At this year’s Proof of Talk, bitstarter.ai is hosting a competition like no other - with aspiring teams pitching a panel including Mog Machine, Mike Grantis, and Const for the chance to win a subnet. But how do Bittensor’s best actually analyze subnets? To find out, we asked Mog to share the process behind his portfolio, and reveal how he built one of the strongest subnet line-ups in Bittensor. 

By Chris Zacharia (@macrozack)

Building a subnet portfolio isn’t easy. 

For two years, I’ve tried to learn as much as possible from the key leaders building on Bittensor, asking: how do they decide which teams to back, and which to pass up? And once they assemble their subnets, how do they integrate them into a meaningful whole?

Even on close inspection, the portfolios of many investors and subnet developers seem haphazard, even scattergun. But a handful of them have emerged with intelligible, well-considered subnet directories that make sense - and make you take notice. 

Mog Machine is one of them. 

His portfolio is one of the most coherent, well-integrated on the network. SN75, Hippius (storage) and SN85, Vidaio (video compression) share clear synergies; TaoStats and SN19, Blockmachine fit together glove-in-hand. And even as a lead investor, Mog’s backing of SN59, Babelbit (speech-to-text translation) and SN44, Score (computer vision) follows the same pattern: it doesn’t take a genius to see how his subnets could form a single decentralized tech stack.

As bitstarter.ai gears up for the Proof of Pitch competition at Proof of Talk, where Mog joins the judging panel, I wanted to understand how he actually evaluates a subnet before he commits. What follows is a tour through his thinking - from commodities to alphanomics, block explorers to enterprise sales, and the one subnet he wishes he'd backed.

There isn't a single thing - there are two. Firstly, it's the team and their ability to understand the market and the product for which they're creating a commodity. And secondly, it's that market itself: the achievability, the pathway to it, the roadmap, and their understanding of why they're using Bittensor to solve a problem - and how that problem would be solved without Bittensor.

Once you understand that distinction, you understand why you're building in the first place. A subnet optimises part of the workflow of a business - usually the procurement of the commodity you monetise or incorporate into your product. To do that, you need a business and a product, or at least a clear path to one. And the commodity has to have value. If it doesn't, you won't have a long lifeline - especially in a dTAO world where emissions, buybacks, and balancing your alphanomics are critical to staying entitled to use Bittensor at all. You're not even entitled to emissions anymore.

I know subnets sitting on the deregistration threshold that have a product - but the value of that product doesn't align with the incentive structure, or they never understood how to optimise and balance those parameters. So I need teams that not only have a pathway to a product, but understand how to balance the economics.

The alignment stems from my knowledge scope. Having not come into the network with a machine learning PhD, I focus on building viable, profitable businesses that are easy to understand. So I invest in subnets whose commodity I understand and where I can add value - through strategic advice, market understanding, and knowledge of the resources we have as a company. That's the main reason for the subnets I've been involved in. It's certainly not by accident.

Honestly, because research isn't my specialist field, so I tend to stay away from it - it's harder for me to assess value and see a clear path to success. If I need to ask Jake, or Fish, or Rob to explain to me what something means, or even whether it's something - and often when I do, I get it, and I think, okay, if we manage to prove this, then it's important - but it's all ifs and buts. There are too many ifs and buts. I'm much more black and white.

So I've always said: a subnet optimises part of the workflows of your business, usually procurement of the commodity you monetise. To do that you need a business and a product, or a path to one. The commodity needs value. If it doesn't, you won't have a long lifeline - especially in a world where emissions flow, buybacks, and balancing alphanomics are critical to continuing to be entitled to use Bittensor. That's what we're constantly trying to do, and each thing we do takes it one step further.

When you build a business, you're solving a problem - and the problem we were solving was our own. As TaoStats, building in Web3, we were spending a lot on our RPC infrastructure and running nodes ourselves. We had in-house expertise, but we knew we weren't doing it as well as it could be done. We didn't have the funds to bring in a team of 20 or 200 to really go out and optimise that RPC - but at the same time it was costing us a huge amount that we knew could be optimised.

And because we were also customers of Finality and other competitors providing RPC infrastructure, we understood the customer story and the customer need. We realised we could run the product ourselves at the same cost we were already paying - we knew exactly what it cost us - and just make it better. That priced us into the market very quickly.

We also understood, through something like Hippius - a commodity-based subnet where you're buying storage - that we only pay miners for what they actually supply and what gets used. That's harder with storage; it makes us more of a need to scale. But with Blockmachine, even with TaoStats as the main customer, we already knew the chunk we could feed directly into the subnet - not as revenue, but as customer usage. Then it's just a question of building on it from there.

Tao.com and tao.app should be using Blockmachine - and they will be soon. It'll soon be the default RPC for Bittensor, replacing the OpenTensor free entry point, and that exposure should reach all miners. So there's no reason not to: if you're running RPC infrastructure cheaper than Blockmachine, then fine, mine on Blockmachine instead. Otherwise you should be using it - unless you're just stubborn and want complete control over your own in-house infrastructure. 

By all means, spend the money. But in the same way so many people build on the TaoStats API because they don't want to run and build their own indexers on top of the chain - Blockmachine is the same. And of course this then extends to Ethereum, Binance Smart Chain, and whatever chains we choose to support after that.

YOLO? No - actually I have a good routine. I wake up and there's a core group I'm very close with: Dubz from Hippius, and Doug from TaoStats, who's also working on Blockmachine. We've got our own little private Discord, just the three of us. The first place I go in the morning is Dubz, because he's two hours ahead of me, and we have a catch-up. My agentic setup is so efficient now that I know everything that's been said overnight - in the Discords, in the ClickUps, what everyone's been working on - and I immediately go and answer anything that needs my attention.

These days I stop work around five or six in the evening, usually when the kids come home from school, and I try not to go back unless I need to - so I miss the American-time stuff. I'll catch up quickly and give my input on anything that's required. But in those early hours, I work on the things I'm working on for those businesses. Right now that's a massive knowledge corpus for the entire Bittensor network - I won't explain how we're using it, because it's quite competitive, but it'll be very useful to every participant in the network. There's something else I'm working on too, that I'll keep quiet for the same reason.

With Hippius, I'm very hands-off - my main role is using it and giving feedback. I'm a heavy user through S3, the agent, the desktop, and I'm constantly feeding back from a consumer perspective. They'll sometimes come to me and say, this design or this one - and I'm good at pulling out fundamental narrative issues in how things are presented. Blockmachine, same thing: the team's very strong, they don't need me, but they want me from time to time. TaoStats is the one I spend the most time in - it's still my baby, my opinion always matters, especially as we're going through a big rebuild right now. That's driven by me and by the feedback I garner from users and from using it myself. I'm the heaviest user of TaoStats out there - and probably the most dissatisfied. Then there are obligations to the network: calls with subnet founders, VCs, external investors, podcasters, people who want advice or a catch-up.

You need both - and they need to understand their own weaknesses. It's very difficult to have a perfect team. Look at Chutes: for a long time Jon’s progress was incredible, but he wasn’t always able to look up from his genius cave to understand the bigger picture. Yet because he had a good team behind him, they're far better on communications, business development, and orientation. Often when I raise something with Jon he'll say, "I don't handle that side" - and that actually makes me happy, because it means he's got people doing more.

The strong teams are the well-rounded ones. I'm critical - I'd say this to his face, though I'd rather he didn't print it - but someone like Fish is super idealistic and super talented, and yet I still think he lacks maturity in business, and I think it reflects in how things get presented and in the momentum a project gains or loses.

There's nothing worse than a hype founder with no fundamentals - Fred Krueger-type vibes - and the thing is, Bittensor sniffs through all that bullshit. But at the same time, look at someone like Barbarian on SN123: incredibly talented developer, but no front-end, no website, nothing. I was asked to do some due diligence there and I said, look - very promising technically, but absolute red flags for investment, because there's no clear communication.

So it's easier to take technical talent and add good leadership to it than to take good leadership and find the technical talent - especially within Bittensor. Technical talent is the primary thing I look for, because the rest can be added. I once went into a meeting knowing I was going to tell a team I wouldn't invest - and by the end of the meeting I was investing, and by the second meeting I was hooking them up with our design team. Look, guys, your front-end is shit, your communication is shit, I love you because you've got experience - but you don't know where it's at in terms of communicating to the audience you need to connect with right now.

The multi-audience deck is an interesting challenge, but we've all got to do it. I keep asking my design team for a single deck we can put in front of anyone and keep updated - one that instantly explains what you're doing in a way everybody understands. Start with the instant introduction in simple language: everyone in the room, including the people who ask for everything to be explained simply, has to get it. Then a slightly more detailed explanation of how you aim to achieve it, maybe without giving away your IP. At that point, go technical on a couple of points - the ones you think are most valuable, most groundbreaking, most important, usually to do with the incentive mechanism. Then draw it back out again: talk about how you'd structure the business, whether there's a product, the addressable market, the monetisation. Then conclude.

That's it. You've explained what you do, you've gone a couple of levels technical, a couple of levels business, and concluded. Everyone takes what they value from it. You've addressed the whole room - the product and business people can drive questions deeper into the business side, the technicals into the incentive and code side, and everyone else can ask generalised questions off the intro and conclusion.

One more piece of advice: English isn't everyone's first language, and sometimes the people best placed to explain the technical detail aren't the best placed to pitch. I saw that judging the subnet ideathon at Hackquest - a lot of really strong technical people who, in a five-minute pitch, just got lost. So get your front-man up there doing the presentation, with the others supporting and ready to answer questions. We do this with Hippius all the time - Dubz isn't the natural front-man, he's deeply technical (and quite funny), so we work well together: I pick things up, bring him in to explain something specific, then know when to pull him back out. Understand your team dynamic. But don't put the hype-man up there with nothing behind him either - he'll get annihilated the moment the questions start.

I said I wasn't doing any more subnets - and I happen to be in deep research and discussion with some people about doing one more. 

I won't say what it is. But there's one thing Bittensor is missing at the moment that's core. The answer, really, is one with clear customers, real usage, and exponential growth - something you just plug in, the same way we talk about Hippius: plug it in and replace S3. There are centralised products we all use every day that we can create optimised versions of on Bittensor, and the people using those products will say, of course I'll drop in a replacement - it's as good or better, and it saves me money. No-brainer. Those are the subnets that are really interesting.

I can't take any credit for Score or Babelbit - I think they're so good I wish I'd had the talent in the team to do that, but I'm fortunate enough to have added value to them and tagged along for the ride. 

But the one I feel genuinely fondly about, for some stupid reason, is 404-Gen - SN17. 

It's a clear product. Such hard work has gone in. They're really using AI and GPUs to do something of value - a big dataset, plugging into the mainstream, having gone through the work of integrating and understanding the channels to adoption. Always well presented, always clear and understandable, fantastic communication. I'm a secret fan. It's a shame I don't own their alpha - I probably should do something about that. 

Because I'm 100% alpha, I'd have to sell some Hippius and Score just to pick up some 404. Maybe I'll do exactly that after this call, so people know I really love them

Watch Mog on the judging panel of this year’s Proof of Pitch competition, June 3rd at 5pm UTC live in the Louvre - or watch online via Bitstarter’s X account.

Get your tickets to Proof of Talk here, with a 30% discount for TaoTimes readers using code TAO.

Bringing TAO to the world.