# How Bittensor (TAO) is Tokenizing Intelligence (Mark Jeffrey) Thread Guy · 2026-03-27 · 29 min · https://www.youtube.com/watch?v=PDj7g--p-qE ## Transcript Thread Guy We have Mark coming on. He wants to tell funds with Jason Calacanis and Rob. Really sharp guys. Obviously everybody knows Jason and he's pretty like generational. I don't know. I heard about a pod. I I was really into it. He's a really sharp guy and so he's going to come on. He can talk about crypto, Tao, all of it. All right, let's get it. Boom. Mark Jeffrey, welcome to the stream, man. How are you? Mark Jeffrey I'm doing well, sir. Thanks for having me. Thread Guy Likewise. Thanks for coming on, man. It's an absolute pleasure. We're all kind of all over the place. There's a lot happening. We're talking Trump and geopolitics. We haven't been doing a lot of crypto, but over the last 2 or 3 weeks, Bittensor TAO has been all over my feed. It's honestly been refreshing to tap back in, talk about some crypto, and talk about everything going on in that world. Happy to have you. Welcome. Do you want to start with a quick introduction to who you are, and then we can talk about some fun market stuff? Mark Jeffrey Sure. I've been kicking around for about 30 years doing tech companies. I'm a little older, and I founded my first company in 1995. I was an engineer to begin with, so I was around for the dot-com boom and bust and worked with Jason Calacanis in those days. That's how I know him from that era. Then I founded a bunch of companies and ended up being CTO at a company that Jason founded. Elon was on the board, so I used to have to give him a report once a quarter. Thread Guy Really? Directly to him? Mark Jeffrey Yeah, when he showed up at the board meetings. He was like, "Okay, Mark. Okay, CTO guy. Tell me what you're doing." He'd pull it apart, of course, because it's Elon. I'd have to dance and answer questions. Sequoia was on the board, too, so we got some face time with unbelievably amazing people. It was like Warhol's loft. All these characters came through. You have to remember, back then Elon wasn't Elon. He was just a PayPal guy. He was successful, but he wasn't the Messiah. We didn't know all the things that were coming, so it wasn't quite like addressing the throne the way it would be now. Thread Guy Were you conscious, at the time, that Elon was actually going to be the GOAT, or not really? Mark Jeffrey No. I was very conscious of the fact that he was a very smart guy, but so were the Sequoia guys, and so were a lot of other people. I also worked briefly with a guy named Travis Kalanick, who founded Uber. Thread Guy Whoa. Mark Jeffrey Before Uber, Travis had a company called Red Swoosh. This was after the dot-com boom and bust, so I was looking around for a job, and there weren't many jobs. Travis said, "Hey, come work for me." I went and worked at Red Swoosh for about a year and a half. It was just me, Travis, and 2 or 3 other guys in a room for about a year and a half. There’s a great line in Amadeus where Salieri meets Mozart and doesn’t know what he looks like. He says, "I played a game with myself. Is genius visible on the face?" For me, the answer was no. I was hanging out with Travis every day, and I didn't realize who I was hanging out with until later. Thread Guy We might have to work on this identifying-talent thing. I don't know. Mark Jeffrey Yeah. Thread Guy You ended up with that. That's crazy lore. What a story. That's awesome. Mark Jeffrey I'll just finish it real quick. I got into Bitcoin in 2013, hence the Bitcoin books behind me. I wrote a couple of Bitcoin books, and then I was in the Ethereum ICO. I interviewed Vitalik on a podcast right before it happened, so I got in on that. I was there for DeFi Summer. I've been kicking around in crypto basically since 2013. Thread Guy Okay, cool. You're super-early Bitcoin, and you're running— Mark Jeffrey And then today, right? I have to talk about today. Right now, I'm a partner at Stillmark Capital, and we're a fund investing in Bittensor subnets. Thread Guy Beautiful. That's kind of a crazy move. We see a lot of Bitcoin funds and ETH funds. We had the DOT mania for a little while, which got pretty far out on the risk curve: Berachain, DOT, AVAX, and Hyperliquid. I'm not sure I've seen a Bittensor fund like this. It's a unique ecosystem because you have this subnet ecosystem, which we actually haven't talked that much about on the stream. How did you get TAO-pilled, and why are you working on this? Mark Jeffrey It was about 2 years ago. I read a report by Messari, written by a guy named Sammy Keseb. Thread Guy Oh, hell yeah. He's been on the stream. Mark Jeffrey Okay, so you know him. He himself is also doing a Bittensor fund. His paper started me suspecting that there was something very important going on here. As I said, I was there early for Bitcoin in 2013 and then again for Ethereum. The more I read about it and understood what was going on, the more I started gaining conviction, over a couple of months, that this was the third great coin: Bitcoin, Ethereum, and then this. This is the most important thing that's happened since Ethereum. Having bought my first Bitcoin on Mt. Gox for $250 back in the day and seeing how that spiraled up, and then seeing the same thing with Ethereum, my opinion quickly became: it's going to happen again, and it's going to happen with this Bittensor TAO thing. How do I get as much TAO as possible? How do I become like the Winklevosses and own 1% of Bitcoin? Of this TAO thing. That was the thinking. I have my own stack from my Bitcoin and Ethereum adventures, but how can I get more? Running a fund was the best way to do that. I had a couple of partners who were thinking about doing it approach me, and then I approached Jason about investing in the fund. Jason Calacanis came back and said, "Can I be a partner in the fund?" We quickly said yes to that. Thread Guy Why is it the most important crypto since ETH? Mark Jeffrey If you look at what Bitcoin did and what Ethereum did, Ethereum looked at Bitcoin and said, "This is super interesting, but this little tiny smart-contracts thing—let's blow that up. Let's make a Turing-complete machine. Let's make the world computer," as Vitalik was calling it. It was a great idea, and it led first to the explosion of ICOs and then to DeFi Summer. We got decentralized lending, decentralized exchanges, and stablecoins. There was this whole rich Cambrian explosion that happened inside the Ethereum and EVM ecosystem. What Bittensor does is take the other side of Bitcoin—the mining side—and make that programmable. It allows the Bittensor chain to subsidize all these AI and other digital commodities into existence and lets this whole other Cambrian explosion happen. I fully buy into the theory that this was going to happen, and it started to happen. The reason you've been hearing about Bittensor so much is that the successes are now becoming obvious to the rest of the world for the first time. Templar Subnet 3 just announced that it completed a 72-billion-parameter training run of a frontier model. It was fully decentralized. So, what does that mean? You and I, if we want to make a new ChatGPT or a new Grok, have to have 200,000 GPUs, which is what Elon has. Thread Guy Disgusting. Colossus. Mark Jeffrey Imagine a Napsterized Colossus coordinated and incentivized by the emissions in a Bittensor subnet called Templar Subnet 3. A lot of people thought it couldn't be done. You could not do AI training in that fashion. Subnet 3 just proved they could do it. It got a huge amount of attention—2 million views on the tweet. They talked about it on the All-In podcast. Chamath brought it up on the All-In podcast while talking to Jensen, and that clip went viral. The CEO, or one of the co-founders, has talked about it twice now and put Templar and that training run in the paper he wrote, calling it out as, "This is the world's largest decentralized training run. We didn't really think this was possible. Now that we know it is, in the future…" The other thing to remember about this is: what is Bitcoin? It's a lot of things, but it's also Earth's largest supercomputer. It's 500 times larger than its nearest competitor. What do we need to train AI? Earth's largest supercomputer. So, how did Bitcoin grow? It incentivized itself into existence. It convinced all these people to donate CPU and electricity to try to mine these coins. Let's do that. It worked once. Let's do it again for AI and all these other projects that need intense amounts of GPU. One of the tweets I saw and liked was from a guy who also used to work at Messari. He tweeted in response, basically saying that the idea of a decentralized training run was initially thought impossible. Now that it's been done, the discourse has switched to saying it's too small, which is a real step-function forward. Thread Guy On a different note, one of my questions for you is: We trade crypto. I trade a lot of crypto, and not all crypto is garbage, but you trade a lot of things based on attention. Not fundamentals, not P/E ratios, not forward earnings—a lot of attention. What was interesting about Bittensor, and why I’ve been talking about it, is that I think it’s the first crypto in a while that has gotten this mainstream tech and Silicon Valley attention. Obviously Jason Calacanis, but then you see Chamath talking about the Templar run to Jensen, and you see—was it Jack Joseph or Joseph Jack, one of these AI guys? Mark Jeffrey J.J. Thread Guy J.J., who’s not a crypto person. If anything, they’re anti-crypto. They’re like, “Okay, cool. TAO.” Why are these tech-adjacent, anti-crypto people talking about Bittensor? What just happened? Mark Jeffrey I think they realized that Bitcoin’s success could potentially be replicated for AI—for doing open-source AI. Bittensor is, in a lot of ways, OpenAI. It’s what OpenAI, the company, should have been but did not become. Part of the reason for that is that with open projects, like open software projects—Linux, maybe, or something like that—the early contributors don’t get anything. There’s no ownership layer. Bitcoin was the first open-source project that had an ownership layer from day one. If you were just some kid in your bedroom mining coins in 2010, you got coins, which meant you had an ownership layer. You were invested in Bitcoin’s success and its future. It’s not stock, but it’s like stock in the sense that it’s ownership. By making mining programmable, Bittensor is basically saying, “Let’s do Bitcoin again.” It has the same economics: 21 million coins in the TAO token, the base layer. It’s also true that there are 128 subnets, each one a different AI project with its own mining contest. They’re all totally different, but each of those subnets also has a coin. Again, those are limited to 21 million. Just like Bitcoin, they’re emitted into a liquidity pool. The only way you can get subnet tokens—and this is extremely important—is by staking TAO into the liquidity pool, which is like a Uniswap system baked into the chain. It’s not an app on top; it’s inside the chain. What that means is that all of the subnet tokens are rehypothecated TAO. It’s another form of the same thing. The subnet-token pools can grow and shrink depending on how well that subnet does in the marketplace, just like any marketplace. But the subnet token itself is very different from Ethereum, Solana, and everything else. They build smart contracts and tokens on top of the gas token, but there is no economic connection. These are tied at the hip. It’s a very different dynamic. If you look at Bitcoin and say, “Bitcoin incentivized Earth to donate absolutely insane amounts of GPU, CPU, and electricity to mine with these incentive economics and contest economics,” it worked really well once. It should work again for this other thing, which requires a lot of GPU, otherwise known as AI. I think people are cluing into that narrative. They’re like, “Oh, yeah, we’ve never tried to do it again. Capitalism worked once. Let’s try it again.” Thread Guy Yeah, let’s run it back. What are you investing in? How are you figuring that out? The subnet scene has a lot there. It took me a while to understand Bittensor, and then I was like, “Geez, there’s a lot under the hood.” What are you guys investing in? How are you figuring out how much capital to allocate to what? How does the fund actually work? Mark Jeffrey Our goal is to raise enough to own 1% of all TAO. Again, just like the Winklevoss twins, that’s the game plan. We’re only 5 months old, so it’s early days. We’ve raised a decent amount so far. We haven’t made any announcements yet. We will at some point, but it’s not quite time yet. However, we have announced that we’ve invested in a couple of subnets. One of them is Ridges, subnet 62, which makes a Claude and Cursor killer. It’s beating them on the SWE-bench tests, and it costs $29 a month versus more than $200 for Claude and Cursor. It’s about 1/10 the cost. We’ve also announced an investment in Targon, subnet 4, which is confidential compute and inference. Intel just published a paper with them that was announced a couple of days ago. Thread Guy I suppose I think I saw you post it. Mark Jeffrey Yeah, I did. We’ve also announced an investment in Hippius, subnet 75. They’re decentralized storage, which would seem like it has nothing to do with AI, but AI needs a lot of storage to crunch things. You have to put the data somewhere at scale. Hippius provides storage for literally 1/400th the price of any other competitor, so it’s extraordinarily cheaper. One of the things we look for at Criterion is obviously a great product. We look for some market maturity, very much like you do when investing in startups. We also look at how the subnet token is going to grow in value. If the subnet token is just a meme coin for that subnet, that’s not good enough. With Hippius, you have to stake Hippius tokens in order to gain the right to mine the Hippius subnet. How do you mine the Hippius subnet? You supply storage to Hippius, and you get slashed if you’re not performing. You have to be responsive and do all that stuff. It deterministically ties the value of the Hippius token to a rise in usage and revenue of Hippius as a product. We like that, and the team was great. We love the people behind it. Other things that I like include Nova, which is AI pharmaceutical molecule discovery. Thread Guy Okay. They were just on This Week in Startups, right? Mark Jeffrey Yes, they were just on This Week in Startups yesterday. So, yes, we like them. We’ve also invested in Score, which is computer vision. Thread Guy Okay. Mark Jeffrey It’s not enough just to slap a camera on an AI. The AI has to understand how to see. What is that blob walking around? Oh, that’s a human. Thread Guy Yeah, what am I looking at? Mark Jeffrey Yeah, this is a car. It has to understand perspective and sight, just like a baby learns how to see. Babies can see, but they don’t know how to see yet. They don’t know how to map the input to the world. We really like Score. Score started out by using its visual mapping for Moneyballing sports. Thread Guy Moneyballing, yeah. I followed the first one; it just took me a second. Mark Jeffrey Yeah, I reversed it. They basically went to European soccer teams that couldn’t afford these big AI scouting systems. Because they can provide the same thing for 1/10th or less of the cost, all of a sudden amateur sporting teams could afford AI Moneyballing and AI visual scouting of great players. Those are just a handful of subnets out of the 128. Thread Guy Okay, so you’re really deep on a lot of these. Mark Jeffrey Oh, yeah. I run a podcast called Hashrate. I’ve interviewed most of these people at least once, from most of the subnets. Thread Guy Nice. I want to watch a couple of the ones you’re invested in. On the LLM side specifically, why would it be possible for a subnet to theoretically beat Claude or Cursor—one of these major AI labs building out either LLMs or some sort of organizational system? Why would a subnet be able to compete with these billion- or trillion-dollar companies? Mark Jeffrey Let me explain it in Bitcoin terms first. Thread Guy Cool. Rip it. Mark Jeffrey Imagine Satoshi didn’t do this decentralized thing. He just woke up one morning, went to Sequoia, raised a bunch of money, and said, “Okay, we’re going to make this alternate-universe SWIFT system and this alternate-universe digital gold. We have to buy a lot of servers all over Earth to run this network.” Imagine the insane amount of cash it would have taken. On top of that, you’d have to hire in Silicon Valley, because that’s where your offices are. If you’re an engineer, you have to move there. If you’re dead weight, you might be able to hide in the company for a long time because there are tons of engineers. Instead of doing that, in the Bittensor world the proposition is: We’re not going to hire anybody. We’re just going to set up a goal—the thing we want people mining for. In this case, with Ridges, it’s: “I want you to make a software agent capable of beating Claude and Cursor.” There are 256 mining slots, so there are 256 possible entrants into the contest. They all submit their entries, usually once a day. Anyone from all over Earth can compete. Thread Guy Mhm. Mark Jeffrey There’s no HR department, there are no salaries, and you don’t have to be geographically anywhere specific. You can be in Turkey or India. You get Earth’s best minds competing against each other, and they don’t have to take a job. They can just do this until they’re bored and leave. If they win, they win—it depends on when you look, but it’s somewhere between $25,000 and $75,000 worth of subnet tokens. You can sell them immediately, or you can hang on to them because you think Ridges might blow up and be worth as much as Claude or Cursor down the road. Because of this fierce competition, with Earth’s best minds trying to one-up each other every day, at the end of each day the winning result is published. So everybody benefits from the cleverness of the last winner. Right? And then it begins again. You cannot beat contest economics that have financial and ownership incentives that are worldwide. That’s what it turns out, right? With only a million or two dollars, Ridges has surpassed what Claude and Cursor had done with hundreds of billions within a very short couple of months. This is the most efficient form of capitalism ever invented. It’s the most ferocious form. Nobody gets paid unless they win. Only the winner gets paid, so there is no dead weight. Thread Guy Can you explain why the decentralized training run was so significant? Mark Jeffrey Yeah. When you train an AI model, the reason why you have to have 200,000 GPUs co-located is because of the intercommunication between the GPUs. They’re basically building a neural network, so there’s a lot of cross-communication that has to go on in order to build an accurate model while you’re training it. The concern with a decentralized model is, “Oh, well, these neurons, if you will, are all over the Earth.” It’s too slow. They have to be co-located in order for it to make sense. I couldn’t speak intelligently to the technique other than to say that the problem they solved was the intercommunication. That was significant. Thread Guy Yeah, so they proved it works in a Petri dish. It’s like, “Well, it worked in the Petri dish. It’s just about scaling up now.” I think they can do it. That was actually a good explanation. What is the bear case for TAO? In what scenario does this not work and your investment in the fund is just, “We messed up. We were close”? Mark Jeffrey There could be a new regulatory regime that arises which seeks to kneecap AI or centralize it. You’re already seeing Bernie Sanders out there saying we’ve got to stop building data centers. Thread Guy By the way, he did the Claude interview, and it was ridiculous. Mark Jeffrey Yeah, yeah, yeah. It reminds me of those cartoons from the 1900s, when electricity first came around and a bunch of people were like, “Electricity? We’re all going to get electrocuted and die.” Thread Guy Yeah, exactly. Every 100 years, we go through one of these. Mark Jeffrey Nonetheless, there are people who could come to power who basically say AI is illegal unless it’s licensed by a national regulator. Maybe only OpenAI, Google, and Grok. Maybe not Grok, because we don’t like Elon. Maybe only licensed AI will be allowed. That’s a big threat, probably, to everybody. Thread Guy Yeah. Mark Jeffrey Could they shut down Bittensor? Probably not. Could they shut us down? Yes. If this was in America, that would affect us, because all of a sudden maybe funding these things would be illegal. I don’t think that’s very likely at this point, but that’s always a danger. Then there are the normal things, like a chain hack. Thread Guy Yeah, yeah, yeah. A disaster on-chain or something like that. Mark Jeffrey I think the regulatory one is really the biggest one that worries me. The other thing is, if we do something stupid in terms of developing new capabilities that adversely affects the way Bittensor operates. For example, there’s now talk about putting lending and shorting directly on-chain, the same way Uniswap is baked into the heart of the chain. It’s not on top of it; it’s inside of it. There are now talks underway, as of last week—you just had Cons on the show—about having lending that would enable shorting baked into the chain, allowing people to bet against subnets they think suck. Right now, there’s no way to do that. If somebody has enough TAO to continually keep buying their own subnet, there’s nothing you can do about it. But if there’s lending and shorting, little guys can offset that with bad sentiment, which might be the correct answer. It might be a crappy product, and the chain should be able to vote with its feet, if you will. This would enable little players to compete against the bigger ones. However, do you remember what happened with ThorChain? Thor Lend? I don’t know if you were around for that. Thread Guy Yeah, kind of. I used to hold RUNE. Ticker RUNE. I love RUNE. I used to hold RUNE back then. Mark Jeffrey Yeah, I’m still a big RUNE fan. But for a period of time, they enabled lending on the chain, which became insolvent and threatened to bring the entire chain down. Thread Guy Wow. Mark Jeffrey Through baking lending into the chain, right? So things like that worry me a little bit. So far, Cons has not steered us wrong at all, and I’ve been worried about things that were not worth worrying about. Cons’s solution to these problems tends to be very good. He has never let me down once. I’ve gone through six worry fests where I’m like, “Oh my God, Cons is making a mistake. He’s totally wrong.” Then, 2 weeks later, I’m like, “No, he was totally right. Look what happened.” Thread Guy Is he in charge? Is he the primary— Mark Jeffrey Well, he used to be. He used to be the CEO of the foundation. He has stepped away from that now. Thread Guy Got it. Mark Jeffrey I could be wrong. I can’t speak intelligently to this because Cons described it, but he has influence and could push it forward. Thread Guy Yeah, yeah, yeah. The foundation is really the one driving all this now. Got it. Who runs the foundation? Mark Jeffrey It’s a bunch of early TAO people. I actually don’t recall exactly who’s in charge of that. Thread Guy Got it. Mark Jeffrey It’s a lot of brainiacs. They sit up in the tower where these decisions are made, and I’m down in the subnets. I talk to them every once in a while, but not really that much. Thread Guy Near the end here, can you share a price target for TAO? Mark Jeffrey I’ll tell you what I think. Not financial advice, but I will tell you what I think because I’ve said it before. I was around in the early Bitcoin days, and I watched Bitcoin go from around $200 to $1,200. Then it got scared, went back to $600, and all the way back down to $250. It sort of did this bobble thing. At some point, it jumped up to $3,000 and then $10,000. All of that happened within the first year after the first Bitcoin halving, around 2013 or 2014. Bittensor just had its first halving. Thread Guy Ah, because you’re looking at it on that halving scale. Mark Jeffrey I’m looking at it from a couple of dimensions, but halving is absolutely one of them. It’s 2013, if you will, in the Bittensor world. What is its price? It’s around $300, and it’s sort of bobbling around $300 at the moment. It’s exactly like Bitcoin. On top of that, you have the subnets. To make a subnet requires TAO. How much TAO? Between $100,000 and $1 million worth of TAO, depending on when you look and how much demand there is. There are also going to be more subnets at some point, probably in the next year. The number is going to go from 128 to 256. That alone is going to create a giant amount of demand for TAO. James Altucher did an analysis of that, and he said, “Oh my God, that alone could drive it up to several thousand dollars,” never mind everything else. Then you have staking TAO into subnets to get subnet tokens. 20% of all TAO that’s in circulation is currently staked within subnets. Another 50% is staked on root, which is kind of the lazy way to stake and get some rewards. Roughly 70% of it is staked, so it’s not moving around. You’ve got things which cost TAO and which cause you to stake TAO. Bitcoin doesn’t have that. In addition to the Bitcoin-like dynamics, you have Ethereum-like dynamics. I think all these things are going to start compounding at some point. I think it’s pretty conservative to think that, if it does only what Bitcoin does, we’ll see something like a $3,000 TAO by the end of this year. Thread Guy Whoa. Mark Jeffrey That’s a conservative bet. It’s only my opinion, but still, of course, we are positioning ourselves accordingly based on thinking that. If you listen to Barry Silbert speak, you become a Bitcoin billionaire. He recently said at a Bitcoin conference that he thinks Bittensor TAO could 500x. It has that kind of potential. Not this year, but just for fun. Thread Guy I respect someone who comes on and gives a price target. That was a sick price. Dude, I appreciate you coming on. We should get Jay Kell on at some point. Mark Jeffrey Yeah, he might do it. Every now and then, he’s getting more excited about TAO, as you’ve noticed. Thread Guy Yeah, maybe we’ll set it up and rip it. I appreciate you coming on. As a sign-off, where do you send people if they want to learn more about it, follow what you’re doing, or watch the interviews you’ve done with founders on the subnets? Where can people learn more about everything you’re doing? Mark Jeffrey I’m at Mark Jeffrey—J-E-F-F-R-E-Y. Hashrate is the name of the podcast where I interview all the subnets. Most of my views are on X, but I also have a YouTube channel that replicates them. Just search for Mark Jeffrey, and you’ll find me. The company is called Still Core Capital, so stillcorecapital.com is for accredited investors only. But if you’re curious about what we’re doing, go there. Thread Guy Sick. Mark Jeffrey, dude, it’s an absolute pleasure. I appreciate you coming on. It was a blast. It was fun to do Cons, then rip you. It was awesome, man. Thanks for the time, bro. I really appreciate you. Mark Jeffrey My pleasure. Thanks, dude. Thread Guy All right, have a good one, bud.