“Just as AI evolved from university labs to industry deployment with the emergence of Google Brain and OpenAI, ZK technology is now transitioning from custom protocols to programmable, general-purpose tools that could reshape how we build blockchain infrastructure.” - Yi Sun, Axiom

Earlier this week, I caught up with Yi Sun, CEO of Axiom – a company focused on building the open infrastructure to help transform ZK from a niche pursuit to an industry-defining force.

Having worked closely with Yi as an advisor to Scroll over the past three years, and collaborating closely building the OpenVM project. I've had the privilege of watching Axiom's journey from the beginning. In late 2022, Yi saw an opportunity to push ZK beyond theoretical cryptography into a practical, general-purpose technology that could have massive impact — an opportunity he felt was so significant that even a small company could help shape the future of the field.

Here's our conversation:

Yi: "I had a relatively longer career in academia. I did a math PhD and went on to become a statistics professor at the University of Chicago, where I was doing research in theoretical probability and high-dimensional statistics. It wasn’t until 2017 that I got into crypto - what drew me to it was how it seemed like the conjunction of my theoretical research interests with a much more practical, real-world impact.

For several years, I stayed tangentially involved in the space, helping out early-stage startups as an advisor, including Gauntlet and, of course, Scroll. In 2021, I started getting more directly involved when I saw that ZK technology was becoming practical enough for real applications."

Me: "ZK went from being the hottest keyword in crypto to being marketing poison. We've seen these narrative pendulum swings before, how do you see this playing out?"

Yi: "ZK, just like many other trends in crypto, often goes through phases where the hype and speculation can be well ahead of the technology. The technology goes through phases where it's overestimated and then underestimated. But probably today, I would say it's a little underestimated.

The progress has been pretty amazing. Back in 2014-2015, when programmable ZK proofs were just getting started, the overhead for generating a zero-knowledge proof compared to running it on your laptop was something outlandish, around 10^10. Now that overhead is around 10^4, which is much more reasonable for real applications."

Me: "Axiom’s success depends on more companies using ZK in real world applications, can you give me three billion-dollar ideas that could be enabled by ZK technology. Where are the business opportunities, and what are the blockers?"

Yi: "The first general category would be rethinking how you build blockchain infrastructure from first principles, given that ZK exists. Justin Drake announced at DEVCON that he wants to do this for the Ethereum consensus layer. This applies broadly to rollups, bridges, and any sort of intent-based protocols.

At the base layer, ZK can be helpful for synchronous composability, cross-roll-up bridging, and communication with much shorter finality. I think the reason we haven't seen as much of that is that, although ZK has been difficult to build with historically, those categories are pretty challenging to deliver any product in, even without ZK.

So, the adoption timeframe needs to be matched up with what's possible technologically in ZK. For example, if you start to build a rollup, it's going to take at minimum 12 to 18 months. And so those roll-ups that are coming to production today were conceived of about a year, a year and a half ago. As we see roll-ups really rethinking themselves from first principles based on the technological upgrades in ZK, we're going to see improved performance, lower costs, and greater composability.

The second category I'm excited about is using ZK for authentication, which is interpreted pretty generally. There are many different cryptographic formats of data across the world, both within blockchains and outside them. Every blockchain has some signature scheme, but there are also many types of signatures — OAuth, email signatures, etc. ZK can translate between these formats seamlessly.

The third category is privacy applications. This has lagged behind computational integrity uses because privacy use cases require proving on the client side, which is much more constrained. But I think performance will soon reach a point where we can deploy real privacy-preserving applications."

Me: "Can we break down the first category into business cases that application builders can understand? How do you see this playing out on a 5-10 year horizon, assuming we get to 1 billion people on-chain doing 10 transactions per day?"

Yi: "On a 5-10 year timescale, I think it's safe to assume that the cost of proving and latency will drop to very close to zero. There will always be some overhead over native execution, but for protocol design, we can assume the overhead will be small, maybe 10-100x in cost and potentially real-time latency.

The biggest unlock is the ability to offer more synchronous composability between different execution zones. If you think about the rollup-centric roadmap, you can have different levels of decentralization and censorship resistance on different sequencers. ZK enables faster cross-rollup assertions than L1 block time, allowing L2s to establish more trust-minimized communication protocols."

Me: "Looking at the current landscape, there are a few potential paths for the Ethereum roadmap to play out. What do you think about these different paths?"

Yi: "I think the future of Ethereum is likely to be path-dependent. I believe there will be multiple rollups because there's a fundamental human desire for diversity. There are two key trade-off dimensions: sequencer decentralization versus throughput/cost, and differentiation in execution environments and VM functionality.

We're already seeing alternative VM rollups like Eclipse launch. As ZK makes interoperability easier, rollups will need to differentiate more to capture market share. They'll compete both on the social level with different applications and ecosystems, and on the technical level with different features and execution environments."

Me: "After such a long academic career, what made you decide to start Axiom? Was it the market opportunity, or is there a deeper mission you want to see realized?"

Yi: "It wasn't strictly seeking out market opportunities. I've been looking for places where it's possible to make a more direct impact and shape the course of technology more than is possible in academia. Academia is really suited for theoretical problems where you have a blank canvas exploration, but the impact isn't always clear.

In ZK specifically, I felt we were at a really interesting transition point, similar to what happened in AI around a decade ago. Previously, AI was the domain of university labs. Then with Google Brain, OpenAI, and now industry labs deploying it, we're seeing much more rapid progress through this fusion of research and engineering.

With Axiom, I felt like this was going to start happening in ZK in late 2022. We're seeing a transition in cryptography from very custom protocols to more programmable and general-purpose technologies."

Me: “With AGI (artificial general intelligence) seemingly around the corner, what are your thoughts on the future?”

Yi: "I used to work in AI — though back then we still called it deep learning. It's been impressive watching the progress from the outside. As these models get better and AI agents can do more and more, it calls into question what we consider uniquely human.

Take software engineering — it used to be this sought-after, unique domain. Now, you can just ask Claude or ChatGPT, and it has had a commoditizing effect. But I think that there's a balance between ‘this is a commodity we can use to replace humans’ versus the ability to enhance humans and make the strongest engineers even more productive.

In terms of the role of cryptography in that, I do think that AI creates an oversupply of things that we used to consider human. This actually increases the need for digital scarcity. In the past, if you saw a photo that looked real, you knew it was real. Soon, that won't be true, but we'll still want ways to socially determine what content we value.”

Me: “And to finish up, can you drop advice for the young people out there looking to enter the world of work with so many non-linear technology changes on the horizon.

Yi: “My advice would be two things: First, don't assume AGI will be all-powerful. Whatever form AI takes, it won't be a perfect, sentient being. It'll have idiosyncrasies, and humans will retain some comparative advantages. Second, AI will de-emphasize purely technical skills and emphasize social skills more. Cryptography is actually the most social form of technical skill – a protocol is only relevant if humans accept it. I think it's one of the technical areas least threatened by AI."

Narratives change, keywords change, but the fundamentals of technology shifts will span over decades and reward those who are persistent and long term focused. I’m looking forward to the next few years as ZK becomes much more mature and usable, many billion dollar projects will emerge, and many ways to make the world a more open and interoperable place using applied cryptography.

Follow Yi on Twitter: @theyisun

https://www.axiom.xyz/

“Just as AI evolved from university labs to industry deployment with the emergence of Google Brain and OpenAI, ZK technology is now transitioning from custom protocols to programmable, general-purpose tools that could reshape how we build blockchain infrastructure.” - Yi Sun, Axiom

Earlier this week, I caught up with Yi Sun, CEO of Axiom – a company focused on building the open infrastructure to help transform ZK from a niche pursuit to an industry-defining force.

Having worked closely with Yi as an advisor to Scroll over the past three years, and collaborating closely building the OpenVM project. I've had the privilege of watching Axiom's journey from the beginning. In late 2022, Yi saw an opportunity to push ZK beyond theoretical cryptography into a practical, general-purpose technology that could have massive impact — an opportunity he felt was so significant that even a small company could help shape the future of the field.

Here's our conversation:

Yi: "I had a relatively longer career in academia. I did a math PhD and went on to become a statistics professor at the University of Chicago, where I was doing research in theoretical probability and high-dimensional statistics. It wasn’t until 2017 that I got into crypto - what drew me to it was how it seemed like the conjunction of my theoretical research interests with a much more practical, real-world impact.

For several years, I stayed tangentially involved in the space, helping out early-stage startups as an advisor, including Gauntlet and, of course, Scroll. In 2021, I started getting more directly involved when I saw that ZK technology was becoming practical enough for real applications."

Me: "ZK went from being the hottest keyword in crypto to being marketing poison. We've seen these narrative pendulum swings before, how do you see this playing out?"

Yi: "ZK, just like many other trends in crypto, often goes through phases where the hype and speculation can be well ahead of the technology. The technology goes through phases where it's overestimated and then underestimated. But probably today, I would say it's a little underestimated.

The progress has been pretty amazing. Back in 2014-2015, when programmable ZK proofs were just getting started, the overhead for generating a zero-knowledge proof compared to running it on your laptop was something outlandish, around 10^10. Now that overhead is around 10^4, which is much more reasonable for real applications."

Me: "Axiom’s success depends on more companies using ZK in real world applications, can you give me three billion-dollar ideas that could be enabled by ZK technology. Where are the business opportunities, and what are the blockers?"

Yi: "The first general category would be rethinking how you build blockchain infrastructure from first principles, given that ZK exists. Justin Drake announced at DEVCON that he wants to do this for the Ethereum consensus layer. This applies broadly to rollups, bridges, and any sort of intent-based protocols.

At the base layer, ZK can be helpful for synchronous composability, cross-roll-up bridging, and communication with much shorter finality. I think the reason we haven't seen as much of that is that, although ZK has been difficult to build with historically, those categories are pretty challenging to deliver any product in, even without ZK.

So, the adoption timeframe needs to be matched up with what's possible technologically in ZK. For example, if you start to build a rollup, it's going to take at minimum 12 to 18 months. And so those roll-ups that are coming to production today were conceived of about a year, a year and a half ago. As we see roll-ups really rethinking themselves from first principles based on the technological upgrades in ZK, we're going to see improved performance, lower costs, and greater composability.

The second category I'm excited about is using ZK for authentication, which is interpreted pretty generally. There are many different cryptographic formats of data across the world, both within blockchains and outside them. Every blockchain has some signature scheme, but there are also many types of signatures — OAuth, email signatures, etc. ZK can translate between these formats seamlessly.

The third category is privacy applications. This has lagged behind computational integrity uses because privacy use cases require proving on the client side, which is much more constrained. But I think performance will soon reach a point where we can deploy real privacy-preserving applications."

Me: "Can we break down the first category into business cases that application builders can understand? How do you see this playing out on a 5-10 year horizon, assuming we get to 1 billion people on-chain doing 10 transactions per day?"

Yi: "On a 5-10 year timescale, I think it's safe to assume that the cost of proving and latency will drop to very close to zero. There will always be some overhead over native execution, but for protocol design, we can assume the overhead will be small, maybe 10-100x in cost and potentially real-time latency.

The biggest unlock is the ability to offer more synchronous composability between different execution zones. If you think about the rollup-centric roadmap, you can have different levels of decentralization and censorship resistance on different sequencers. ZK enables faster cross-rollup assertions than L1 block time, allowing L2s to establish more trust-minimized communication protocols."

Me: "Looking at the current landscape, there are a few potential paths for the Ethereum roadmap to play out. What do you think about these different paths?"

Yi: "I think the future of Ethereum is likely to be path-dependent. I believe there will be multiple rollups because there's a fundamental human desire for diversity. There are two key trade-off dimensions: sequencer decentralization versus throughput/cost, and differentiation in execution environments and VM functionality.

We're already seeing alternative VM rollups like Eclipse launch. As ZK makes interoperability easier, rollups will need to differentiate more to capture market share. They'll compete both on the social level with different applications and ecosystems, and on the technical level with different features and execution environments."

Me: "After such a long academic career, what made you decide to start Axiom? Was it the market opportunity, or is there a deeper mission you want to see realized?"

Yi: "It wasn't strictly seeking out market opportunities. I've been looking for places where it's possible to make a more direct impact and shape the course of technology more than is possible in academia. Academia is really suited for theoretical problems where you have a blank canvas exploration, but the impact isn't always clear.

In ZK specifically, I felt we were at a really interesting transition point, similar to what happened in AI around a decade ago. Previously, AI was the domain of university labs. Then with Google Brain, OpenAI, and now industry labs deploying it, we're seeing much more rapid progress through this fusion of research and engineering.

With Axiom, I felt like this was going to start happening in ZK in late 2022. We're seeing a transition in cryptography from very custom protocols to more programmable and general-purpose technologies."

Me: “With AGI (artificial general intelligence) seemingly around the corner, what are your thoughts on the future?”

Yi: "I used to work in AI — though back then we still called it deep learning. It's been impressive watching the progress from the outside. As these models get better and AI agents can do more and more, it calls into question what we consider uniquely human.

Take software engineering — it used to be this sought-after, unique domain. Now, you can just ask Claude or ChatGPT, and it has had a commoditizing effect. But I think that there's a balance between ‘this is a commodity we can use to replace humans’ versus the ability to enhance humans and make the strongest engineers even more productive.

In terms of the role of cryptography in that, I do think that AI creates an oversupply of things that we used to consider human. This actually increases the need for digital scarcity. In the past, if you saw a photo that looked real, you knew it was real. Soon, that won't be true, but we'll still want ways to socially determine what content we value.”

Me: “And to finish up, can you drop advice for the young people out there looking to enter the world of work with so many non-linear technology changes on the horizon.

Yi: “My advice would be two things: First, don't assume AGI will be all-powerful. Whatever form AI takes, it won't be a perfect, sentient being. It'll have idiosyncrasies, and humans will retain some comparative advantages. Second, AI will de-emphasize purely technical skills and emphasize social skills more. Cryptography is actually the most social form of technical skill – a protocol is only relevant if humans accept it. I think it's one of the technical areas least threatened by AI."

Narratives change, keywords change, but the fundamentals of technology shifts will span over decades and reward those who are persistent and long term focused. I’m looking forward to the next few years as ZK becomes much more mature and usable, many billion dollar projects will emerge, and many ways to make the world a more open and interoperable place using applied cryptography.

Follow Yi on Twitter: @theyisun

https://www.axiom.xyz/