Outstanding article. The only elements I would add is a broken justice delivery system and insufficient legal safeguards to right of speech. These are necessary conditions to any story, and perhaps the AI one specifically.
An insightful article. My view is that the Indian government should exit the business of directly funding startups and instead focus on fixing the fundamentals of the ecosystem, such as ease of doing business, expanded capital sources (e.g. convertible debt), reliable infrastructure access including electricity, quality of research and talent, and overall quality of life. The reasons are straightforward:
1) As you noted, previously announced programs have delivered limited results.
2) There is no shortage of private capital globally for strong ideas and capable founders.
3) In a four trillion dollar economy that is now the fourth largest in the world, innovation should be driven primarily by the private sector through sustained R&D and reinvestment of profits rather than dividends and stock buybacks.
In the current environment, positioning the government as a savior and relying on micromanagement is a recipe for stagnation.
I agree that they shouldn’t be in venture funding. And they should make their capital mobility and tax rules better to allow private capital to flourish.
The DISCOM problem and the Tiger Global tax ruling sit at the heart of why the 10,372 crore IndiaAI Mission will underperform its ambitions. Compute without reliable power is unusable. Startup capital without predictable tax outcomes will not come in the volumes needed. The pattern is consistent: India builds the policy announcement first and the implementation infrastructure second, sometimes never. The one structural shift that could change this is the account aggregator framework. If it scales the way UPI did, it gives India a data advantage that no other country can replicate quickly. That is the version of India AI that is genuinely defensible at a global level.
Eloquently articulated. Hope it reaches the right audience & has an impact just like your advocacy which found its way in right sizing AI rules and not constraining them in typical bureaucratic fashion.
Fantastic piece. Reads like the chapters of AI readiness playbook for India.
Really clarifies that India’s AI bottlenecks are not model gaps (as much) but state‑capacity gaps - power, fabs, land, taxes and that without fixing those, the ‘AI buffet’ is tempting but will stay to be out of reach.
The author's analysis is a vital 'reality check' for the AI Summit’s high-gloss ambitions. The core tension is clear: India has developed a sophisticated, innovation-first regulatory framework, yet it remains tethered to a 20th-century physical and fiscal reality.
To truly 'move the needle,' the government must address three critical 'heartburn' factors:
The Energy-Compute Gap: We cannot have 'AI Sovereignty' without baseload power. While the SHANTI Act is a landmark step toward nuclear energy, its success depends on the state’s execution capacity—which historically lags. Until we fix the DISCOM rot and the political economy of electricity pricing, the 21-year tax holiday for data centers remains a solution to only half the problem.
The Capital-Tax Paradox: There is a glaring disconnect between MeitY’s promotion of startups and the Tax Department’s aggressive, retrospective stance on exits, highlighted by the Tiger Global ruling. Private venture capital seeks predictability; without it, India will remain dependent on government 'compute-for-equity' grants rather than the agile, high-scale private capital that fuels global leaders like OpenAI.
The Execution Bottleneck: From the literal 'soft soil' under the Dholera fab to the 3,000-hour compliance burden for subsidies, the gap between announcement and implementation is our greatest risk.
India’s true competitive edge lies in its talent for enterprise integration and 'frugal' model development—like Sarvam’s niche-leading performance. However, for this talent to scale, the government must move beyond 'Wedding Buffet' policy-making and commit to the boring, iterative work of cross-ministry and state-level structural reform. Without it, we are simply building digital skyscrapers on clay foundations.
It's a great article with food for thought for everyone involved and interested!
Enjoyed reading this! As a side note, it reminded me of Ian Fleming's flashy James Bond thrillers v/s John le Carre's "boring" bureaucratic depiction of the spy business
I can barely fathom the amount of research and inherent multi-sectorial understanding that has gone into this, and can only read in complete awe. Thank you for the balanced tone, and especially for making it comprehensible to the layperson.
Excellent analysis—you've done what tha AI experts haven't: look at the actual binding constraints rather than the headlines.
Your point about India's advantage being in enterprise integration & adoption rather than foundational model training maps directly onto something I've been thinking about. I recently wrote about why the US and China are racing toward different AI finish lines—America optimizes for engagement (AI girlfriends, viral ChatGPT moments, Moltbot), China optimizes for deployment in the physical economy (factories, hospitals, power grids).
India doesn't fit neatly into either track & I think that's an advantage. You're right that India's got talent specifically in the unglamorous integration work that makes AI actually useful inside messy enterprises. But there's another piece you touch on but don't fully develop: India has unique data infrastructure that can't easily replicate.
UPI + Aadhaar + the account aggregator framework creates a consent-based data sharing architecture. This is digital public infrastructure—it's also a data moat for training models on financial inclusion, fraud detection, voice-based interfaces in 22 languages & use cases that emerge only when one is forced to serve people at rupee-scale with zero-bandwidth tolerance.
The constraint you identified—frugal innovation at massive scale—is exactly what makes India the testing ground for "everyday AI" that could serve as a template for other poor countries. When you write that "deploying AI in the real world will require vast amounts of integration work, sandboxing for liability, niche customization," you're describing India's comparative advantage. But you're also describing why India might end up defining what "useful AI" means for most of the world, not because of better models but because of bigger constraints.
Your diagnosis of the political economy gridlock is spot-on. The question I ask: Can messy, hybrid, unscripted experimentation (India's 'advantage') produce better long-term results than either state coordination (China) or venture capital concentration (US)? The DPI story suggests yes. The DISCOM story suggests no. We'll soon find out which pattern will prevail.
Every time I have to use an Indian credit card in the US I’m reminded of how broken policy making is in India. The overbearing requirement to OTP, not allowing subscription payments (think that has changed a bit now) reeks of being afraid of technology and an illiberal approach to markets.
I’m worried we’re seeing the same in AI as we’ve seen with payments (eg. the risk of deepfakes being the tentpole of the policy, and missing other larger AI safety or socio-economic implications)
Outstanding article. The only elements I would add is a broken justice delivery system and insufficient legal safeguards to right of speech. These are necessary conditions to any story, and perhaps the AI one specifically.
100%; backlog of cases needs to be displayed in a big billboard where all indians can have a look at like a score board only here less is more!
An insightful article. My view is that the Indian government should exit the business of directly funding startups and instead focus on fixing the fundamentals of the ecosystem, such as ease of doing business, expanded capital sources (e.g. convertible debt), reliable infrastructure access including electricity, quality of research and talent, and overall quality of life. The reasons are straightforward:
1) As you noted, previously announced programs have delivered limited results.
2) There is no shortage of private capital globally for strong ideas and capable founders.
3) In a four trillion dollar economy that is now the fourth largest in the world, innovation should be driven primarily by the private sector through sustained R&D and reinvestment of profits rather than dividends and stock buybacks.
In the current environment, positioning the government as a savior and relying on micromanagement is a recipe for stagnation.
Again, a great article!
I agree that they shouldn’t be in venture funding. And they should make their capital mobility and tax rules better to allow private capital to flourish.
Shruthi Madame,
Your article is profoundly researched - each line reflects your remarkable standard. I thoroughly enjoyed the AI
Wedding Buffet.
SRINATH GR.
Thank you for your kind words.
The DISCOM problem and the Tiger Global tax ruling sit at the heart of why the 10,372 crore IndiaAI Mission will underperform its ambitions. Compute without reliable power is unusable. Startup capital without predictable tax outcomes will not come in the volumes needed. The pattern is consistent: India builds the policy announcement first and the implementation infrastructure second, sometimes never. The one structural shift that could change this is the account aggregator framework. If it scales the way UPI did, it gives India a data advantage that no other country can replicate quickly. That is the version of India AI that is genuinely defensible at a global level.
Eloquently articulated. Hope it reaches the right audience & has an impact just like your advocacy which found its way in right sizing AI rules and not constraining them in typical bureaucratic fashion.
following the path of anne krueger! she would be proud, shruthi! what’s krueger’s take on the EU auto deal? can you bring her on again, PLS ?!?!
Fantastic piece. Reads like the chapters of AI readiness playbook for India.
Really clarifies that India’s AI bottlenecks are not model gaps (as much) but state‑capacity gaps - power, fabs, land, taxes and that without fixing those, the ‘AI buffet’ is tempting but will stay to be out of reach.
The author's analysis is a vital 'reality check' for the AI Summit’s high-gloss ambitions. The core tension is clear: India has developed a sophisticated, innovation-first regulatory framework, yet it remains tethered to a 20th-century physical and fiscal reality.
To truly 'move the needle,' the government must address three critical 'heartburn' factors:
The Energy-Compute Gap: We cannot have 'AI Sovereignty' without baseload power. While the SHANTI Act is a landmark step toward nuclear energy, its success depends on the state’s execution capacity—which historically lags. Until we fix the DISCOM rot and the political economy of electricity pricing, the 21-year tax holiday for data centers remains a solution to only half the problem.
The Capital-Tax Paradox: There is a glaring disconnect between MeitY’s promotion of startups and the Tax Department’s aggressive, retrospective stance on exits, highlighted by the Tiger Global ruling. Private venture capital seeks predictability; without it, India will remain dependent on government 'compute-for-equity' grants rather than the agile, high-scale private capital that fuels global leaders like OpenAI.
The Execution Bottleneck: From the literal 'soft soil' under the Dholera fab to the 3,000-hour compliance burden for subsidies, the gap between announcement and implementation is our greatest risk.
India’s true competitive edge lies in its talent for enterprise integration and 'frugal' model development—like Sarvam’s niche-leading performance. However, for this talent to scale, the government must move beyond 'Wedding Buffet' policy-making and commit to the boring, iterative work of cross-ministry and state-level structural reform. Without it, we are simply building digital skyscrapers on clay foundations.
It's a great article with food for thought for everyone involved and interested!
Enjoyed reading this! As a side note, it reminded me of Ian Fleming's flashy James Bond thrillers v/s John le Carre's "boring" bureaucratic depiction of the spy business
I love that juxtaposition.
I can barely fathom the amount of research and inherent multi-sectorial understanding that has gone into this, and can only read in complete awe. Thank you for the balanced tone, and especially for making it comprehensible to the layperson.
Terrific article! Most comprehensive.
Excellent analysis—you've done what tha AI experts haven't: look at the actual binding constraints rather than the headlines.
Your point about India's advantage being in enterprise integration & adoption rather than foundational model training maps directly onto something I've been thinking about. I recently wrote about why the US and China are racing toward different AI finish lines—America optimizes for engagement (AI girlfriends, viral ChatGPT moments, Moltbot), China optimizes for deployment in the physical economy (factories, hospitals, power grids).
India doesn't fit neatly into either track & I think that's an advantage. You're right that India's got talent specifically in the unglamorous integration work that makes AI actually useful inside messy enterprises. But there's another piece you touch on but don't fully develop: India has unique data infrastructure that can't easily replicate.
UPI + Aadhaar + the account aggregator framework creates a consent-based data sharing architecture. This is digital public infrastructure—it's also a data moat for training models on financial inclusion, fraud detection, voice-based interfaces in 22 languages & use cases that emerge only when one is forced to serve people at rupee-scale with zero-bandwidth tolerance.
The constraint you identified—frugal innovation at massive scale—is exactly what makes India the testing ground for "everyday AI" that could serve as a template for other poor countries. When you write that "deploying AI in the real world will require vast amounts of integration work, sandboxing for liability, niche customization," you're describing India's comparative advantage. But you're also describing why India might end up defining what "useful AI" means for most of the world, not because of better models but because of bigger constraints.
Link to my piece: https://rajeshachanta.substack.com/p/spectacle-vs-scaffolding
Your diagnosis of the political economy gridlock is spot-on. The question I ask: Can messy, hybrid, unscripted experimentation (India's 'advantage') produce better long-term results than either state coordination (China) or venture capital concentration (US)? The DPI story suggests yes. The DISCOM story suggests no. We'll soon find out which pattern will prevail.
Excellent article. Thank you
Big fan mam,from lumding a small town in assam
Brilliant piece. Appreciate the balanced tone.
Every time I have to use an Indian credit card in the US I’m reminded of how broken policy making is in India. The overbearing requirement to OTP, not allowing subscription payments (think that has changed a bit now) reeks of being afraid of technology and an illiberal approach to markets.
I’m worried we’re seeing the same in AI as we’ve seen with payments (eg. the risk of deepfakes being the tentpole of the policy, and missing other larger AI safety or socio-economic implications)
Great work! With this, Shruti, you do qualify to be an AI expert!
Thanks! But I am not sure I have that much more to say about AI to join the AI expert bandwagon.