Building India's Foundational AI Model - Is it really necessary?
A candid opinion by a super early-stage Indian founder
Post ChatGPT launch, I think DeepSeek-R1 is the only model that shook the tech world and hailed as AI’s sputnik moment as they found a way to train model in 1/30th computational cost. It uses Mixture-of-Experts (MoE) Architecture and Auxiliary-Loss-Free Load Balancing technique. In layman terms, it activates only 37B parameters per query out of 671B parameters i.e, only necessary/ relevant parts of the model get activated than the entire model, which led to 95% reduction in GPU usage. As a result, NVIDIA lost $600B market cap, the biggest ever for a U.S. company. The funny thing is- DeepSeek used tens of thousands of NVIDIA H100 & thousands of H800 AI GPU’s.
Let’s put DeepSeek, Stock crashes, US market etc aside for a moment. DeepSeek is a Chinese foundational AI model. This aha moment started a heated debate in India about India’s original foundation layer AI model. I’ve seen a few Indian CTO’s/ Founders tweeting on X inviting Indian tech diaspora to join the mission of creating India’s 1st foundation layer AI model. That’s a good sign & I, as a founder building application layer AI company AIspire Labs, would crave to build on top of India’s foundational AI layer than Google/ OpenAI/ DeepSeek.
The real deal isn’t it. I slept over AI & started digging deeper asking a question - “Is it really need of the hour for India to build a foundational model?”. There’s no doubt that India can build its foundational model in not more than 18 months. By the time, the world could have found another breakthrough/advancement and India’s AI chapter gets questioned again and the never-ending vicious loop of comparison and chasing US, China or other continues.
So,
What should India focus on?
What makes us a global leader in AI?
Is it just limited to thoughts (or) Do I really have a plan?
Yesssss, I really seem to have a plan to make India - A true global AI leader!
Foundational Thoughts:
Let’s go back to the real question- What exactly is a foundation layer AI model? A breakthrough happened with google publishing its landmark research paper “Attention Is All You Need” introducing a new deep learning architecture “Transformer” (Heart of present day LLM’s). They predict the next word, and it changed the world.
I look at the present day LLM’s as Generalist AI models. I’m guessing the next biggest breakthrough will definitely be not a generalist AI but a specialist/ expert AI (I’m not talking about fine-tuning layer/ vertical LLM’s). I believe it could really disrupt (not just displace) the multi-centenary civilizational brain wiring of humans.
Healthcare, Finance, Commerce, Education, Entertainment, Travel and more industries/ sectors, Expert Layer AI is need of the hour to achieve AGI (achieving AGI as Sam is claiming won’t be possible with the current generalist systems). And, what’s important for the expertise? DATA-DATA-DATA-DATA. Facebook built its own foundational model Llama and Google launched its own model Gemini as they own data (social media & search). You may question me about Mistral, Claude or GPT other foundational layer AI models data sources - but they have controversies (or) no mentions about the data sources.
Data (& RealTime Data) is 1000X super value and happy India realized it and drafted stricter Data Governance policies. But we should realize our strength of data available in/of India and that leads us into building next breakthrough - World’s 1st Expert Layer AI Model from India.
Potential Areas for creation of Expert Layer AI Models in India:
Expert Layer AI for Government- Unique Identification Authority of India- The world’s largest data management project. I’m not sure about how much control the government exerts on its data usage but there exists a huge opportunity to build an Expert Layer AI for governance. This Expert Layer AI for governments trained on UIDAI data could help efficiently serve the people of India - whether it’s delivery of welfare schemes to eligible candidates, government tender management and processing, Income tax payables and infinite possibilities opens up and save billions of dollars of tax-payer spending from going wasted without creating DOGE (Department of Government Efficiency) like system in India. OpenAI realized this possibility and partnered with US government to separately build and manage government functions and so launched ChatGPT Gov for modernization of public services in US.
India has successfully managed world’s largest data management project, and I believe there’s an opportunity to build Expert Layer AI for Government of India and can scale it globally as Expert Layer AI for Government as a Service.
Expert Layer AI for Finance- Unified Payment Interface- the world’s most advanced payment system built by NPCI with 443 million transactions taking place in a day. This is huge. I believe there’s a possibility to build an Expert Layer AI for Finance to help us strategize the personal investments, payment memory, accounting for companies and lot more with infinite possibilities. It learns from our finances, executes a strategy and only human approval needed.
If we can leverage our leadership in UPI to build expert layer into finance, I strongly believe we could generate few trillion dollars of value globally with our Expert Layer AI for Finance.
Expert Layer AI for Health - I’m not yet sure about the origin/ status of National Health Interface but I believe this is the most sensitive data that directly impacts human lives. I believe if we build an Expert Layer AI for Health that wins human trust, we can announce AGI is achieved officially! We can build a super strong recommendation system about health, medicines, doctors, appointments and a personal doctor to every citizen of world would be possible with this creation of Expert Layer AI for health. This is super complicated, but we can start building in this health immediately. The possibilities are really mind-blowing.
There are a dozen usecases where India can build Expert Layer AI with our own data for the world. The idea is to build 100X of Steve Jobs knowledge for entrepreneurs, 100X of Einstein for researchers, 100X best doctor to the world, 100X of best financial maverick to every citizen etc! Expertise is what drove human civilization forward and this Expert Layer AI will disrupt the way humans think, work and live the life.
But, yeah, we should build our own foundational model as soon as possible to not let Indian’s real-time data go for other countries. It should be done to protect our country, but creation of Expert Layer AI should be achieved to regain our global technology leadership after many centuries.
Reminder: Expert Layer AI is the only way to achieve AGI. Let ‘s go, India.
I would love to end this by remembering the most powerful words by Steve Jobs.
Don't be trapped by dogma - which is living with the results of other people's thinking. Don't let the noise of other's opinions drown out your own inner voice.
Life can be much broader once you discover one simple fact: Everything around you that you call life was made up by people that were no smarter than you and you can change it, you can influence it, you can build your own things that other people can use. Once you learn that, you'll never be the same again.
Let’s build Expert Layer AI from India.
Notes: I know there are huge resource limitations (computing, funding, talent etc). But I believe expert layer AI is the only option for us to lead this general-purpose technology.
I would love to hear your thoughts/ opinions/ discussions/ collaborations etc. Just drop me a mail at - ajreddy.start@gmail.com.
Happy building :)