Subnet Session with Ken from Zils: Subnet 127

In this episode of Revenue Search, Siam Kidd and Mark Creaser welcome Ken from Zils AI to explore how smaller, specialised AI models could transform the way businesses make decisions. Unlike traditional large language models that generate lengthy responses, Zils focuses on delivering fast, accurate decisions with confidence scores, allowing businesses to automate tasks more efficiently and at a fraction of the cost. Ken explains how Zils uses Bittensor miners to continuously improve its decision models, which businesses can customise using their own data. From customer support and healthcare claims to warehouse automation and AI agents, the discussion explores how these models can make everyday decisions faster, cheaper and more reliable.The conversation then turns to Zils' business model, early revenue and growth strategy. Ken reveals that the platform is already generating revenue, with customers paying for model training and API usage. Zils currently charges $0.20 per million training tokens, subject to a $2 minimum, and $0.06 per million inference tokens, with several customer pilots already underway, including healthcare applications. Siam and Mark explore the challenges of attracting enterprise customers, the importance of effective marketing and how Zils could establish recurring revenue as businesses integrate its models into their operations. They also discuss the potential for smaller AI models to become a major part of the wider AI economy, bringing affordable intelligence to industries where larger models are simply too expensive or impractical.

Revenue Search welcomes Ryan, founder of Carbon, to explore how Bittensor could revolutionise the world of engineering and product design. Drawing on his experience in the US Air Force, Virgin Galactic and Boeing, Ryan explains how Carbon uses AI-powered physics models, known as neural operators, to replace slow, expensive engineering simulations with near-instant predictions. By allowing engineers to rapidly test different materials, shapes and designs, Carbon aims to dramatically reduce development costs and accelerate innovation across industries including aerospace, automotive, robotics, cooling systems and manufacturing. Ryan also explains how Bittensor miners compete to develop specialised models, while validators rigorously test their accuracy and reliability before they can be used in real-world engineering applications.The conversation then turns to Carbon's commercial strategy, revenue potential and future plans. Ryan outlines an initial focus on smaller engineering firms, with the goal of developing bespoke physics models for paying customers before expanding into reusable tools for entire industries. The team also plans to use future revenue for alpha buybacks, directly supporting the subnet's growth. Siam and Mark explore whether Carbon could go beyond supplying prediction models and eventually offer complete AI-powered engineering solutions, designing better products rather than simply providing the tools. From more efficient aircraft and drones to entirely new approaches to manufacturing, the discussion highlights Carbon's potential to turn cutting-edge AI research into practical, commercially valuable technology.

Revenue Search welcomes back Rob from AdTao (Subnet 21) to discuss the project's progress, the challenges of operating within Bittensor's rapidly changing ecosystem, and its plans to transform the Google Ads industry. Drawing on his experience building and selling successful advertising and software businesses, Rob explains how AdTao aims to replace traditional PPC agency guesswork with data-driven predictions. With access to approximately 10,000 Google Ads accounts representing over $1 million in daily ad spend, AdTao uses Bittensor miners to develop prediction models that forecast the impact of advertising decisions over 7, 14 and 28 days, helping businesses improve campaign performance and make smarter decisions.The conversation also covers AdTao's expansion beyond its own software into a wider AI-powered advertising infrastructure, including MCP integrations that allow users to access its intelligence through tools such as Claude, ChatGPT and Grok. Rob outlines plans to significantly expand the dataset, introduce free and usage-based pricing, and offer a fully managed alternative to traditional advertising agencies. Crucially, AdTao announces that 15% of all new revenue will go towards automated alpha buybacks, creating a direct link between commercial growth and subnet value. Siam and Mark also explore AdTao's competitive advantage, its revenue potential, and why combining an established business with BitTensor's decentralised intelligence could create a significant opportunity.

Revenue Search sits down with the Agent Comms team Tom, Chris and Ashton to explore a potential new Bittensor subnet designed as a trust layer for human and AI communication. The project aims to combine Kaspa for ordering, sequencing and transaction records with Bittensor for incentives, using miners to index, store, serve and prove communication data. At its simplest, Agent Comms is building infrastructure that lets people, businesses and AI agents communicate, share files, publish data and verify interactions without relying on a single central provider. The team walks through four planned products: a Slack/Discord-style communication app, a decentralized Kaspa RPC provider, AI-to-AI coordination and payment rails, and ZK proving services. They discuss how encrypted messages and files are stored off-chain while Kaspa records proofs and timestamps, with Bittensor miners supporting availability and delivery. The commercial model includes server subscriptions, protocol licensing, RPC subscriptions, ZK proving fees and planned alpha buybacks from revenue, though Siam and Mark challenge the team on messaging, go-to-market, miner scoring, and whether the product can be explained clearly enough for the market.

This episode is a Bittensor catch-up rather than a subnet interview, with Siam back from three weeks in California and Mark writing from Spain. The main focus is the pace of recent BitTensor “upgrades,” especially Root Reborn, validator baskets, deregistration mechanics, conviction, chain buys, and the looming arrival of shorting. Both remain strongly bullish on Bittensor, but argue that the ecosystem is becoming difficult for investors, subnet owners, and potential new teams to navigate because rules are changing quickly and often without enough notice, roadmap clarity, or explanation. Mark frames his recent article as a call for better process rather than a complaint: the issue is not that changes are bad, but that constant changes create fatigue, uncertainty, and make it harder for serious teams and institutional capital to commit.A major theme is whether price alone is the right mechanism for deciding which subnets survive. They argue current deregistration can become a popularity contest, while real businesses, especially enterprise or research-focused subnets, often need 6–24 months to generate meaningful revenue. They compare this to startups like Tesla, Apple, Anthropic, or enterprise software companies, where patient capital is required before results arrive. Siam also highlights that markets inside Bittensor are still inefficient: tweets, Telegram comments, and short-term narratives can move subnet prices far more than fundamentals. They discuss how Root Reborn and chain buys may create new investment metas, but sudden future changes could instantly break those strategies, which is exactly why clearer process matters.They also debate conviction and whether it actually protects investors from rug pulls. Both are sceptical, with Siam calling it “virtue signalling” and Mark arguing it may restrict good teams from using alpha to fund growth while not necessarily stopping bad actors. They also warn that conviction plus shorting could increase the power of deep-pocketed players to influence, pressure, or even attack subnets. The closing message is that Bittensor still has huge talent and potential, but the network’s incentive mechanism needs refinement so value, talent, capital, and long-term builders are better aligned. They finish by reinforcing that they are not bearish or moaning; they are pushing for improvements because they believe Bittensor can become much bigger if the ecosystem becomes easier to understand, invest in, and build within.

In this episode, Mark and Siam discuss why they now view Const’s rapid protocol changes more positively: although constant changes can make institutional investors nervous, they argue Const is in “founder beast mode,” trying to fix Bittensor’s core mechanics before the network scales. They focus heavily on “Root Reborn,” especially the distinction between emissions and injections: miner/validator/subnet-owner emissions are not what’s changing; the new formula affects liquidity-pool injections, rewarding subnets with stronger root proportion, stronger EMA price, and lower miner burn. They frame this as a way to stop zombie subnets, reduce gaming, and reward teams that are actually producing value. They also cover the new “naughty list” / injection-eligibility list, arguing it is less a punishment list and more a forcing function for subnets to prove their incentive mechanisms work. New subnet registrations now start with injections off by default, meaning teams need to be ready from day one rather than buying a slot and sitting on it. Mark and Siam debate whether this kills the secondary subnet-slot market, why new subnets may now have a better setup than buying old slots, and how the “Goldilocks subnet” might be a new project with high root prop, high starting price, low burn, real product-market fit, and revenue/buyback capacity from launch. They briefly touch on conviction, impending shorting, and then plug DSV’s upcoming “Inside Bittensor” event in London, where they’ll reveal DSV’s first subnet, Subnet 69.

This session starts with Mark and Siam briefly addressing the decision to pivot away from “Handshake”: it didn’t gain the traction they hoped, the team’s focus drifted toward building trading/mining “skills” that didn’t feel like a scalable revenue path, and they chose to “rip the band-aid off” rather than keep a zombie project alive. They argue that in AI (especially on Bittensor) things change fast, so it’s better to redeploy talented builders quickly, even if that decision attracts criticism. The main segment features Subnet 103 “Djinn,” led by Harry Crane (stats professor with deep prediction markets/sports betting background) and Iosef Gerstein (finance + scientific startups). Djinn’s core idea is a “genius–idiot network”: profitable bettors (“geniuses”) have an edge but get limited/banned by sportsbooks, while the mass of recreational bettors (“idiots,” i.e., accounts that can still place bets) can execute. Djinn is building a trustless, privacy-preserving information exchange to connect the two—geniuses sell time-sensitive betting signals, buyers execute them wherever they have access (Bet365, Polymarket, etc.), and the protocol uses Bittensor miners/validators for verification and obfuscation (including decoys) so neither Djinn nor outsiders can easily steal signals. Revenue is expected to come from transaction fees (they discuss ~0.5% as a working idea), with buybacks/utility mechanisms considered to support alpha; they’re currently moving from fake-money testing to low-stakes real-money testing and say they’ll launch only when they’re confident it’s safe.

This episode, Mark discloses that DSV is already invested in today’s subnet, but they’ll still ask the awkward questions. They bring on Koyuki (“special k”) from San Francisco, who shares her background in AI (web2 + web3), how she joined the Bittensor Foundation/OTF as Head of AI, and then dives into her slides on Subnet 78, Vocence.Koyuki pitches Vosens as a decentralized “voice intelligence layer” on Bittensor, targeting the rapidly growing voice AI market and competing with incumbents like ElevenLabs by being more open, cheaper, and driven by Bittensor incentives. She shows that Vocence already has a live studio product (TTS/STT, voice cloning/design, text-to-music, API) and outlines how miners submit models that validators score across nine dimensions (script accuracy and naturalness weighted highest), with winning models becoming the new baseline for inference. On revenue, she describes a credit-based SaaS model (consumer + API, with enterprise as the big upside), plans for buybacks into a treasury, and an emissions burn condition if no model clears a defined improvement threshold. The discussion then focuses on the “Turing test” problem for voice agents—latency, filler words, interruptions, and overlapping speech—and Koyuki claims a new “style trajectory TTS” approach will make agents sound truly human soon. Siam offers a $5,000 wager that Vocence can produce a voice agent he can’t detect as AI by the end of the month, and Koyuki accepts, with some talk about testing via a phone-call scenario and adversarial off-script questions. They wrap by noting the prior Vocence slot issues/deregistration risk and arguing this time is different due to stronger leadership, a live product, faster shipping, and early traction.

In this episode, Bob from Subnet 48 (quantum compute) gives a grounded overview of quantum computing: huge long-term promise (materials, batteries, drug simulation), but today’s machines are still “NISQ” (noisy, intermediate-scale, not error-corrected at useful scale). Subnet 48’s pitch is essentially “Airbnb for quantum computers”—miners run real quantum workloads, users submit quantum circuits, and the network executes them cheaper than traditional access. Bob shows OpenQuantum.com as the front-end marketplace, listing multiple hardware providers (IonQ, Rigetti, IQM, AQT) with current machines in the ~20–50 qubit range, and explains that most jobs on OpenQuantum are being executed via Subnet 48.The conversation then veers into the big scary question: quantum risk to crypto. Bob distinguishes SHA-256 (mining) from elliptic curve cryptography (ownership/signing) and argues the nearer-term threat isn’t quantum “mining Bitcoin faster,” but breaking signature security unless chains migrate to post-quantum schemes. He mentions industry roadmaps and research suggesting timelines could be tighter than people assume, and plugs Subnet 63 (Enigma)—a prize-driven subnet designed to incentivize public breakthroughs in cryptography rather than vague claims.

In this Revenue Search episode, the hosts sit down with Aldo from Subnet 55 (NIOME / “Neural Intelligence in Omics”)—a project tackling one of the messiest problems in biotech: how to make genomic/biodata usable for research and AI without turning it into a privacy and cybersecurity nightmare. Aldo walks through why the status quo is broken, pointing to repeated breaches and misuse across the industry (from direct-to-consumer testing firms to major institutions), and makes the case that “compliance” doesn’t equal “security” when hackers are actively targeting sensitive health data.NIOME’s approach is twofold. First, through the wider genomes.io ecosystem, individuals can store their DNA data in encrypted “vaults” where the user remains the owner and controls access—rather than handing away rights to hospitals or platforms. Second, the subnet’s core mission is to generate synthetic genomic / biodata at scale—so pharma, biotech, and researchers can train models and run analyses without exposing raw identifiable datasets. The roadmap is built around a structured series of predictive challenges (starting with cystic fibrosis / CFTR), with commercial interest already forming around bespoke challenges, licensing outputs, and data brokerage partnerships (e.g., bringing external datasets into the synthetic pipeline and sharing revenue when that data is used). The big idea: make biodata safe, precise, and scalable and use Bittensor’s open, inspectable “under-the-hood” model development to build trust versus black-box approaches.