Why now, for the India Semiconductor opportunity? Seed to Silicon part 1

Sudipto:

It became a geopolitical issue where chip became the new oil and every country had to rely on self-capability to build. It says that it’s a 20 billion dollar market. Today we’ll get to 60-70 billion dollars over the next 5, 6, 7 odd years. All the change that’s happening in the world AI EV 5G RISC- V has made it a very interesting time and you have this line. India is both right to play today and we at least internally believe our right to win in the global semiconductor market. As a founder maybe the best thing to do is take up a particular industry use case, build an ASSP which is specific to that industry but can be used across multiple different companies and start with that.

Nivedith:

Hi, everyone. I'm Nivedith and I'm part of the deep tech investing team at Matrix Partners. And today we’ll be talking about the semiconductor opportunity in India. Touted as probably the world’s most complex industry they have a very important role to play in our lives, they power our PCs, our laptops, our phones and even our basic electronics like fans and air conditioners. They’ve been widely talked about of late thanks to all the amazing work that the government has been doing. And today we’ll be diving deep into the industry and opportunities for Indian entrepreneurs. We have Sudipto with us, Sudipto leads B2B commerce and deep tech investing for Matrix Partners and has made a plethora of investments in B2B commerce companies. Some notable ones include Vegrow, Captain Fresh, FarMart amongst many others. I think my first question stems from the introduction itself. Sudipto, we’ve made plenty of investments in traditional supply chain companies and manufacturing companies but we haven’t invested in a semiconductor company yet, why is that?

Sudipto:

Great question to start. First of all, welcome, Nivedith, to the deep tech team. And I agree we’ve done multiple different companies in B2B hopefully a bunch of them become large, we’re super bullish on B2B. But coming to semiconductors the topic for our discussion it’s not that we did not want to it’s just that the opportunity to invest in semiconductors in India probably did not exist 5-6 years back. Let me take a little bit of historical perspective to venture capital investing. In some shape and form the VC industry was created to invest in semiconductors, in the US we have a Silicon Valley, so all started with silicon and semiconductors and investing in cutting edge companies in silicon. Even if you look at Matrix as a franchise Matrix US have been big believers of hardware and semiconductor investing. I don’t know if you can see in the video in the back but we were early investors in Apple, investors in Sandisk, Xilinx, Oculus through the 1980s and ‘90s and more recently also have done a bunch of investments in semiconductors. Same for Matrix China, where the opportunity for semiconductors showed up 3-4 years back. We feel that today finally the time is right to build multiple large companies in semiconductors in India which is why we’re super bullish, we want to invest, which is why we hired you so that time seems to be right.

Nivedith:

Before we did deeper let’s do a few one on one questions. What is a semiconductor and let’s take a relatable example for our broader audience. At the essence of it what is on the piece of silicon and how does it really interact with the broader world.

Sudipto:

So I did electronics years back and I’ll try to answer it in a little bit of simple layman’s term but maybe we’ll start with the fundamental building block which more often than not is a transistor. A transistor is something which acts as a switch or is an amplifier, so that's the fundamental building block of any chip. You take a few transistors together and you start getting logic gates, you can have AND gate, OR gate which is the simple logic functions. You take more of those and you generally get simpler chips called integrated circuits which does certain functions. You take billions of these transistors and put on a silicon sort of wafer and that actually becomes a chip which can do numerous complex computations and sometimes it’s a CPU, it’s a GPU, it’s a microcontroller, it’s a microprocessor depending on the use case. Typically designers of those chip use hardware description language like Verilog to design chips. Sometimes there are existing IQ blocks that they use in that chip but that's what a chip looks like. Now the way it interacts with the real world you write code and say CEC plus python a compiler converts that into a machine language and then the machine language interacts with that piece of hardware through pre-defined instructions at architectures. So that's how a single transistor goes into billions of transistors into a chip and gets convert and sort of your c code gets converted into machine language and then all of that sort of comes together.

Nivedith:

That sounds very complex, Sudipto. But how does it all come together to the extent that an Apple or Dell places the chip into our PC or our laptop, what does this remarkably complex supply chain really looks like?

Sudipto:

So there are different parts to the supply chain, right, to build a chip it’s actually extremely complex because different parts of the chip making process happens at different parts of the world. So most of the – when somebody starts building or designing a chip you don’t design the entire chip from scratch, you take pre-existing IP blocks which do certain functions really well and start from there. Now companies like Synopsis and Cadence specialize in that. Then you take those IP blocks, add your own IP, get the circuit done well and then you create a chip with companies like Nvidia, AMD etc are famous for doing that well. Now that design goes to something called a fab or foundry where it’s taped out on to a real silicon. To tape that out you need extremely advanced machines that uses photolithography which comes from Europe. The tape out typically happens in Taiwan in foundries like TSMC which are very popular which is why Taiwan is also a geo political sensitive area and then it goes to countries like sort of Korea and China for it to get packaged and then gets distributed in the world. So single piece of chip and this is we’re just talking about the chip, the electronics assembly has not started yet. It’s designed in the US, it’s sort of manufactured in Taiwan using machinery in Europe, packaged in China or Korea and then goes to a Foxconn where they’re doing electronics assembly and they put the chip into essentially the phone. So extremely complex supply chain but extremely well oiled and run efficiently over 15-20 years till chip disruption started happening post Covid and which is why semiconductors has become such an interesting sector to invest today.

Nivedith:

So in whatever you talked about in the previous answer I didn’t hear the mention of India anywhere and India so far has had a very limited role to play in the global semiconductor value chain. The best, the largest fabric design companies like AMD and Nvidia are based in the US. Taiwan has the most sophisticated foundries that are manufacturing chips on leading process nodes like 3 nanometer, 5 namometer. But you do mention that there is an opportunity for India today, so what has really changed at the heart of it?

Sudipto:

So first the reason we did not hear about India is the chips were designed in the US putting a lot of cases by Indians. And they could have been in the headquarters in the US or sort of in the same company’s offices in India. So India has had a role to play, Indians have had a role to play but may not necessarily India, so that’s one. I think what fundamentally changed is if you look at it from 200-2018, ’19 there wasn't significant leaps of innovation happening in the semiconductor supply chain and everybody was happy with the way things were getting built. Then first Covid happened which resulted into chip shortage which we’ve all heard about. Second, as you realize every digital product out there in the world actually has a chip inside and it became a geopolitical issue, chip became the new oil and every country had to become sort of self-reliant on their own capability to build chips. Third, the Indian government obviously recognized that there’s a need for chips getting developed in India. To do that they did two things, one, is design linked incentives, DLIs to incentivize more and more people to design chips. And then bunch of PLIs which are production linked incentives to boost stimulus in end markets the chips are required for which means auto, EV, consumer electronics production, EMS, ODM etc. And we always had the design pool at increasingly over the last 5-6 years more and more senior talent and entire design teams had started moving from the US to India. So a combination of geo political factors, India’s need to be self-reliant and government incentives, exiting talent, the end market starting to show up and all the change that’s happening in the world in terms of AI, EV, 5G, new instruction architectures like RISC-V coming in has made it a very interesting time. And India is both right to play today and we at least internally believe our right to win in the global semiconductor market.

Nivedith:

I think you touched upon the emergence of end markets, right, and that’s a very interesting point. Today how large is the semiconductor opportunity in India and what are some promising subsets of these?

Sudipto:

So if I go by sort of secondary data, right, at least it says that it’s a 20 billion dollar market. Today we’ll get to 60-70 billion dollars over the next 5, 6, 7 odd years. So the challenge is you can build the best semiconductor in India but finally who do you sell to. If you're building it for cloud you need to sell it to Amazon, if you're building it for cars traditionally you would have needed to sell it to Tesla or a General Motors. So, the end – if you were building for phone so obviously sold it for Apple, Samsung, Xiaomi. So what happened was we did not have the end customer, so it was very difficult and semiconductor chip development is generally you develop it with a customer to understand their needs and preferences and build for them. So that never existed in India.

But today if you see what India has done in terms of auto and EVs especially in two wheelers where we’re probably in numbers second largest two wheeler producer in the country. If you look at EVs obviously we’re very proud of what Ola Electric has done but across both traditional incumbents as well as new companies in EV two wheelers, three wheelers, four wheelers are starting to boom. India is at the forefront of 5G and telecom, so all these industries are here and now today and they would want companies which actually build products customize to them. So if you go and talk to a Nvidia or AMD, Indian companies are nowhere in the roadmap but they’re producing chips customized for India which is why we need Indian companies that take Indian auto and EV requirements and build for India. Take India 5G communication requirements, build for India. As we have a more robust electronics manufacturing ecosystem, consumer electronics, fridges getting built and ACs getting built in India chips for them. So I think that's where the opportunity is. The end user manufacturing has come to India so we also need semiconductors that go into those end products.

Nivedith:

And you also spoke about the rich talent base that we have, right, back in the 1990s Texas Instruments was the first global semiconductor company to really set up shop in India. They were followed suit by the likes of Qualcomm, Micron, Intel and a bunch of others. So while today there’s no doubt that some of these setups house probably some of the best design talent in the world today. What has the evolution been from a captive back office to hubs of innovation?

Sudipto:

So I think that was also one of our biggest questions, so when we started doubling down on semiconductors we went and met a lot of senior folks across all of these companies. We met Anand at Micron, Naveen at Marvel, a bunch of folks at Intel, Nvidia etc and the striking thing is in the last 4-5 years a lot has changed. So when I was graduating out of IIT in 2012 I had gone to Texas Instruments, from then to today I think at some point of time India was a back office, then a verification office. But today entire host of products are getting built directly out of India. So the quality of talent has significantly improved and that is our right to play, without that we actually could not have played because this is a very technically complex industry. But today Indian officers of large MNCs have really good quality talent, on top of that you go in the Valley and you look at any large company or any large startup like Tenstorrent or SambaNova etc you’ll find hundreds of Indian engineers actually powering those companies. So combination of talent in India plus Indian talent in the Valley makes us super excited on what they can build if they focus on building for India and then eventually for the world.

Nivedith:

So I think it’s clear now that India does have a very important role to play in the global semiconductor value chain. But for an Indian company what is really the right to win?

Sudipto:

So I think first as we said the end market is here so if you're building for Indian customers the right to win is extremely clear. You have the talent and you're building for Indian customers, you’re focused on the way the building is catering to their need so it’s clear right to win. Now when I think about an Indian company winning in the world the right to win is, a, very good quality of talent but at obviously cheaper rates than in the Bay Area. 2, all the DLI etc incentives that the government has given which is subsidizing your cost. Third, you’re building for India through which you're testing your products. But then once the product actually works you can take it to the world the way a lot of SaaS companies in India have done in the past. So if I look at in the entire supply value chain of semiconductors in fabless design the right to win for India is extremely clear where you build for India and then you take it to the world. In the other parts of the supply chain it’s still very difficult I think, obviously a startup it’s very difficult for a startup to either build a foundry or a packaging or a photolithography machine, it’s also difficult to become an IDM where you integrate both your chip design capability and start manufacturing on your own. So those I believe the thing will take years to come and a lot of it will be incumbents, large global MNCs setting up foundries or packaging units in India but on the fabless design space India has a clear right to win in Indian market and once you’ve developed high quality products at comparative prices hopefully potentially take it to the world.

Nivedith:

So I think we’ve touched upon the India opportunity a little bit, let’s maybe spend a little bit of time on some global trend lines that are really defining the semiconductor landscape today. Back in 1965 Gordon Moore who was one of the founding fathers of Intel made a prediction that has largely help true till today, but as we move to more advanced process nodes, TS is even talking about a 2 nanometer process node now. Experts are saying Moore’s law has largely played out, now while there is a school of thought that says that advanced technology like different types of transistors, 3D architecture will continue to advance Moore’s law, they will be prohibitively expensive to begin with. What are the commercial implications of this?

Sudipto:

So I think first it’s fantastic to see first of all Moore’s law play out, like something that was predicted in 1960s more or less working out over 30-40 years is fantastic to see. And what Moore’s law did is essentially every two years you could get a faster and cheaper chip from the same large companies like Intel and Nvidia and that would take care of your compute needs. Now with Moore’s law of flattening you can't expect every 1-2 years that general purpose chips become actually faster and cheaper year on year. And I think the trend line that we’ll start seeing because of that is more custom silicon and sort of companies building chips focused to their end use cases which can become then more power efficient cheaper and faster for their use cases. And actually you're already seeing that, right, if you see all the hyper scalers, Amazon has Graviton, Google has TPUs, Meta has its own custom chip. You talk of EV players Tesla has its own suite of own custom silicon chip, Dojo etc. So increasingly we’ll see more custom silicon that is cheaper, more efficient, both in terms of processing and power for a particular use case and a company versus only general purpose chips because of flattening of Moore’s law.

Nivedith:

So let’s talk about RISC-V now. That's another trend line that’s really shaping the semiconductor landscape. We know that its an open source instruction set architecture developed at UC Berkeley and we’re very proud of the fact that our very own SHAKTI – Microprocessor that was developed at IIT Madras is also based on RISC-V and they’ve had a very crucial role to play in RISC-V. But currently today what are the inherent advantages that RISC-V provides and where is it in its adoption journey?

Sudipto:

So first as you mentioned SHAKTI Labs out of IIT Madras is one of the other pivotal reasons which has brought semiconductors in India into limelight so fantastic job done by the team. Also bunch of really great promising startups that have come out of the lab. I think overall RISC-V as a trend line is exciting is because x86 was obviously a closed architecture used mostly in CPUs. If you had to work with ARM you have to pay a very high fat licensing fee which is okay for multinationals which are worth billions of dollars but very difficult for startups. So what RISC-V enables you to do are two things, a, it’s open source so you don’t need to pay and two you can customize that architecture for your own needs. You can have the best custom ISA that solves for your particular use case. So really good for the ecosystem and especially for startups where they can take an existing ISA and starts building on top of that without having to pay ARM a very big loyalty fee. Now in terms of where it is in terms of journey I think in terms of hardware the ISA is already developed at a point where you can actually take the ISA and sort of design your hardware at the same level probably will get with x86 or ARM. I think the place where it lacks a little bit is on the software stack where if you take the AWS or the cloud ecosystem there are multiple layers of software, an ecosystem that has been built to power cloud servers and those servers today are either in on x86 or on ARM architecture. So I think that software ecosystem will take a few years for you to develop. So if I were a founder and thinking about whether to use RISC-V or not in any application which is the closed ecosystem where you both design the hardware and the software say for example embedded systems use cases. The entire hardware plus software stack is built inhouse. Or you take use cases like data storage, not cloud, anyway it’s a contained ecosystem in EVs. I think RISC-V is already there in capability both in terms of hardware and software. If it’s an open architecture where you have sort of interact with bunch of different softwares for example in the cloud I think it’s a few years out there till the software ecosystem around RISC-V matures to a point that you can start building for the cloud using RISC-V.

Nivedith:

Just a follow up question to the question on risk five. What are some notable companies globally or in India that have kind of moved the RISC-V adoption along?

Sudipto:

So I think globally if I think about probably one of the most popular names is a company called Tenstorrent which is building both CPUs and sort of AI accelerators on RISC-V. There’s also Ventera Microsystems which is building CPUs for data center. A bunch of other promising startups if I only talk about RISC-V I think those two will be top of mind. Closer home in India also there’s a very strong RISC-V ecosystem, as I mentioned there are bunch of good companies like InCore and Mindgrove that have come out of the SHAKTI Labs and increasingly as you know we meet a lot of really good companies but early and that are adopting RISC-V.

Nivedith:

I think no other trend line has probably created more hype and more importantly more value creation in the artificial intelligence trend line. Nvidia’s market cap soared to 1 trillion plus after Jenson Wayne gave estimates for quarter 2 for CY23. SambaNova today is valued at upwards of $5 billion all the way back in 2019 Intel acquired Habana Labs for $2 billion to kind of further their AI specific product offerings. How do you see this playing out and does India have a role to play in this?

Sudipto:

So I think the AI is probably the most popular buzzword today and rightfully so. And if you look at at least on the AI stack as AI has exploded over the last probably 12 odd months the biggest beneficiary have been semiconductor companies. Because if we have to make an AI model work you first need to train it. And the company with the best combination of hardware and software to train is Nvidia, they have their H100 chips, the H100 GPUs for training along with the entire software stack of CUDA on top which enables them to give the best possible offering for anyone who is trying to build large training models which is why you’ve seen so much of value creation. Although if you look at recent sort of tests on performance benchmarks I think both AMD and Intel thank God they acquired Habana and now they’re actually quite close to Nvidia in terms of performance and training chips but it’s the software ecosystem is where Nvidia has actually done really well. Now coming to India I think even companies in the US large multinationals are finding it very difficult to compete with Nvidia so it’s not that easy to compete with Nvidia directly for training purposes on cloud. But there are lot of other use cases, so two things. First those costs, the chips are extremely costly and it’s supply constraints. So if you can create more chips of similar quality and at a lower price you will obviously have takers for it. But second more interestingly given how large the training market is all the companies are focused on training market for large language models. Other lot of use cases for example Ed use cases for AI etc which large MNCs are not focusing on today. And I think as you take a AI first lens to the world a lot of the chips will be re written because a lot of the end use cases will be rewritten AI first. And keeping that in mind you can take existing use cases on the edge and start building chips which is AI first and then sort of build. So I would probably say competing directly head on with Nvidia and AMD today difficult both on hardware and especially on the software stack layer but lot of use cases outside that especially on edge computing where there’s a role our Indian companies can actually compete and build large companies.

Nivedith:

And given this positive outlook that we have at Matrix on the semiconductor industry in India how many investments do you see us doing over the next say 3-5 years?

Sudipto:

So we’re super bullish on this space and we believe semiconductors is in some shape and form like software where a use case is a use case, you can have large companies. For example you can have multiple companies just building AI accelerator chips. You can have companies catering to specific use cases, if you take Texas Instruments, right, it has 80,000 SQs, has probably 60,000-70,000 SQs. So if you take just the battery and start building chips for battery and assuming millions of vehicles in India in EV gets sold which means millions of batteries are getting sold in India. You can build a large company just on that, so we believe multiple large companies will be built, difficult to say a number, it will also depend a little bit on the entrepreneurship that that gets created in this sector. But we hope to make multiple investments every year.

Nivedith:

So, Sudipto, given that you mentioned India does have a role where will the next set of Indian founders really come from and if someone is looking to build for India what should they be looking to build?

Sudipto:

So I think I’ve answered both the questions a little bit before, we spoke about the design talent, right, Indians design talent both in India and in the Bay Area. So semiconductor is obviously designing semiconductor chip is an extremely technical thing to do which means people with prior experience of having built cutting edge chips either at large MNCs or startups or universities at the right position to build something so I think the talent will come from India and who are already working in MNCs or are in the Bay Area and working cutting edge chips. And there are hundreds and thousands of those as India is the second largest pool of semiconductor design talent. What should they build I think we’ve spoken about how the Indian semiconductor industry will boom, we’ve spoken about how end markets like EV and auto, 5G and communication, actually defense is an industry if you look at consumer electronics may not be smart phone but the rest of the consumer electronics will boom in India and there will be a captive market that you can start with. I think as a founder maybe the best thing to do is take up a particular industry use case, build an ASSP which is specific to that industry but can be used across multiple different companies and start with that. And then once you do that deliver that, have a order revenue book of maybe 10 million dollars raise a lot of money on top then you go and then start computing in the US to build a really large company. But sort of that's the playbook I will take, come to India, take a end market, take a specific use case, build an ASSP , build a strong revenue book and capability, both tech as well as sales and then raise money and take it to the world.

‍