Watch the full video of this panel discussion on Gamma Prime’s YouTube channel
Panel discussion with Michael Terpin (Founder and CEO of Transform Group), Christian Sauer (Founder of soonami.io), Catrina Wang (General Partner at Portal Ventures), and Amit Mehra (Partner at Borderless Capital)
Host
Blockchain, driving innovation and investment in web3. Let’s try to answer and disclose this topic on a separate note and feed our audience with key insights.
So, before we dive into the topic, please introduce yourself to our audience and give us two or three key facts why we should listen to you for the next… 30 minutes, Yuri? Yuri, 30 minutes? Okay, let’s go. Michael, do you want to be the first?
Michael
Hi, I’m Michael Terpin. I’ve been in Bitcoin and crypto since early 2013. I started Bit Angels, which is in 30 cities right now, and the Tokenized conferences. I’ve also worked with almost 500 projects, going back to the early days of helping Ethereum and Tether launch. These days, I’m working on quite a few decentralized AI projects. I think it’s one of the three big narratives for this wave, and I’ll talk about that a little on the panel.
Chris
Hi, I’m Chris from Soonami. We’re a tokenized ecosystem for startups. We started this during COVID and thought there needs to be a supportive structure for founders, borderless and permissionless, and that’s what we built. We just launched our token and we’re on Polygon, so have a look. Happy to be here.
Catrina
Hi everyone, my name is Catrina. I’m a general partner of Portal Ventures. We are a $120 million fund dedicated to backing first checks into pre-seed companies. How I got into crypto was back in 2017. I helped Deloitte Strategy, when I was in management consulting, launch its global blockchain practice. Before joining PortaI worked at Protocol Labs. They were quite early to AI Web3 and I started writing about AI Web3 thesis in 2023.
Amit
Hey, I’m Amit. I’m a partner with Borderless Capital. I was the first hire and became the first partner. We have both a VC and a liquid strategy. On the VC side, we currently have two active funds. One focuses on cross-chain interoperability, working with more than 15 L1s closely as LPs and with Wormhole. The other focuses on DePIN, where we are actively deploying into AI-Web3 as one of the active categories. We are also in the process of launching a growth fund, and the other is a long-biased liquid fund.
Host
Let’s start with the first question. Before we dive into difficult topics let’s start with the primitives. It took us over 10 years to learn what blockchain is, what interoperability is, oracles in blockchain, and now AI. It took us another few years to learn what LLMs and AGI are. What type of primitives should we know, should we be aware of, to dive deeper into AI?
Catrina
I’m happy to start. The building blocks, abstracted away from crypto, for AI are primarily three things. One is the algorithm, the second is data, and the third is GPU power. As you start thinking about the intersection, look into these building blocks fundamental to AI and where crypto can actually help. A couple of use cases we’ve seen over the past year or two: on the GPU side, crypto gives a good mechanism to crowdsource GPU supply. Nvidia and big tech dominate the supply of GPU compute power, whereas in a decentralized crypto world everyone can contribute their spare compute power. One of our portfolio companies is generating $10 million in annual contract value just by decentralizing and providing cloud compute. It’s giving at least 80% cost savings compared to centralized GPU providers. That’s one of the pillars. I’ll let the others chime in, but the three foundational building blocks are important to know.
Michael
So AI right now is growing much faster than crypto. Crypto showed us the way of decentralization, tokenization, and immutability, but adoption has been slow. We’re 16 years in now and we don’t even have half a billion people. AI got there in a year. AI is solving problems very quickly. At the end of the day, if AI is already at a billion and a half people and reaches 50% of the world in two or three years, which I think it will, Bitcoin adoption might still take another 10 years. Put the two together — do you really want your data and agents owned by Microsoft or Google, or would you like to control them yourself? That’s the promise of decentralized AI and decentralized agents.
I’m working on a few projects I can talk about a little bit, but I’ll tell you one I’m not involved in because they didn’t raise any money at all, but it sets the bar very high. Anybody read about Base 44 this week? Base 44 was a one-person startup. His prior company had raised $125 million from VCs, had 100 employees, several rounds. Then he had to leave because he was a young Israeli drafted into the army. He came back, was replaced as CEO, and had a new idea. His girlfriend was trying to build a website with AI and found it clunky. So he asked, how do I use Claude and other tools to make it easier to just write what you want and have it build a website?
He started this company in January. He sold it for $80 million this month — five and a half months after starting. He had no employees, just coded nonstop, posted results online, used social media to share successes and failures, and his beta testers did the same. He ended up with a quarter of a million users, $3.5 million in recurring revenue, and sold to Wix, which has 180 million users. Sold for $80 million plus upside.
That’s what we need to look at for decentralized AI to take off — making AI easy for people to onboard. We’re not there yet.
Chris
I fully support that. We started four years ago with our investment thesis investing in Web3 and AI. I also believe that Web3 has a little problem at the moment, looking at the UX, looking at how everybody’s building rails, and they don’t really match. The trains are not going on these rails, and that is a problem. We wanted to have Web3 and AI connected in these projects that we’re supporting, but 80% of the founders that we’re supporting at the moment are either in Web3 or in AI and are not matching the two technologies together, where it would be so powerful to develop the next level of usability for Web3 as well.
I have the fear that at the moment Web2 is more or less taking over Web3 because they just have better usability and use the AI tools, and the industry is a little bit at a loss of what to do.
Amit
Maybe building upon what Christian, Michael, and Catrina said, I think definitely there’s a friction between hardcore AI folks and hardcore Web3 folks. I think it’s very similar to what you are seeing in gaming. Gaming also used to be like: are you really a gamer, or does Web3 fit in here, or does it create more friction? So it will be interesting to see how it evolves.
Building upon what Catrina said earlier — infrastructure — the other two pillars we are very bullish about. One is training. Even with these big LLMs like Claude and ChatGPT, the idea is that they need to be refined. There is just not enough money or power to do it in a centralized way, so the solution to that is decentralization.
For example, one of our protocols, Bagel, is like Hugging Face in a Web3 way. How you monetize ML researchers and creators to go and fine-tune these models which they are putting out, and incentivize them in a Web3 way. That is a very beautiful use case where the combination really makes a lot of sense.
The other is infrastructure — how you actually benchmark how these models are performing in specific use cases, how GPT is performing, how Claude is performing, and actually do it. Because these models are so smart that once they realize how you are tracking or gauging them, they can perform in a way that they come to the top. So how do you do it in an unbiased Web3 way?
There’s a company, one of our protocols, called Layer Lens, which is working on benchmarking the infrastructure piece. Those are some interesting primitives which are evolving.
Host
Yesterday I went to AGI conference, and we had a really interesting talk with Sentient about fractionalization of AI in Web3. Sentient promises that the AI agent will be able to execute any trading operations through protocols like Jensen. Michael mentioned this startup I’ve never heard about, Base 44.
What other key names are worth mentioning at this panel — who drives the development of AI in Web3 apart from Jensen, Sentient, BitTensor? Maybe you have something to add.
Michael
Those are the base platforms. BitTensor I think is still way undervalued, and I keep picking up more as it drops below 300 or 330. It’s a no-brainer that it’s gonna be 10,000 in the next couple of years.
Host
Do you hold BitTensor? Do you hold TAO?
Michael
I’m sorry?
Host
Do you hold TAO?
Michael
Yeah, of course. That’s one. Also Morpheus — I think Morpheus is wildly undervalued right now. You’re having this Bitcoin cycle highs. The Bitcoin dominance has been up to 65–66%. Coupled with stablecoin growth also getting crushed, alts are down and the AI tokens are down, in some cases 80–90% from their December highs.
I think it’s just the narrative. Mentor talked about this a little bit earlier. The narrative for making money right now is: buy stock in a US company that is going to buy Bitcoin for 1x and then go public for 2 or 3x. It seems like easy money until all of a sudden, like many other things, it gets oversubscribed.
I wrote the book Bitcoin Supercycle, where I talk about fairly predictable trends. There’s only been three years in the history of Bitcoin you could have lost money: 2014, 2018, 2022. These are all two years after the halving in the fourth quarter. That’s going to happen again in 2026.
So all these long-only Treasury companies that don’t hedge their bets are going off a cliff, and people will be shorting MicroStrategy — they’re already doing that. That’s when Jim Cramer will come on TV again and say: “Sell all your Bitcoin, it’s going to zero, it’s a bad asset class.” And of course, that’s the best time to buy.
I think the next layer — the application layer — is really where the action is right now. I’ve got a few investments in that area. We just closed some funding — we still have a little bit left in the round — but it’s called AI Quant Labs. It’s aiquant.fun, and it lets you use AI agents to do DEX trading.
You set up your parameters, it costs about a tenth of an ETH to set them up, and it analyzes the smart contracts of 100,000 coins a day and picks the 30 you should be trading given your strategy. So far, the average beta tester is up. We take 10% of the profitable trades. Our estimates are we’ll be able to get about $4,000 per user per year with very low costs. At 10,000 users, that’s $40 million a year, and at 100,000 we’re already built to scale up to unicorn-plus.
Host
Once again, repeat the name of the company?
Michael
AI Quant Labs. It’s aiquant.fun.
Host
Is it investment advice?
Michael
I invested. I’m just giving some alpha.
Host
Thank you. Kate, any names we should be aware of?
Catrina
Well, to be honest, everyone will turn to AI. A year before it was doing more of an OS narrative, now it’s AI OS. Polygon has AI in there. Everyone wants to dovetail into what’s trendy. So it wouldn’t surprise me if a lot of these big names become AI.
Ultimately, as an investor, it’s about parsing through the noise. Which company actually needs to do it? I talked to Prime Intellect, which is actually doing interesting things. Recently, the founders won, and not sharing my own bag but someone else’s.
In general, it’s important to figure out what’s a real AI company authentic to their core value proposition, and what’s just riding a trend.
Host
Can you specify a real AI company? According to the graphic simulator, we have two namings: one is Tesla — companies not very well promoted but with core technology. And Edison — companies with questionable technology but really good promotion. From your perspective at Portal Ventures, how do you evaluate AI startups?
Catrina
How do I what?
Host
Evaluate. How do you do due diligence? What are key signs that this is a good AI startup for you?
Catrina
It’s interesting. I wrote about AI OS back in 2023, but from then to now we only made one investment — Exhibit. That’s generating more than 10 million, and…
Host
This was my angel investment.
Catrina
Wonderful! The team had real traction, whereas a lot of the agentic narrative was just launching verification. Let me make an important distinction. The fact we didn’t invest or it didn’t pass our evaluation is more because of the business model, not because the technology isn’t important.
As an investor, the business model is the most important. The challenge I have with the intersection of Web3 and AI is: how do they monetize this? What is the willingness to pay? For example, verification. In decentralized AI it’s very important: I need to verify if this is a deepfake, if the output is coherent with the input. But just like privacy, everyone screams it’s super important, but when it comes to paying for privacy, most people say no.
So the reason we didn’t invest that much in Web3 is I still struggle with this model. Some tech needs to be built, but you need to look for the Venn diagram of what’s commercial and what has product market fit.
Host
Amit, do you have something to add?
Amit
I think the only other very interesting area coming up is agentic AI and all the infrastructure that has been built. Eventually, the idea is entire funds, for example, and the most prime for disruption is fintech. How can you have an entire hedge fund? There have been several cases where an agent was picking tokens, investing in tokens, and generating very good returns, more than actual human beings. So I think the two more base protocols like MPC, what Coinbase released X402, and a lot of startups that we’ve spoken to, are building upon these protocols to work on agent-to-agent payments, agent-to-agent communication. Those would be the other.
Chris
I like the Morpheus by the way, that’s an interesting project. They have a very interesting way of funding the whole infrastructure by a fair launch model supported by staked Ethereum. We took that model and want to offer it to pretty much any other company out there. To me this is really an interesting name.
Of course, the issue with the Morpheus funding model, which worked brilliantly for them, was that the thesis with Morpheus was the first time it was done. It’s been replicated by a few of their portfolio, their ecosystem. You stake your Ethereum on Lido and contribute the interest to the project, and they ended up having 400 million dollars staked. The main people behind the project were David Johnston and Eric Voorhees. Most projects don’t have iconic billionaire founders that could be staking that entire amount themselves, and then other people are like, “oh, if they’re building this.”
When you have a newer startup, if somebody stakes a million dollars you only get 30,000 funding. When it becomes 400 million all of a sudden you’re talking about money. It’s difficult to do that, but there are some interesting models. Morpheus ended up spinning out of Venice, which is Eric Voorhees’ decentralized AI platform, and then that spun out the Venice token VVV. That first day it went over a billion dollars. It’s pulled back, but I think it’s still a good project that will probably recover. I’m bullish on that one because it has use cases. You use the VVV token to buy inference on the Venice search. You can use it as web2 and also as web3.
I think there aren’t that many hybrids where it’s easy to come in with zero crypto knowledge, but also more profitable to utilize with web3. That’s what’s missing right now. Hopefully in the next three to five years, most projects in decentralized AI will have very easy onboarding for the web2 user, or they just won’t be competitive.
Host
Michael, coming back to your vision up to 2030, five or six years from now — what is overhyped right now, and what is underhyped? What should we pay attention to, and what should we slow down?
Chris
I think overhyped is a little bit…
Host
AI and meme coins?
Chris
AI and making money, and underhype to me is the purpose. There was this nice book Start With Why. For a lot of people, the why right now is just “I want to make money,” and I think that doesn’t really help. I’d love to see more purpose in projects, to really understand why they are there and how society benefits. That’s how I entered the market five years ago when we all wanted to build a decentralized layer for humanity. Now we end up with meme coins. That’s the most underrated part of the industry.
Michael
I’d say overrated would be layer ones and layer twos that are specific to AI. I don’t think you need new layer ones, I don’t think you need new layer twos only to do what the other hundred layer twos are doing, but with different coins instead of Polygon’s.
Underhyped is probably that 90% of the agents are going to die, but the 10% that survive are going to be giant and used by millions of people profitably in the next five years. In five years from now, you’ll be doing hundreds of transactions a day — whether trading like with AI Quant, or with another portfolio company, Jacques Voorhees’ Icecap AI, a shopping bot that answers all the questions like a salesperson. It increases lift of sales: a physical store has 40% sales conversion, an online store has 4%. Because it’s a horrible experience if you have questions — you look in a database, a help file, maybe a bad bot. But if you have a human-sounding agent, not just a chatbot, that helps you through the sale and is trained like a salesperson — “oh, you’re looking at that, I have another color, want to see how it looks? We have a sale, want to know about it?” — all it has to do is raise the sales rate by 2%. That’s a 50% increase from 4 to 6% conversion.
That’s another use case I’m bullish on: AI sales agents.
Catrina
What’s overhyped? In the context of crypto, autonomous agentic trading. I’ve seen many trading decks doing that, but realistically as humans we still want control. Imagine you have an agent doing all crypto trading for you with no say — who’s going to do that? We want co-pilots, not full autonomy. From a technical standpoint, agents are largely non-deterministic, whereas a smart contract is deterministic. That mismatch can cause issues.
What’s underhyped? I think there’s real opportunity with data contribution. Before, why would I contribute my data for training OpenAI is doing? But now, with token incentives and on-chain verifiability and ownership tracking, I can actually get a share if they develop an LLM based on my contribution. Then there is proof I want to share this. In data contribution, I haven’t seen many credible startups, but it would be interesting.
Amit
I think generally, within the firm, I’m not very excited on the infrastructure side. One of the things we see is startups using VC money to buy GPUs, data centers. It’s very capital intensive but not scalable. We’re more bullish on the challenges centralized LLM companies are facing. There are things in AI that can’t be done centrally — refining models, data attribution, governance. I think companies hitting those ideas are underhyped and they’ll really grow big.
Chris
Maybe one thought: with the purpose I mentioned before, imagine if everybody used AI agentic trading and all these AI agents fought each other. That’s interesting. Another thought — using AI for marketing. It’s easy now to do all marketing and LinkedIn posts with AI. What does that mean? We only have a certain number of eyeballs, a limited amount of attention. Additional marketing power from AI will dilute the actual eyeballs viewing it. This will happen in many markets. AI enables something, but if everybody can do it, it dilutes itself.
Host
That’s a topic for another big conference. As media, we question ourselves — how do we stay authentic and unique?
Michael
I just wanted to defend AI agentic trading. The number of wallets has increased from 70 million last year to 200 million DEX wallets. I agree you don’t want agents learning on their own, fighting each other. With AI Quant, you program it, it’s your responsibility. At the top of our leaderboard, someone made 660% profit last month; the bottom lost money. If you’re losing money, you adjust parameters — tokens from 50,000 to 10 million, or 10 million to a billion — until you find a formula that works. Strategies at the top are hidden, but you can buy them and copy-trade for a fee we split with the creator. That’s a winning model, and I’m very bullish on it.
I also agree on not using VC money to buy chips. Chamath Palihapitiya on the All In podcast said any VC who greenlights a $50 million check for chips at a $1 billion valuation should be fired. Chips will be 90% off in a year, and you’ve already put the money into hardware. Instead, put it into software engineers, then use your valuation to take out debt and buy chips. Nvidia is happy to give you payment terms.
Host
We are wrapping up the panel, so I want to finish with a funny, down-to-earth, simple question. I just arrived from San Francisco. I visited all demo days like A6NZ, Y Combinator, HF0, and the new trend is that each company has to pass an AI test.
What it means is that all those accelerators and VCs are using AI for internal startup optimization. So I want to highlight just two AI tools that are must-haves from Silicon Valley. One is Superhuman — a lot of sales teams are required to use it daily, though not the C-levels because we need to keep an authentic voice of tone. Another one is Upgrade, a recent alumni of the HF0 accelerator. They’re trying to answer the question: what marketing activity actually brought users, payments, etc. So very down-to-earth — one or two AI applications that are must-use for any successful startup?
Catrina
Superhuman is great. Granola is strong for AI note-taking. Vimco is great for calendar — like a better version of Calendly. If you don’t want to use Calendly (which is also good, though not exactly AI), Vimco is an option. Notion actually has a great co-pilot to help you edit, and since then everyone’s been building similar tools, but they had the first-mover advantage.
Michael
I’m just looking for things that are not ultimately going to be sold to Google or Microsoft, where they own all my data and learn on it. I want to have things I control, and that’s the promise of decentralization in DeFi — and now also the promise of decentralization in AI.
I’ve mentioned sales agents, I’ve mentioned sales trading. The last portfolio company I’ll mention is Alpha AI. The founder is the former head of strategy for NASDAQ, and he’s building an agentic Bloomberg terminal, both physical and on-chain, that lets you plug in a range of decentralized modules for research and trading.
Chris
We have a couple of interesting AIs in our portfolio. One is called Mind Keeper — if there’s a transition from one employee to another, it allows the new employee to ask the AI what the other person actually did. That’s a useful case. Then we have Andrual (or Dearflo), which provides automated email inbox support.
Amit
I think a lot of things come up. Maybe, other than that, is Cursor.
And eventually what would be very useful is benchmarking these tools. We have a portfolio, and others like Layer Lens, to benchmark which model actually does well.
Host
I think we are ready with a post for Cointelegraph with your advice. Thank you, Yuri.
Thank you, our speakers. Please applaud. Thank you.
