Full Transcript

Dhruv Sharma: Hey there listeners, it's Friday, June 19th. We're streaming live. This is stream number 102.

Today, Utsav and I are speaking with Vishnu of Eigen Labs. Vishnu, welcome to the show. Thank you so much.

Vishnu Patankar (EigenLabs): Thank you so much. Happy to be here.

Utsav Somani: Good to have you with us, Vishnu. And glad that you're in the same time zone as us. I believe you're based in the US, but you're traveling to your hometown, which is Bangalore. So good to have you in the same country.

Vishnu Patankar (EigenLabs): Yeah, very much excited to be back. And I love coming back here. There's so much going on. Can't miss it. We have people all over the world. In the blockchain world and now more and more AI, wherever we can get the talent, we will go. And Bangalore is a top place for us.

Utsav Somani: Awesome. So let's start with Eigen Labs. Can you introduce the company to our listeners?

Vishnu Patankar (EigenLabs): Yeah, we are, think of us as a neo research lab. You can see a lot of these AI companies that are valued north of one billion dollars, they raise humongous amounts of money, they have free revenue. And so the potential for the science and research that these teams do is what they're being valued for. So think of us in that light.

We are promoting what's called open agentic science. And we believe that coordination between people is going to be very important in the future, especially where power is getting centralized with a few vertically integrated frontier labs.

Utsav Somani: And why this coordination technology and what's the science behind this?

Vishnu Patankar (EigenLabs): Yeah, so it has its most modern roots in the blockchains themselves, where smart contracts coordinate people's actions. They are transparent, they're cryptographically verified. What it does is it brings down the barrier for people to trust each other. And so, if I wanted to transact with you as a counterparty, how do you know that you will hold up your end of the bargain? The way blockchain does it is the code is the law. So long as you can see the code and it's executed by many validators and attested, you get the correct execution.

Coordination technology is even more than that — money itself is an age-old coordination mechanism. The constitution of whichever country you're talking about is a coordination mechanism. The founding fathers built the preamble and constitution so that people can operate on those rails. So this has been there for many ages. And now it's in the digital domain. It's a fun time to be around.

Dhruv Sharma: Vishnu, in your mind, how do you draw a distinction between decentralized technologies like blockchain versus what we have in AI right now, which — I don't want to say it's centralized — but the truth is there's a handful of model companies?

Vishnu Patankar (EigenLabs): Yeah, it's a great question. I think it's something every other day any mainstream media outlet, any think tank — whether it's the Guardian, the New York Times — are all talking about.

In one way, you need a group of people to start off. You can't have a completely decentralized start because you can't concentrate your resources — you'd have chaos. But over time, as the technology matures, you're going to have a need for getting efficiencies again. This is what we're seeing in the Frontier Labs. But if that's left unchecked — absolute power always leads to absolute imbalance.

You've heard of the K-shaped economy. You've heard of the recent IPOs, the first trillionaires. There's no doubt about the ideas themselves with AGI and space technologies. But you're creating single points of failure. It's no different from when in computer science you build a distributed system and you want more peer-to-peer nodes and peer-to-peer protocols so that they can scale.

Another thing I'll say: this time it's different. If we don't have coordination technologies, everyone outside of the 10,000 people working in Frontier Labs — which is 8 billion minus 10,000 people — are going to be left out of the future economy. Unseen wealth is going to be a problem in the future.

Dhruv Sharma: There's an editorial from the North Coast and the Financial Times where he says that AI is going to need its own tax code, because power and wealth are concentrating in a handful of hands.

Utsav Somani: Elon Musk said that saving up is going to be irrelevant now, because post-AGI the world looks very, very different. What are your thoughts on this post-AGI world?

Vishnu Patankar (EigenLabs): That's a good question. I'm a person of science, an engineer. You could have Dyson swarms with intense sources of energy feeding the Earth. When I was growing up I was watching Terminator and these self-evolving AIs. I also worked in silicon at Intel. It's really interesting to see how NVIDIA has captured the market there.

Ultimately, all the things I just talked about — whether it's energy, silicon, some other material — I do think in the future there's going to be abundance. Robots building buildings. If you look at your everyday wants and all of that was available at zero cost — what is the purpose of life after that? You get into more philosophical, spiritual things.

Will it happen in our lifetimes? I think it's unlikely. But a few hundred years from now, it's definitely going to be there. In that land of abundance there's going to be some pain, some displacement of jobs. The right kind of governance and policy, where tech folks work with governments to generate the right long-term incentives — that's going to be very important.

Dhruv Sharma: Vishnu, very topical — the AI export bans. If governments step up and ban the export of state-of-the-art models, certain countries are going to get ahead of others. The difference between haves and have nots, even at the sovereign level, is going to keep widening.

Vishnu Patankar (EigenLabs): One has to look back at nuclear technology. When the U.S. came up with nuclear technology, it wasn't like Russia invented it on their own — there was a leak from people in the U.S. who were building it. The reason they did that was they believed that asymmetric power would be ultimately bad for the world.

I think it's a similar thing with AI. You're going to have people even in the frontier labs move out and share knowledge with open source models. China will catch up. India will catch up. Something as powerful as this — the only way to counter it is for everyone to have it and make it an equalizer. Using AI to design very targeted protein molecules that could solve health problems for specific people — those are the things we should be focusing on. Utsav Somani: What are the thoughts that you and the company have developed on verifiability of AI?

Vishnu Patankar (EigenLabs): AI is fantastic in execution. But there are two other bookends: the taste aspect and the verification. On verification itself, there are two aspects: correctness — did the code meet its specification — and execution verification — once a piece of software was going to execute, did it execute as it was supposed to? Was the binary changed? Was there a security bug injection?

What we have done with EigenLayer is on-chain attestation of execution. The combination of formal methods with cryptographic verification is where the most exciting work is happening.

Dhruv Sharma: Does verifiable AI also mean the technology can verify for you if what you see is what you get — if you're being served the model you've paid for?

Vishnu Patankar (EigenLabs): That's exactly right. The idea is there's this technology called trusted execution environments. So you have Google confidential compute, AWS Nitro. In a secure environment, you make sure that the code, the model, the weights that are there is exactly what was promised to you. One way is the trusted execution environment itself and its attestation. The second way — if you want to re-execute it, it needs to provide the same result again.

For that, we've built something called deterministic verifiability. When you rerun a model, it gives you exactly the same output. We control batch inference, temperature, tuning parameters so that non-determinism is not available in the system. So if you run a prompt 100 times, you get exactly the same output tokens every time. That doesn't happen today because of KV cache handling, prefix computation, optimizations these labs do. But if you're willing to spend about 20% more, you can get deterministic inference.

Utsav Somani: What are your thoughts on decentralized AI versus verifiable AI? How are those two things different?

Vishnu Patankar (EigenLabs): There's also distributed, right? I would say verifiability is more of a superset, or a close peer, to decentralization. For something to be verifiable, it probably has to always be decentralized — there has to be no single counterparty that you can trust for that output property of verifiability. But if something is decentralized, it's not necessarily verifiable.

Dhruv Sharma: Would you say post-training techniques like RLHF also involve doing the same thing — human validators, say, if you're trying to find a model for application inside a law firm, will have lawyers verify its output?

Vishnu Patankar (EigenLabs): Yes, and you bring up an important point. A majority of data is sitting inside corporate firewalls and boundaries — the proprietary data. The frontier labs would love to get their hands on it. I know for a fact that OpenAI is building a confidential compute version that a big bank wants to use, because the bank doesn't want to give away its trade secrets while training the bigger model. That kind of customization — whether it's RLHF or any other kind — that's the secret sauce that any company has.

Dhruv Sharma: How are the frontier labs serving customers like that? Are they forward deploying engineers? On-prem deployments?

Vishnu Patankar (EigenLabs): OpenAI is now having these on-prem VPN, trusted execution environment, confidential compute containers that even their own engineers cannot see. With enterprise AI, I think next year is going to be the big year in terms of actual adoption. Initially people talked a lot about agents but there wasn't much ROI yet. That's coming — all the way from payment trails, governance, formal verification especially for high stakes use cases. On a stock trade of a billion dollars, 99.99% is not good enough. You need formal methods.

This is something I had worked on in my master's thesis 25 years ago on hardware. Very exciting to see it come to life in software.

Utsav Somani: You've had amazing roles before — Kraken CTO, CTO of StockX. What has that time taught you about trust?

Vishnu Patankar (EigenLabs): StockX is a marketplace to allow access to hard-to-find items — a $3,000 sneaker, a Pokemon card, collectibles. These are asset classes in their own right, just like you invest in a stock. We found that 30% of customers were not using those sneakers — they were simply investing to resell at a higher price. So we said, why don't we represent that with an NFT on chain? You save the shipping fee and gross margins on both sides. The customer finds it cheaper, the company is more profitable. That's really how I got into crypto professionally.

With Kraken, we built a brand new data center — the fastest cryptocurrency exchange out there, microsecond and nanosecond level latency to execute trades. It's been a really fun journey.

Utsav Somani: Google put out a report that quantum computing was going to break the encryption in blockchain. A lot of people got scared. What are your thoughts on that?

Vishnu Patankar (EigenLabs): This goes back to one of the thesis we have — open agentic science. The keyword is open, so more people can participate and have agency, not just the Frontier Labs.

If you go to ECDSA.fail, you'll see what happened. It was basically an 18-year-old and a 24-year-old. Google published a paper saying ECDSA — the encryption protocol that not only Bitcoin is based on but also TLS — can be broken by a quantum computer, not in 2035, but maybe five years faster, around 2030. Just imagine if TLS encryption is broken — banks not being able to function. That's unacceptable.

Using open agentic science — an open network where anyone can come and try to solve it with their own model, cloud, Codex, an open model — we just created that marketplace for people to submit solutions. We got 44% better than the Google result. Which means Q-Day, the day encryption gets broken, moves even quicker than 2030.

The beauty is it was done by amateurs, people not necessarily world class in quantum computing, just using their agency and the models available. They could actually beat a lab like Google. Very exciting.

Utsav Somani: As a final closing question, Vishnu, what's next for Eigen?

Vishnu Patankar (EigenLabs): More about providing agency for everyone to participate in this post-AGI economy. The question is what is everybody going to be doing when this happens? There's going to be a land of abundance. But there's also going to be purpose in everyone's jobs and lives. We want to be that place to enable all of that through coordination technology.

Utsav Somani: Thank you so much for coming on the show. Wishing you the very best.

Vishnu Patankar (EigenLabs): Thank you so much.

Utsav Somani: We're moving on to our next guest. We've got Chetan from AUM Ventures. Chetan, welcome to the show.

Chetan Mehta (AUM Ventures): Hey, hi. Thanks. Thanks for having me. Great to see you all.

Utsav Somani: What's the story behind the name AUM?

Chetan Mehta (AUM Ventures): Personally, as a spiritual person, I wanted a deep-rooted Indian spiritual word. AUM is something which is not only Indian but transcends. It's the source — from where you find your source when you say AUM.

Utsav Somani: What's the technical explanation? What's the fund size? What are you investing in?

Chetan Mehta (AUM Ventures): The story of AUM started quite a few years back. I was investing for a family office doing tech investing globally — invested in several global ecosystems including Israeli deep tech, quite a few companies listed on NASDAQ, had exits. As I was meeting global investors, one common feedback was they had zero exposure to India.

Personally, on my calling and passion, I wanted to work with very early stage founders. I started AUM Ventures focused on three things: back entrepreneurship and Indian founders, back innovation from the country, and take the innovation globally. Because that's where India can really leapfrog into being a global startup ecosystem.

In my earlier role, I had invested in SpaceX in 2019. So space tech was something I felt not many were investing in. I started meeting a lot of founders, spoke to ISRO, and met the SkyRoot team. SkyRoot became the first investment from the fund.

Today in fund one, we've done five space tech companies. With Semicon, when the government started talking about Semicon in India, we started looking at that sector. We've done one investment, Azimut AI — at that time, no one was investing in chips. We took a bet on this founder. Today the chip is taped out, they've won a couple of OEM contracts, and the government is talking to them.

With fund two, the thesis is to double down. What we're seeing globally is a mix of two or three things converging. One is geopolitics — every country is looking inwards for sovereign capability, reducing strategic dependence. So space, Semicon, defence are becoming national priorities.

Second, with advancements in AI and ML, the speed at which you can build from idea to product has become much faster. Third is capital — globally, software SaaS is almost dead. Money is chasing IP-led, hardware-led startups. In the last 10 years, the biggest companies that have made money are all deep tech — SpaceX to Nvidia to the AI companies.

Dhruv Sharma: Chetan, you brought up interesting points. One, almost all countries are looking inwards because of geopolitical instability. But at the same time when we talk about the opportunity, we start talking about building global companies. On a 10-15-20 year scale, what's going to be insulated and what's still up for grabs when it comes to exports?

Chetan Mehta (AUM Ventures): I believe all these areas are tremendous opportunities to build global companies. Take SkyRoot — it's a global company, and it can and will be one of the top space tech companies globally. The launch business has the biggest moat in space tech. If you build reliability and have a regular cadence of launches, there's a big market.

In sectors like space, defence, Semicon — today they are also an ecosystem play. NVIDIA still has to rely on TSMC for fabrication. So how do we capture the value in India? First is the sovereign part — build national capability. Second, take sectors like defence: with what's happening, we want more capabilities in India, so there's demand from the defence budget. Third, space tech is global from day one. In deep tech, you have to think global from day one — you can't build from India and only think global after five years.

Utsav Somani: You've done one defence investment that saw an exit to LAT Aerospace — Sharan Shakti. Talk to us about how you found that investment and the journey.

Chetan Mehta (AUM Ventures): We had done an event at IIT Delhi. Just after that, Karan, the founder of Sharan Shakti, did a cold email to us on LinkedIn. We met him a couple of times. There were five team members, all from IIT Delhi, who had started building a counter-drone solution with some prototype. They wanted to build a full stack drone plus radar tech.

We came in as the first check, worked closely with them. For what he was building, he wanted to raise a larger round. He went to Dipinder as well as part of his fundraise. Dipinder was also building LAT Aerospace. He felt this requires a lot of investment — why don't you guys come and join us? And that's how it happened, and the exit happened as well.

Utsav Somani: The question we've had amongst ourselves — deep tech investments underwrite a huge cycle. Huge CapEx, huge R&D, huge exit cycles. Can India fund these cycles all the way to exit?

Chetan Mehta (AUM Ventures): Traditionally, investments in deep tech have not really flowed — it remains a sector where it's still not capitalized enough. Last year India got 12 billion as FDI and only 200 million went into space tech. There's so much more we can do as a country.

But I don't think it's a misconception that deep tech requires a lot of CapEx — if you compare with 10-minute delivery businesses, they've required more capital than deep tech. Skyroot has done it under 100 million dollars, while a Virgin or SpaceX would have taken a billion. Frugal innovation is India's advantage.

On exits — capital markets are also looking for opportunities like these so that retail can also participate. From a duration standpoint, good quality founders can shorten the time from start to commercialization. Any startup you have to give seven to ten years anyway — Zomato took 12 years to IPO. Deep tech may take seven to fifteen years. SpaceX took 24 years. Good things do take time.

Utsav Somani: Let's bring one more friend on the show. Chiran, welcome. Where are you dialing in from?

Chiran Phanindra (Cosmoserve Space): Thank you for having me on the show. I'm at VivaTech Paris. I can give you a quick glimpse of what's happening here.

Utsav Somani: Please. Do you want to take your camera?

Chiran Phanindra (Cosmoserve Space): Yeah, I'll take you around just for a minute. So yeah, this is the India pavilion. We have three India pavilions. It's a huge event — all Indian companies across all sectors of VivaTech, including robots, AI, space, automotive, mobility.

Out of 30 startups selected, six were selected to pitch directly to Prime Minister Narendra Modi and President Emmanuel Macron. They came to our booth. I told them we are building space debris solutions — we are sending a robotic spacecraft to collect space debris and remove it. Modi ji pitched for us to Macron, saying India has the capability of launching satellites and also pulling them down to Earth while they are dead. It's an honor to hear such a statement from Modi ji. And Macron said they're keen to collaborate with technological advancements in India and with companies like us.

Utsav Somani: Wow, what a story. Chetan, how did you guys meet? What was Chiran building and how did the journey evolve?

Chetan Mehta (AUM Ventures): Pawan and Chiran had worked together while at ISRO. Pawan is the founder of SkyRoot. At that time Chiran was still at ISRO, thinking about Cosmoserve, and wanted to brainstorm. Pawan connected me to Chiran and we started talking. At that time, Chiran was maybe serving his notice period.

We had a lot of back and forth on his business plan. Chiran comes from a technical background and had never prepared a business plan.

Utsav Somani: Deep tech founders — they must be so obsessed with problem solving that they don't have time for polished pitches. Maybe AI helps in that regard now.

Chetan Mehta (AUM Ventures): Yeah, it's also about having clarity on how much you want to raise, what are your milestones. I asked Chiran how much dilution he was looking at — and Chiran had never heard the word dilution. For him, dilution was a chemistry word.

Utsav Somani: And he did not know what dilution means. So it tells you something.

Chiran Phanindra (Cosmoserve Space): Yeah, I don't know anything about equity. On the first meet, he asked me to select a place. I thought, an investor is coming to hear your pitch, you need to be at a grand place. So I chose a hotel for a brunch. It was a noisy place. The next day he wants to meet again, and this time he selected the place — a coffee shop. Then I understood that pitches happen in coffee shops, not five star hotels. That's where I started my investor journey.

When Chetan first asked me about dilution, I was dumbstruck, silent for some time. Then he asked about valuation — I could correlate some term between dilution and valuation from that. And later he asked if the lead investor is giving me DD. For me, DD is demand draft. The lead investor is sitting in the US — why would he give me a DD? Then I understood DD means due diligence. In this industry, all these terms are new. Every day we learn new terms. Adaptation is more important, I feel, for any human being.

Dhruv Sharma: This is so refreshing, Chiran. Maybe we should start an online petition — instead of space tech founders having to learn about dilution, maybe the data room should include CAD files and simulation data that VCs should learn to interpret.

Chetan Mehta (AUM Ventures): Chiran has gone 360 — today he goes and pitches at events. Great to see him transforming as a startup founder.

Utsav Somani: What have you learned about the space industry that has stuck with you, Chetan?

Chetan Mehta (AUM Ventures): Space is so humbling, it requires a very different level of mindset. I was speaking to Pawan — in your case, the failure is broadcast live, in front of everyone. You send a rocket and something happens. A lot of ministers are attending. It requires a very different level of conviction.

In space you can't really iterate too much. You have to be best in class in terms of your tech, best in class in terms of processes before you put a satellite or launch vehicle to space. And from day one, you have to think globally. All our portfolio companies are already talking to global companies — on the supply chain side for requirements, and on the customer side for LOIs.

Dhruv Sharma: Chiran, let's start with you explaining — what is space debris? Why is it important to remove it?

Chiran Phanindra (Cosmoserve Space): There are about 50,000 odd objects out there in space. Of that, only 12,000 are operational satellites — just one fourth. The rest — around 40,000 objects — are man-made junk in space. These 40,000 objects create a great threat to the operational satellites. Unless we remove this junk, they will impact operational satellites and create more and more debris. If that cascading effect happens — what's called Kessler Syndrome — future generations will never be able to use space again the way we do. No GPS, no video calling, no applications.

Elon Musk has filed for an IPO and is building data centers in space. There should be space in space. So we are building a robotic spacecraft that goes to space, collects the space junk, and removes it.

When I first tell this idea to people, they think we're doing a nonprofit business — why would someone pay to remove space junk? But if a car gets stuck on the road, you can't leave it there. You call a towing van. Same regulations are now coming in space, forcing satellite operators to remove dead satellites. We can't leave dead satellites there forever. That has created the market. We are building those towing vans and refueling depots in space to create a sustainable space for future generations.

Dhruv Sharma: The one big difference of course is that space debris is not stationary. It's in orbit and could be on a collision course with your flight path.

Chiran Phanindra (Cosmoserve Space): Exactly. SpaceX, for their satellites, does thousands of collision avoidance manoeuvres every day. Two satellites come closer, one is moved from its path so the other can pass by, then it returns. Satellite operators are burning more and more fuel for these manoeuvres.

Utsav Somani: Can't all satellites just be regulated to have collision avoidance technology?

Chiran Phanindra (Cosmoserve Space): The problem is that objects coming from different directions and at very high speeds can't always be detected in time. It's a passive debris problem — the junk itself has no propulsion.

There are about 50,000 tracked objects. But there are a million more smaller objects below 10 cm — too small for ground radar to track but large enough to puncture a satellite. So the problem is much larger.

Utsav Somani: What's the status of Cosmoserve? Where are you in the journey?

Chiran Phanindra (Cosmoserve Space): We started the company about two and a half years back. We have 32 people now. We've done our preliminary design review and critical design is ongoing. We have ground-tested our propulsion technology. The plan is for a demonstration mission — our prove-it moment — followed by commercial missions.

We're also exploring the servicing market: refueling and life extension for existing satellites. A satellite that runs out of fuel still has perfectly working systems — if you refuel it, you extend its life by 5-10 years. The operator saves hundreds of millions of dollars versus launching a new satellite.

Indian regulators and government have been very supportive since the Space Policy 2023. We're in conversations with both Indian and international satellite operators.

Utsav Somani: Chetan, what are you most excited about for the next one or two years coming out of India?

Chetan Mehta (AUM Ventures): We'll see a lot more companies coming in space, Semicon, defence, advanced manufacturing. I'm seeing significant interest now in hardcore hardware-led, IP-led businesses. One thing I'd love to see is more corporate participation — corporates coming in as design partners for early-stage startups. That would be wonderful for the ecosystem.

Government is doing its bit. I hope the RDI program and others are implemented at pace. But overall, super excited. There is so much great innovation from India. These ideas can become global businesses. Super excited to back these next generation founders.

Utsav Somani: Thank you so much for coming on our show. Wishing you the best.

Chetan Mehta (AUM Ventures): Pleasure. Thank you so much.

Utsav Somani: Alright, Chiran. We'll talk about the tow truck of the space world in slightly more detail.

Utsav Somani: Last seven decades, we launched only 20,000 objects. Now, the next one decade, one million satellites are planned to be launched into space. From 20,000 to one million in just one decade. If we do nothing, there will be multiple collisions happening every second.

Chiran Phanindra (Cosmoserve Space): Exactly. It's a cascading effect. There are about 50,000 tracked objects, but actually a million more smaller debris below 10 cm that ground radar can't track. We've launched 20,000 objects in seven decades but in the next decade, one million are planned. That changes everything.

Utsav Somani: Thank you Chiran. Really appreciate it. We'll see you all on Monday at 4 PM. Bye bye, folks.