles, astronomical knowledge, and engineering techniques that were genuinely advanced for their era. That achievement required no embellishment. But celebrating invented capabilities is easier than studying real ones. The hype makes the scholarship harder.

Labour arbitrage, all the way down

If the Dognosis example were isolated, it would not be worth writing about. But it is not isolated. It is a structural pattern that runs through Indian tech, Indian research, and now Indian AI discourse. And it has a name that Indian tech people should recognise, because they built their careers on it.

Labour arbitrage.

For thirty years, the Indian IT services industry ran on a simple proposition: take work designed and architected in the US or Europe, execute it with talented but cheaper Indian engineers, capture the margin on the cost differential. Infosys, TCS, and Wipro became some of the largest companies in the world this way. But India did not become a software product economy because of them. The value capture stayed upstream. The architectural decisions stayed upstream. The IP stayed upstream. India provided the labour.

Indian research is now following the same pattern.

Look at the joint doctoral programmes that India’s top institutions run with Australian and European universities. These are well-regarded programmes. The supervisors are competent. The research is technically sound. But the economics tell a different story.

India’s national PhD stipend, set by MHRD, is ₹37,000 per month for the first two years and ₹42,000 after that.⁸ That is roughly $440. In the IIT Madras-Deakin joint PhD programme, students spend three years in India at these rates and one year in Australia at AUD 35,500 per year.⁹ In the IIT Hyderabad-Deakin equivalent, according to the programme’s own press release, Deakin’s contribution while the student works in India is a top-up of AUD 150 per month.¹⁰ One hundred and fifty dollars. For a doctoral researcher.

Now run the numbers from Deakin’s side. A domestic Australian PhD costs roughly AUD 35,500 per year in stipend, or AUD 142,000 over four years. Through the joint programme with IIT Hyderabad, they get three years of research at Indian rates (their cost: about AUD 5,400 in top-ups) plus one year in Australia (AUD 28,600).¹⁰ Total: roughly AUD 34,000. Same research output. Same co-authored papers. A joint degree badge. At twenty-four percent of the cost.

This is not collaboration between equals. It is offshoring knowledge work, wrapped in the language of international partnership.

The pipeline that was never built

Sidu Ponnappa published a rigorous analysis titled “True Frontier Labs in India,” laying out the capital, compute, and talent gaps between India and the AI leaders.¹¹ The numbers he found are devastating. India’s entire IndiaAI Mission budget is $1.25 billion over five years. A single US hyperscaler spends $528 billion on AI in one year. India’s national GPU pool is 38,000 units. A single private US lab, xAI’s Colossus, has over a million.

But Ponnappa frames the talent problem as brain drain: talented people leaving India for better opportunities abroad. This is true but incomplete. It misses the more fundamental issue.

India does not have a research pipeline for frontier AI. It is not that the pipeline leaks. It is that the pipeline was never built.

The Stanford AI Index 2026 makes this painfully clear.³ Between 2014 and 2025, the academic institutions that produced the most notable AI models were Stanford (26), Tsinghua (26), Carnegie Mellon (25), UC Berkeley (20), University of Washington (19), Oxford (17), and MIT (15) (Chapter 1, Section 1.1, Figures 1.1.6 and 1.1.7). The number of Indian institutions on that list is zero. No IIT. No IISc. No IIIT.

IIT Madras gets exactly one mention in the entire 425-page report, and it is for running a crowd-sourced evaluation tool for Indian language models. Not for producing frontier research.³

India does appear prominently in one ranking. It has 50,460 “top AI authors and inventors,” making it the second largest contributor globally behind the United States.³ This sounds impressive until you look at the education breakdown.

Only 18.5% of India’s AI authors hold PhDs (Section 1.8, Figure 1.8.3). Compare that to the UK (51.1%), Australia (50.5%), the United States (42%), or South Korea (42.5%).

Nearly 34% of India’s “top AI authors” hold only a bachelor’s degree. In a system where many Indian universities need publication for course grades, a significant share of these papers are mandatory term papers, not frontier research contributions.

Three compounding failures

The talent challenge is three problems that multiply rather than add.

First, the pipeline does not produce frontier researchers. India’s R&D spending is 0.64% of GDP, compared to China’s 2.68% and the US’s 3.45%. But even if you tripled the spending overnight, you would not create multi-generational research clusters. Tsinghua’s KEG lab had fifteen years of pre-training research before the LLM era began. No Indian institution has an equivalent lineage. You cannot buy institutional depth. It has to be grown.

Second, it is not just that people leave. It is adverse selection. The researchers capable of frontier work are precisely the ones who leave for US and European labs offering packages of $1 million to $200 million, world-class compute, and dense peer clusters. Those who stay are not operating at the frontier of either applied or fundamental research. This is not a criticism of individuals. It is a statement about a system that pays its top PhD researchers less than a junior taxi driver earns.

The Stanford data confirms this: India has the world’s largest net outflow of AI talent at -16.9 in 2025.³ Over 10,000 top Indian AI professionals moved to the US between 2019 and 2024. Zero senior frontier-lab researchers from OpenAI, Anthropic, DeepMind, Meta, or xAI have moved to an Indian lab. Not one.

Third, no international talent flows into India either. DeepMind draws researchers from UCL, Cambridge, and across the globe. OpenAI pulls from every continent. Indian labs work with a homogeneous talent pool, and the Stanford report carries no data on significant researcher inflows. The absence is itself data. Frontier breakthroughs correlate with diverse teams bringing different training traditions and methodological instincts. A homogeneous pool, however large, has structural limits on the frontiers it can push.

These three failures compound. Even if you solved capital and compute tomorrow, you would still lack the researcher clusters, the institutional culture, the international cross-pollination, and the domestic career paths that would make staying or returning rational.

The Indic AI distraction

The most politically charged example of the hype pattern is the claim that India is building “sovereign AI” through Indic language models.

Here is what is happening. Companies like Sarvam and initiatives like BharatGen are taking open-source foundation models (LLaMA, Mistral) and fine-tuning them on Indian language data. This is useful work. India’s linguistic diversity is genuinely underserved by global models, and local language AI has real applications in healthcare, agriculture, and government services.

But it is not frontier research. Fine-tuning an open-source model on regional language data is applied adaptation, not architectural innovation. It is the AI equivalent of localising a software product for a new market. Valuable, yes. Frontier, no.

In theory, applied work can build the experience and institutional muscle that eventually leads to frontier research. I genuinely believe that pathway exists. But in practice, it rarely happens in India. The pattern is more predictable: early success generates hype, the hype becomes the product, and the hard pivot from adaptation to invention never arrives because the incentive to attempt it has been diluted by premature celebration.

The automobile industry provides the cautionary tale. Tata acquired Jaguar Land Rover in 2008. Mahindra acquired Pininfarina, the legendary Italian design house, in 2015.¹² Both acquisitions were celebrated as proof that Indian industry had arrived at the global frontier of automotive engineering.

Nearly two decades later, the technology and architecture still flow from the acquired companies to the Indian parent, not the reverse. Tata’s 2023 SEC filing shows it is licensing JLR’s Electrified Modular Architecture for its own EVs, not the other way around.¹³ According to Mahindra’s own website, one of Pininfarina’s first projects for the group was redesigning the car dealership experience.¹⁴ The Battista hypercar, Automobili Pininfarina’s flagship, runs on Croatian Rimac powertrain technology.¹⁵ Tata and Mahindra’s domestic vehicles remain mass-market products.

The frontier capability was acquired but never absorbed.

This is what happens when the culture rewards the announcement of capability over the patient, unglamorous work of building it. And it is exactly what is happening with Indic AI.

The damage comes when this work is reframed as evidence that India is competing with the US and China in AI. It is not. It is using their outputs. And the reframing diverts policy attention from the structural investments that might close the gap over decades: compute sovereignty, PhD pipeline reform, research institution building.

And a contradiction

The Stanford AI Index 2026 contains an internal contradiction that captures the hype problem perfectly.³

In Section 1.8 of Chapter 1, the data shows India’s net AI talent flow is -16.9, described as “the largest net outflows” of any country measured (Figure 1.8.6). Then in Section 8.3 of Chapter 8, the same report states: “India is transitioning from a net exporter to a net absorber of talent.”

These two statements appear in the same document, sixty pages apart. The raw data says India bleeds more AI talent than any other country. The narrative framing says India is turning the corner. Even credible international institutions find it hard to resist a more positive framing. The numbers say one thing. The narrative says another.

If Stanford does it, imagine what happens when the incentives are commercial or political.

What I am not saying

I am not saying India has no talent. It clearly does. The fact that Indian-origin researchers lead some of the world’s most important AI labs is proof of the raw material.

I am not saying applied AI is worthless. India’s strength in services, its digital public infrastructure (UPI and Aadhaar are genuine, world-class achievements), and its population-scale deployment capability are competitive advantages.

I am not saying India should try to match the US or China dollar for dollar. That is neither possible nor necessary.

What I am saying is that hype actively harms India’s chances of building AI capability. Every time a prominent voice claims India will disrupt AI in a decade without specifying a single structural change that would make that possible, they make it easier for policymakers to substitute announcements for action. Every time derivative work is reframed as frontier innovation, it reduces the pressure to fix the PhD pipeline, the research funding, the compute infrastructure, and the institutional incentives that keep India in the labour arbitrage trap.

Shaw’s tweet will be forgotten in a week. The structural gaps it papers over will persist for decades.

Global capital has already noticed. On 28th April 2026, Ruchir Sharma, Chairman of Rockefeller International, said at Express Adda: “Most foreigners have taken the view that India is a loser in the AI race.”¹⁶ He called India an “anti-AI play” for global investors. In his thirty-year investing career, he said, he has “never seen such indifference towards India.” Foreign investors have withdrawn over ₹1.8 lakh crore from Indian equities in 2026 alone.¹⁷ The IT services boom of the 2000s gave India a clear participation route. The current AI wave has not. The hype at Davos in January did not convince the people who allocate capital for a living.

What would honesty look like?

An honest Indian AI strategy would start by admitting where India stands. Not second in the world. Second in volume, with 18% PhD depth, the world’s largest talent outflow, zero presence in notable model production, and national compute capacity below the Stanford report’s threshold for measurement (Section 1.2, Chapter 1).

It would ask why Indian institutions accept these asymmetric arrangements in the first place. The answer is not complicated. Prestige and resource constraints reinforce each other. A joint degree with an Australian university boosts an IIT’s international ranking metrics. The foreign partner provides access to equipment and techniques that Indian labs cannot afford. The MHRD stipend rates are set nationally. Individual institutions cannot compete on compensation even if they wanted to. The system is designed to produce volume and international brand association, not frontier talent. Everyone behaves rationally within these incentives. The incentives are the problem.

From that honest baseline, a few things follow.

Own the applied strategy explicitly. Stop comparing to China’s frontier push. India’s services DNA, digital public infrastructure, and population-scale deployment capability are genuine strengths. UPI processes over 14 billion transactions per month. Aadhaar covers 1.4 billion people. These are world-class systems built with engineering depth. Build on them without pretending they are something they are not.

Fix the talent economics, and be specific about the cost. India currently has roughly 200,000 PhD researchers across all disciplines. Raising MHRD stipends to ₹100,000 per month (on par with lower-tier European doctoral salaries) would cost approximately ₹24,000 crore per year, or about $2.8 billion. That is less than 0.1% of GDP and roughly twice what India currently spends on IndiaAI Mission over five years. It would not solve everything, but it would change the signal: that India values research talent enough to pay for it.

Invest in institutional depth, not announcements. This takes decades, not budget cycles. But it is not impossible. AI4Bharat at IIT Madras, despite its limited resources, has built a genuine research programme around Indian language AI that produces usable tools, not just papers. It is a small example, but it shows what focused, sustained effort can achieve when the goal is building capability rather than collecting applause. Can India replicate that patience across dozens of institutions for fifteen to twenty years without the hype cycle destroying focus?

Behind every prediction sits a structural change that would have to happen for it to be true. Most predictions do not name one. The applause does not seem to mind.

Sources

1. Kiran Mazumdar-Shaw (@kiranshaw), April 30, 2026.

2. Ashwini Vaishnaw at World Economic Forum, Davos, January 21, 2026 (“India clearly in the first group”).

3. Stanford University, “AI Index Report 2026,” April 28, 2026.

4. Kulgod, A. et al., “Canine Olfaction Combined With Bayesian Modeling for Multicancer Detection From Breath Samples: A Phase II Study in India,” Journal of Clinical Oncology, April 2026.

5. Mayukh Panja (@mayukh_panja), April 29, 2026.

6. Dognosis press release on JCO publication (confirms Medical Detection Dogs UK collaboration).

7. Dognosis, Inc., Science & Technology page (US patent disclosures, corporate structure).

8. IIT Hyderabad, Department of CSE, PhD fellowship details (MHRD rates: ₹37,000/₹42,000 per month).

9. IIT Madras-Deakin University Research Academy, Joint PhD Programme.

10. IIT Hyderabad-Deakin University Joint Doctoral Program, stipend and programme details (confirms AUD 150/month top-up at IITH, AUD 28,600/year at Deakin). PRNewswire, April 12, 2021.

11. Sidu Ponnappa, “True Frontier Labs in India: What do we need to make it happen?”.

12. Mahindra acquisition of Pininfarina S.p.A., December 2015.

13. Tata Motors / JLR, MoU on EMA platform licensing, November 2023 (SEC filing).

14. Mahindra & Pininfarina partnership details (dealership concept, design collaboration).

15. Automobili Pininfarina (Rimac powertrain, Battista production).

16. Ruchir Sharma at Express Adda, April 28, 2026 (“Most foreigners have taken the view that India is a loser in the AI race”).

17. BusinessToday report on Sharma’s remarks and foreign investor outflows.

18. Praveen Mohan YouTube channel statistics (2.4M subscribers, ~504M total views).

19. Jignesh Patel, “Invention of bicycle credited to ancient India based on misleading photos,” Alt News, November 23, 2019 (Dr. R. Kalaikovan theorises that the carving dates from a 1920s renovation). The first-hand account is Nahla Nainar, “Bringing the past to the present,” The Hindu, June 12, 2015: Kalaikovan found the carving in 1980 and “theorised” the 1920s date.