Who drives deep-tech innovation?

It is early April 2025, and an important debate has been spark by certain remarks about Indian startups from Mr. Piyush Goyal, the Minister of Industry and Supply of India. Here is what is said 1,

“Are we going to be happy being delivery boys and girls… Is that the destiny of India…this is not a startup, this is entrepreneurship… What the other side is doing – robotics, machine learning, 3D manufacturing and next generation factories,”

This is a very interesting take, and has led to a number of debates. The entire startup ecosystem, big folks like Mr. Mohandas Pai, the usual (useless) influencer community and a lot of other people have shared their comments both for and against this sentiment.

I generally don’t find myself interested in such debates, I just stick to me and my work. But this one is quite interesting, and I believe I have a perspective to share, so thought of writing this article real quick. In this article, I make an attempt to understand who is generally responsible for innovation in different countries across the world, especially such ground-breaking, deep-tech innovation which Mr. Goyal is talking about. Probably at the end of this article we will have a good answer to this question. When I say deep-tech, I don’t refer to just AI and ML related innovation but anything deeply technological - for example, the semiconductors and microprocessors are included in the category too.

1. Who drives deep-tech innovation in the USA?

1.1 Rise of Operations Research

Consider these examples.

1. You enter a McDonalds, you see exactly 3 cash counters, 3 automated ordering systems, some 20 workers in the kitchen doing different things, a highly efficient kitchen facility, a burger pipeline that gets you a burger in 5 minutes. What is this restaurant optimized to do? How many customers can it handle?
2. You are at Zara’s clearance sale, and you want to buy something before it goes out of sale. How much do you pay for a clothing? How much should Zara sell each piece of clothing to maximize its revenue? Or should it maximize its revenue? or should it go for maximizing sales and emptying all its inventory? Who sets these prices?
3. You order that book from the US, it’ll come to your home in 2-3 days. How does this work? Millions of goods are transported each day, billions of dollars worth goods are stored in large inventories and warehouses, tight delivery times are promised not just to corporate customers but also for an average regular customers.
4. I can go on and on.

From a simple restaurant to designing manufacturing facilities to running logistics, what is behind all these? It is Operations Research.

Operations Research (OR) is applied mathematics, in the sense that mathematics is used to model business problems and helps in decision-making. It helps you in setting a certain price for every piece of clothing if you want to maximize revenue, it helps you in coming up with the exact number of cash counters and design of the burger pipeline, or the route your book has to take to reach you within 3 days and so on. I would not go into gory technical details, but know that this is the engine that runs the world (but many of us would not even know this).

Today, you can see that OR is used in many different contexts - facility design, pricing, inventory management, routes & freight planning, transportation, queuing and service-design and so on - I won’t be wrong if I say that all the commercial businesses which make a significant difference in people’s lives use OR today. But when did this even come to happen? Did commercial businesses always use mathematics to model different parts of their business? Was this even common 100 years ago for businesses, business owners and managers to use OR? Let us see.

To know this, let us start with the erstwhile definition of OR. Here is the definition from the US Department of Defense from the previous century3.

The analytical study of military problems undertaken to provide responsible commanders and staff agencies with a scientific basis for decision on action to improve military operations.

Continuing, we come to know there are four main applications of OR in military decision making.

1. the development, testing, and performance evaluation of weapons and other equipment;
2. the design and evaluation of military organizations, tactics, strategy, methods, and policy;
3. the evaluation of human performance and behavior; and
4. the design and evaluation of effective management structures and procedures

This has been taken from a fantastic book on the history of OR. Let us delve into something lighter. Checkout this article2 published on INFORMS on the history of OR. To summarize it, OR was essentially born during the early 20th century, to primarily improve military operations.

OR was in its early phases during WW-1, but it peaked during the WW-2 where it was used in a variety of ways. It was realized that OR is an insanely important tool/technology/innovation that rose during the war times. Post the WW-2, people started applying OR to commercial non-military business use-cases: From military operations to business operations.

After WWII, the concept of applying the techniques from military operations research to business operations flourished in both the U.K and U.S. The transition was smooth due to practices and attitudes that were receptive to types of analysis used by the military O.R. teams.

And from there on, OR, which has rose to its peak in military operations, was used extensively in business operations and decision-making (for the type of ones I listed at first). Today, business operations continue to get increasingly efficient, thanks to OR.

Now, you tell me, who drove the development and usage of Operations Research? It can be said without a doubt that it was the different governments, their defense departments, their army and military departments in collaboration with mathematicians, physicists, chemists who built OR. From a finance point of view, it was government money that helped develop OR. Then different business owners, entrepreneurs, managers came in and started using OR to make their business operations better. Obviously using OR for commercial non-military purposes has given rise to a good deal of innovation in the field of OR, so business owners and management have played a role in active development of OR.

Atleast in the case of OR, the government led the ground-breaking innovation which was then adopted and actively continued by business owners and managers.

1.2 Rise of Semiconductors

Semiconductors play an integral role in our lives. They are essentially in every device we use today. The AI-boom that we are witnessing is made possible because of the advancements happening in the field of semiconductors.

What was the beginning of the semiconductors? The theoretical beginning was again early 20th century, with a number of developments over the decades. But things started to materialize post 1950s when actual companies started to form around semiconductors, when things moved from pure academic research (William Shockley invented the world-changing transistor in 1947) to industry, engineering, business, manufacturing and production. I think if you ask anyone, one would say that it is because of pure capitalism, the efforts of a great number of academicians and companies like Texas Instruments, Integrated Electronics (Intel), the TSMC, Samsung, now Nvidia and visionary businessmen and technophiles who shaped the semiconductor industry for what it is today. But that is only part of the story. Listen to this discussion 4 with Chris Miller, who has authored a fantastic book called “Chip War” 5. The book describes in detail how different governments, their money and policies, wars, nationalism, political and power rivalry shaped the semiconductor industry. At one point, even though these companies (TI, Intel etc.,) were functioning with some customers, it is the department of defense (DoD) of the US government that became their biggest customer. In essence, the government funded these companies so that it could use semiconductors (memory chips, microcontrollers, microprocessors) in military equipment (bombs, missiles, aircrafts, you name it). With that the ground-breaking innovations were made in the field of semiconductors, global supply chains were built to supply cheaper and better chips.

Once the industry matured, a number of consumer-product companies were born which started using semi-conductor chips in commercial (non-military) products. Japan led the world in consumer electronics, where it bought chips from different companies and started making consumer electronics at insanely small prices, again bankrolled by the Japan government (talking about Sony here).

Today, governments do fund semiconductor innovation but companies themselves are big enough to bank-roll a good-deal of research.

1.3 Rise of Internet

While many different academicians performed research in switching theory and such, it is very well known that it was the Defense Advanced Research Projects Agency (DARPA), an R&D wing of the USA’s Department of Defense and the scientists, academicians and engineers in it who essentially invented the Internet. It was known as ARPANET back in the days when it was not a public network but a network for military/defense purposes. As with any technology, it was developed and when it became mature, a number of private players picked up and built companies and entire industries out of it. A number of large telecommunications companies, Networking companies were born. Then Internet usable for general people was born (the WWW/World Wide Web for example).

Not just telecom companies but a number of internet/technology companies were born once the Internet was mature enough - Microsoft, Apple, Google, Facebook, Amazon and so on. These companies cater mainly to consumers and the commercial markets. But as they grew in size, I believe they became quite collaborative with different governments, where governments funded them for different types of technologies.

Today, these private corporations are behemoths themselves and are no less powerful than governments.

Now coming to Internet in India. India has had access to the Internet for quite some time, but it went through a massive revolution in the last decade, thanks to Reliance Jio. This was possible primarily with the support of the government, but the dirt cheap Internet data which we continue to get today is atleast partly bankrolled by the government. If one simply delves into internet companies’ financial statements, it becomes clear that the indian telecommunication companies are piling up massive government debt year on year, standing at over 4 lakh crores rupees as of 2024 6.

1.4 The Rise of AI

The theory behind AI & ML was built by mathematicians, computer scientists in the 20th century itself, but it all materialized with the rise of compute (and networking), both, as we saw were primarily driven by governemnts then actively developed by commercial companies.

I think the development of AI is a bit different from the rest of the examples we saw earlier. I believe it was these large private corporations who actively funded and drove the innovation in the field of AI. If you take a look at history of OpenAI, it as been private corporations, banks and large venture capitalists pouring billions of dollars into it. Even prior to that, it was corporations funding AI research (both corporate and academic) for commercial purposes.

I recently listened to a podcast with one Mr. Ben Buchanan, who was a central advisor Biden’s White House for AI policy-making7. Whatever we have seen in this article is essentially reiterated by him - that major ground-breaking technological innovations previously were driven by the governments but AI is primarily driven by these large private corporations. It is a very interesting discussion, please do listen to it.

All that on one side, I think now AI in USA will get the much needed impetus by the government. President Trump recently announced the The Stargate Project, which is essentially a massive private-sector investment for the development of AI8.

The initiative brings together some of the biggest players in technology and AI. The project begins with an immediate $100 billion investment and plans to expand to $500 billion over four years. SoftBank is led by Masayoshi Son, who will serve as chair and be responsible for financing the venture. OpenAI’s Sam Altman will oversee operational responsibilities, and Oracle’s Larry Ellison will join the leadership team, bringing his expertise in data center infrastructure. The collaboration also includes other tech giants, such as Microsoft, Nvidia, and Arm.

A number of private corporations, all working on the development of AI, with the government stands as a backbone to this and facilitates the process.

I think with that, I have covered major deep-tech breakthroughs in the last 70-80 years.

1.5 What about other nations?

Other competing nations follow a very similar model, primarily China, at the heart of its development story is its government. In the 1960s-70s, it was roughly where India was but today it is rivalling in USA on the technological front. This kind of rapid development within 4-5 decades is simply not possible without the government’s vision and strategy.

2. Sad case of India

Due to several reasons, India remained starkly backward when it comes to deep-tech. The liberalization of our economy post 1990s was a golden opportunity. We first started with building companies that exported software and services to the West but remained as software exporters even in 2025. The big IT corporations like Infosys, TCS, Wipro etc., for whatever reason did not delve into deep-tech when they were the only companies who had the slightest chance at it. Even today, even if these companies want to close shop, the government won’t let it happen because they provide jobs to people and provides a stable middle class to India - both of which are important.

India unfortunately missed a number of deep-tech breakthroughs due to timing, policial reluctance, lack of vision and strategy. You can see how each breakthrough is built on the previous one. The viable manufacturing and production of semiconductors was possible through efficient business processes, logistics and supply-chain. Internet was possible because of specialized/custom hardware/semiconductors for the Internet. AI-ML requires a lot of compute, which is made possible today through Cloud and data-center infrastructure, which is essentially compute, storage and networking at its core. I think it can be said that because government drove initial breakthroughs, supporting the deep-tech companies financially and regulatorily, after a point which the deep-tech companies were capable enough to drive development of AI and with now the USA government roping in different private entities to drive further development of AI. India does not have any of the backbones needed for development of AI (and such deep-tech) today.

There are a number of problems India faces today from a deep-tech innovation point of view.

2.1 Indian VCs are not ready

The Venture Capitalists who fund your typical Indian Startups (which are simple e-commerce aggregator platforms, labor consolidation platforms, fast supply chain platforms) will be ready to invest a maximum a couple of hundred million dollars on companies. It is not in their capacity or intent to invest tens and hundreds of billions of dollars. They are in the business to see certain returns after a reasonable point of time, which is not a deep-tech which is in its nascent phase (from a business model point of view) can promise. This reminds me of the infamous and rather embarassing exchange between Sam Altman and Rajan Anandan, Rajan being a fairly big VC today. Rajan essentially asks if its possible for Indian startups to build foundational models at $10 million (and not say $100 million). I am not sure if this was supposed to be light-hearted humor or if Rajan was asking a serious question. But either ways, I, as an Indian felt very embarassed with Altman’s brutal response. Here is some info on Rajan’s VC fund, PeakXV Partners[10]

Over the last 17 years of our operations in the region, Peak XV has grown to manage over $9bn in capital across 13 funds and invested in over 400 companies, of which ~40 companies have achieved revenues of over $100MM. Our team of professionals work across 5 offices – Bangalore, Mumbai, Delhi, Singapore and Dubai and span 14 nationalities.

From this, I understand where Rajan is coming from, but I feel he should have known where Altman is coming from. OpenAI started with hundreds of millions of dollars, but it has been dealing with tens of billions of dollars but now probably in hundreds of billions of dollars with President Trump’s initiatives. This tells me VCs have a capacity constraint (understandable) but possibly a lack of vision as well.

2.2 Usual reluctance of IT Services companies

The development of Generative AI was first started by a number of big people and private corporations coming together. Unfortunately, I don’t think we can expect anything similar in India. I don’t expect IT services behemoths to contribute to any meaningful development of AI. Further, if Generative AI becomes powerful enough, the software exporting business model these services companies are built on will come under threat. The maximum they will come close to Generative-AI and deep-tech is simply using them in building AI-based solutions and work-flows for customers, which is just IT-services, nothing more than that. Anyway, I actually want to delve into the mindset of the people towards deep-tech who run these IT services companies. Narayana Murthy had given an interview a while ago 12. To be very honest, this was genuinely a painful read for me. It starts like this.

“We have not been able to build large databases, and without big data, AI (artificial intelligence) has no value. A large language model (LLM) doesn’t make any sense. Basically, the Indian mindset is still not oriented towards problem definition and problem-solving,”

He continues with his IT-services mindset.

He argued that India should focus on building solutions atop existing LLMs—an idea supported by prominent technologists like Infosys co-founder and Aadhaar architect Nandan Nilekani and venture capital firm Accel.

and goes on to pat himself on his back for all the IT services jobs he and his contemporaries have created.

“We should all be grateful to the IT services industry; we should all salute them,” he remarked, defending the sector against criticisms that it has not driven innovation.

This line of thought is shared by the current Infosys Chairman Nandan Nilekani as well13

“Our goal should not be to build one more LLM. Let the big boys in the (Silicon) Valley do it, spending billions of dollars. We will use it to create synthetic data, build small language models quickly, and train them using appropriate data,”

Grandeur mindset without wanting to do the actual work.

“The Indian path in AI is different. We are not in the arms race to build the next LLM, let people with capital, let people who want to pedal chips do all that stuff… We are here to make a difference”

Mr. Mohandas Pai, the former CFO of Infosys says something similar14. But he was very honest with his take, which is what I appreciate.

The reason is simple – lack of capital. “Who will give $200 million to a startup in India to build an LLM?” said Mohandas Pai, head of Aarin Capital and former CFO of Infosys, when asked about the lack of innovation from Indian IT. “Why is nothing like Mistral coming from India,” he asked rhetorically.

He continues,

“There is nobody,” he answered. “Creating an LLM or a big AI model requires large capital, time, a huge computing facility, and a market. All of which India does not have,” said Pai.

He opens up completely.

Pai also believes that even if they have enough capital, expecting Indian IT companies to build an LLM is not the right approach for India.

and Instead of services companies, Indian startups and product companies should invest more into R&D, which he agrees is also difficult because investors do not put in so much money. Most Indian AI startups are powered by jugaad, not VC money, which makes it harder for them to put those funds into R&D.

But where are the product-based companies in India, and that too who have the risk appetite and financial capacity like this? Unfortunately, there are almost zero such firms.

I don’t want to take you down a depressing path, but just want to make it clear the systemic problems and lethargic mindsets (in the illusion of grandeur) we are dealing with.

All that aside, I was of the opinion that it will be Reliance who today has the money, power and government support to take up such risky deep-tech development, even though it has not actually done any deep-tech type work previously. But there is a high chance it may not take up any such thing due to some recent developments9. If it just becomes a middleman, it might end up making a lot of money but there is deep-tech innovation involved, very similar to Reliance contracting its entire Jio project to Cisco instead of making any meaningful attempt at innovation in networking deep-tech.

2.3 Lack of vision

I was trying to think who has the vision to carry such deep-tech projects and companies. We do not have big leaders from private corporations, product heads from big firms who have vision and experience in such technologies. Apart from that, it is fair to say that even the meagerly capable people are leaving the country for better pastures. I personally want to believe that we have plenty of academicians who have what it takes to make it, especially the ones at research institutes like IISc, JNCASR and the top IITs. But just having vision is not enough, you need a leader and risk-taker to support such visionaries.

Adding to that, we have startup founders who are into labor consolidation, e-commerce aggregation, fragmented-market aggregation, a whole lot of “we are a middlemen between A & B, and we make money through providing a platform connectin A &B “ - these are the typical business models you see in “big startups” of India. I almost laughed when Urban Company launched its “Maid as a Service” or “Insta Maids” as they call it. Same with BlinkIt delivering iPhones, or Zepto advertising about delivering cars in minutes. I personally see all these as a net negative, highly unsustainable models which are simply burning essential VC money. It is better to think of making their dark-stores based model profitable (which burns cash like anything at the moment) instead of these theatrics. Anyway, those are for another day. People who build companies like Zepto, Lenskart, Urban Company, Boat, BharatPe and so on - do not have vision and depth to build deep-tech companies.

And this lack of vision and intent is shared by parties in previous sections as we saw.

2.4 The government

We have gone over many stakeholders. Now let us come to the government, represented by Mr. Piyush Goyal in this case. His criticism on Indian Startups might make someone think that he is right in his criticism. But the issue here is as we saw in section 2.3, criticising people who mostly cannot build deep-tech companies in the first place is cruel. You are simply putting the blame on them, it is not their job to be very honest. They have built successful or valuable businesses, providing certain services and products, good for them and good for us. But there is no point in criticising them and undermining their efforts.

Mr. Mohandas Pai has been brutally honest when it comes to deep-tech efforts in India. Referring back to the previously seen article[14],

While speaking at Cypher 2024, Pai called on the Indian government to significantly increase its investment in AI. He pointed out that although the Government of India spends INR 90 lakh crore annually, only INR 3,000 to INR 4,000 crore is allocated for innovation, a sum he referred to as “peanuts”.

He continues,

“The Government of India should invest INR 50,000 crore in AI. They need to support AI startups by providing grants and enabling them to invest in and develop solutions like ChatGPT,” he said.

INR 4000 CR is around $450 Million. INR 50,000 CR is close to $6 Billion. Just to put things into perspective, the Congress Party’s welfare schemes in the state of Karnataka will cost around INR 50,000 CR ($6 Billion), and similar scehemes in Madhya Pradesh by the BJP will cost around INR 20,000 CR ($2.4 Billion).

Adding to that,

Moreover, most of the companies and startups in India are focused on building vertical and specialised AI models such as using smaller language models and implementing them for healthcare, legal, insurance, and similar sectors. “They have got their models, they are implementing them and they would be the major players in this market,” added Pai.

and finally,

The only problem with that is that we might have all the Krutrims of the world, but India will not have a Mistral AI to call its own.

Now let us get into what the government and powerful folks have to say. Let us start with Mr. S Krishnan, the secretary of the MeitY (Ministry of Electronics and Information Technology) of India15.

Building a large language model (LLM) from scratch may not be worth the effort, suggested S Krishnan, secretary of the Ministry of Electronics and Information Technology (MeitY) on November 6. Instead, Krishnan advocated for adapting existing LLMs to cater to specific sectors.

He continues,

“It may not be worth the effort to build an LLM on our own, will be instead better to adapt it to different sectors,” Krishnan said at a Microsoft event, while in conversation with Microsoft AI’s CEO Mustafa Suleyman in Bengaluru.

All this aside, the Indian government has setup something called IndiaAI16. Mr. Krishnan talks about that.

During his conversation with Suleyman, Krishnan spoke on the IndiaAI Mission, explaining that the Indian government has put in $ 1.3 bn in to the initiative, $500 mn for access to AI compute and so on.

This initiative will provide folks with datasets, compute and such resources to train and work with LLMs and AI in general. Irrespective of this, I think policy and mindset decides what this compute and datasets are used for, and work on foundational models is not encouraged/incentivized by the government as of now.

2.5 Coming back, who funds deep-tech innovation?

After all this, let us come back to our original question. Who funds deep-tech innovation?

Well, if India was in a similar situation as the USA, large private corporations would have funded it automatically. But we lack such companies. It is through repeated cycles of government support for several previous deep-tech innovations that today these private corporations are able to fund AI, and the government has also pitched in with President Trump announcing the Stargate initiative. These cycles of innovation - where there are small companies with break-through ideas (take Intel or Texas Instruments), the government funds it, becomes their primary customer throughout its formative years, and finally the technology is mature enough to sustain through non-government consumer markets. India did not have these repeated cycles of deep-tech innovation, it is a relatively new country and is still trying to get its house in order. So, today if we are asked the question on who should fund AI or any such deep-tech innovation in India, no VCs, no IT service companies, no such entities can/will do it. The cycle should start with the Indian government - it is the only entity with such huge risk appetite and at times even vision for such technology.

If the Indian government starts the cycle of innovation today which the governments of developed countries have done for decades or even close to 2 centuries now, one day we will have Indian private corporations big enough to fund deep-tech innovation possibly without government support.

2.6 Conclusion

To conclude, I think it is very clear that if we want India to develope and train foundational models, have its own ChatGPT, it has to come through government funding. There is no other way to fund REAL AI Research in India.

Cheers,
Adwaith

References

1. Mr. Piyush Goyal’s comments: https://cio.economictimes.indiatimes.com/news/consumer-tech/ aman-gupta-sides-with-piyush-goyals-comments-on-indian-startups-to-focus-on-deep-tech/120040976
2. History of OR, an INFORMS article: https://pubsonline.informs.org/do/10.1287/orms.2023.03.04/full/
3. History of Operations Research in the US Army: https://ia801009.us.archive.org/18/items/DTIC_ADA468757/DTIC_ADA468757.pdf
4. Chris Miller’s podcast: https://www.youtube.com/watch?v=v5yztfMWXqA&pp=ygUMQ2hyaXMgTWlsbGVy
5. Chip War by Chris Miller: https://archive.org/details/chip-war-by-chris-miller
6. Indian Internet Debt: https://www.deccanchronicle.com/business/economics/india-telecom-debt-crosses-4-lakh-crore-in-fy24-1841837
7. The Government knows AGI is Coming: https://www.youtube.com/watch?v=Btos-LEYQ30
8. Trump’s AI Push: Understanding the $500 Billion Stargate Initiative: https://www.forbes.com/sites/garthfriesen/2025/01/23/trumps-ai-push-understanding-the-500-billion-stargate-initiative/
9. Reuters Article - “OpenAI, Meta in talks with Reliance for AI partnerships”: https://www.reuters.com/technology/artificial-intelligence/openai-meta-talks-with-reliance-ai-partnerships-information-reports-2025-03-22/
10. PeakXV Partners, About Us Page: https://www.peakxv.com/about-us/
11. Narayana Murthy’s comments on LLMs: https://timesofindia.indiatimes.com/technology/tech-news/infosys-founder-narayana-murthy-on-why-india-should-not-build-its-own-ai-models-the-indian-mindset-is-still-not/articleshow/115304666.cms
12. Narayana Murthy’s Interview on LLMs: https://www.moneycontrol.com/technology/indian-mindset-not-oriented-towards-problem-definition-and-solving-infosys-founder-narayana-murthy-on-why-building-an-llm-doesnt-make-sense-article-12867098.html
13. Nandan Nilekani and LLMs: https://www.moneycontrol.com/technology/infosys-chairman-nandan-nilekani-let-the-big-boys-in-the-valley-build-llms-we-will-use-it-to-solve-real-world-problems-article-12850483.html
14. Mohandas Pai on LLMs: https://analyticsindiamag.com/it-services/mohandas-pai-explains-why-india-doesnt-have-a-mistral-ai-yet/
15. Mr. S Krishnan on LLMs: https://www.moneycontrol.com/technology/indias-ai-llm-path-may-lie-in-adaption-not-building-from-scratch-suggests-meity-secretary-s-krishnan-article-12859599.html
16. India AI: https://indiaai.gov.in/