This episode reveals how a miner-first design on Bittensor is enabling weather predictions 30% more accurate than state-of-the-art models, targeting a multi-billion dollar B2B market.
The Origin of Subnet 18: A Blend of Friendship and Expertise
- The creation of Subnet 18 (Zeus) began with a casual conversation that quickly escalated due to a unique convergence of talent. Wouter Haringhuizen, who has a Master's in AI, and his co-founder Travis, an econometrics specialist, decided to explore creating a subnet. The idea for a climate-focused network came from Travis.
- In a moment of serendipity, their colleague Eric, who was sitting nearby, revealed he had conducted research on AI climate forecasting at Cambridge. This immediately catalyzed the project.
- Team Dynamics: Wouter emphasizes the importance of personal connection and good team culture, highlighting his friend Travis, who jokes he was "hired for the vibes" but is a critical, hard-working member of the team.
- Rapid Development: With Eric’s specialized knowledge, the team developed the initial code within days and leveraged existing connections to launch the subnet.
Partnership with BitMind: From Miners to Mentees
- The team’s relationship with BitMind (Ken and Dylan) was foundational to their launch. Wouter explains that his team initially operated as a mining group on other subnets, including BitMind's. By actively reaching out to subnet owners, they built a strong rapport.
- This pre-existing relationship evolved into a formal partnership, with BitMind providing mentorship and technical guidance for Subnet 18.
- Deep Involvement: Ken and Dylan are not passive partners; they actively review code, participate in weekly meetings, and hold a stake in the subnet. Wouter notes, "Every time we do like a pull request or an update, they look into it... I would say they're like pretty well invested into the subnet."
- Ecosystem Dynamics: Wouter points out that post-DTA (Dynamic TAO Allocation)—a Bittensor mechanism that distributes ownership to miners—subnet owners are now forced to build strong relationships with their miners, as miners hold a significant stake and influence over the network's future.
Strategic Rollout: Temperature, Precipitation, and Wind
- Subnet 18’s strategy is to incrementally add and master core weather variables that are easily understood before expanding into more complex B2B-focused data. They began with 2-meter temperature, the standard measurement seen in everyday forecasts.
- After achieving superior performance in temperature prediction, they have now added precipitation. Wind prediction is next on the roadmap.
- B2B Focus: While these initial variables are familiar to consumers, the ultimate goal is to serve the B2B market. Wouter explains that marginal improvements in accuracy, such as a 0.1-degree better prediction, are highly valuable for businesses like hedge funds or energy companies, but less so for the average consumer.
- Roadmap: The plan is to first cover the three main consumer weather metrics (temperature, rain, wind) and then expand to variables tailored for specific industries.
Beating State-of-the-Art: A 30% Performance Leap
- The subnet's performance is measured against a baseline model that calls a professional weather API, which provides predictions from the historically best-performing model for a given location. Wouter reveals that their miners have collectively achieved a 30% lower error rate on temperature predictions compared to this state-of-the-art baseline.
- Technical Validation: This achievement validates the core thesis of Bittensor: a decentralized network of competing models can outperform monolithic, centralized systems.
- Wouter’s Perspective: "It's absolutely insane that in 2-3 months miners are able to beat state-of-the-art models like that. We definitely couldn't have done that ourselves. So, we're happy that we created an environment where we allow miners to do that."
Wouter’s Journey: From Skeptic to Subnet Owner
- Wouter shares his personal history with Bittensor, which began through a friend's introduction. Initially, he was skeptical and left to finish his studies. However, after the price surge to $700, he returned to mining but struggled to compete.
- This experience of being a miner and facing the high level of competition gave him a deep appreciation for the network's potential and informed his approach to building Subnet 18.
- The Value of Experience: He and Travis, despite their academic backgrounds in AI and econometrics, found it incredibly difficult to beat other miners. This firsthand struggle convinced them of the network's power and provided invaluable lessons for designing their own subnet.
Competitive Landscape: A Different Approach from Gaia (Subnet 57)
- When asked about Gaia (Subnet 57), another weather-focused subnet, Wouter clarifies the key technical difference. Gaia focuses on fine-tuning a single, large model, whereas Subnet 18 allows miners to run their own models locally and only submit the final predictions (inferences).
- Decentralized vs. Centralized Models: Wouter argues that weather is a distributed problem. A model optimized for the Netherlands will outperform a global model in that specific region. Subnet 18’s architecture encourages miners to specialize in different geographic areas, creating a composite forecast that is superior to any single global model.
- Strategic Focus: He notes that his team remains focused on building, viewing the competitive attention from Gaia as a compliment.
The Multi-Billion Dollar Market for Weather Data
- Wouter identifies a massive B2B market for hyper-accurate weather data, citing a multi-billion dollar market size based on industry research papers. The potential customers are diverse and span numerous high-value sectors.
- Key Use Cases:
- Hedge Funds: Trading on predictive models that use weather data to forecast commodity prices (e.g., oil). Wouter confirms they are already in discussions with a hedge fund.
- Energy Grids: Predicting solar and wind power generation to manage grid stability and trade energy futures. As renewable energy sources grow, this becomes increasingly critical.
- High-Performance Sports: Teams like Formula 1 rely on precise weather forecasts to gain a competitive edge.
Alpha Token Utility: Aligning Incentives with Revenue
- The primary utility for Subnet 18’s alpha token is directly tied to the business's future revenue. Wouter explains that while the subnet is currently bootstrapped with TAO emissions, the long-term vision is to create a self-sustaining economic loop.
- Revenue Sharing: A significant portion of the revenue generated from selling API access will be funneled back to the miners as incentives.
- Investor Thesis: This model directly correlates the value of the alpha token with the subnet's commercial success. Holding the token represents a stake in the future cash flows of the weather API business.
Future Roadmap and Core Challenges
- The long-term goal for Subnet 18 is to become a comprehensive weather API that offers state-of-the-art predictions across all standard variables, effectively competing with established players but with a superior product.
- Market-Driven Expansion: Future variables, such as 100-meter wind speed, will be added based on demand from specific B2B markets like the German energy sector.
- Biggest Challenge: The primary operational challenge is the 5-day delay in receiving ground-truth weather data for verification. This makes it difficult to quickly evaluate new miners' performance without granting long immunity periods, which can disrupt the network's incentive balance.
Advice for Future Builders: Start by Mining
- Wouter’s top advice for anyone looking to build on Bittensor is to first become a miner. He argues that this hands-on experience is the best way to understand the ecosystem's technical and economic intricacies.
- Learn the Struggles: By mining, you learn the flaws, frustrations, and complexities of the network from the ground up. This includes setting up validators and understanding performance verification.
- Build with Insight: "You learn everything in Bittensor. And then as soon as you get all those components right... with a good idea, you can actually make it work within like one or two weeks because you know how everything works."
Conclusion
This episode highlights how Subnet 18's focused, miner-centric approach is creating tangible value by outperforming established models. For investors and researchers, this case study demonstrates a clear path from decentralized incentive design to real-world, high-value B2B applications, with tokenomics directly tied to commercial success.