This Week in Startups
This Week in Startups

This Bittensor Subnet Could Cut Drug Discovery Costs in HALF | E2267

This Week In Startups is made possible by: Luma AI - https://lumalabs.ai/twist Every.io - https://every.io Lemon.io - https://Lemon.io/twistPlaud - https://Plaud.ai/twist Today's show: What do drug discovery, the creator economy, and AI vision models have in common? In the case of Metanova, Bit

Featured Speakers

Jason Calacanis Host

Topics Discussed

Episode Summary

Executive Summary: This episode spotlights three BitTensor subnets: Metanova (drug discovery), BitCast (creator marketing/content generation), and Score/Manico (vision AI). Across all three, the hosts and founders argue that decentralized incentives can crowdsource useful AI work, reduce costs, and accelerate commercialization by turning open competitions into production-ready tools.

Main Topics: BitTensor as a decentralized AI marketplace (Priority: 5/5): The hosts frame BitTensor as a network where miners, validators, and subnet operators compete to produce valuable AI outputs, with token emissions rewarding useful work across many different applications. Metanova: decentralized drug discovery (Priority: 5/5): Metanova explains how subnet 68 uses miners to search massive chemical spaces for synthesizable molecules, then narrows candidates through heat-picking, CRO synthesis, and wet-lab validation to support a lean virtual biotech model. BitCast: creator economy and attention mining (Priority: 4/5): BitCast describes subnet 93 as a system where creators generate branded videos and are rewarded based on watch time and engagement, aiming to automate creator marketing and democratize access beyond top-tier influencers. Score/Manico: vision models for production use (Priority: 5/5): Score (subnet 44) focuses on distilling large vision-language models into small, task-specific models that can run on CPUs, enabling practical computer-vision applications without expensive infrastructure. AI commercialization and timelines (Priority: 4/5): The episode repeatedly returns to whether AI can materially shorten drug development, content production, and computer vision deployment, with guests arguing that meaningful gains are already emerging and will compound over the next few years. Prediction markets and the AI bubble debate (Priority: 3/5): The hosts discuss a Polymarket contract on whether the AI bubble will burst in 2026, using it as a lens on market expectations, hedging, and the gap between tech-industry optimism and outside skepticism.

Key Arguments: BitTensor works as a marketplace for intelligence production, where different subnets can optimize for very different tasks while sharing the same incentive structure. Metanova argues drug discovery is a high-cost, high-failure process, so improving virtual screening can de-risk the earliest and most expensive stages of R&D. Metanova’s two incentive mechanisms—molecule submission and chemical search algorithms—encourage both open-source innovation and private, flexible search strategies. The drug-discovery workflow still requires synthesis, toxicity filtering, and wet-lab validation; AI improves the funnel but does not eliminate experimental testing. Metanova believes decentralized R&D can reduce costs through geographic arbitrage and CRO partnerships, while still aiming for regulatory-grade results. BitCast argues creator marketing is bottlenecked by admin overhead, and that automating briefs, validation, and reward distribution can unlock the long tail of creators. BitCast claims watch time is a better reward metric than raw views because it better captures engagement and content quality. Score argues the biggest problem in vision AI is not accuracy alone but production economics; distilling models into tiny, task-specific skills makes them deployable on CPUs. Score’s Manico product turns a prompt into a full computer-vision pipeline, including model selection, fine-tuning, deployment, and SDK generation. Across all three subnets, the guests emphasize that adversarial miner behavior can be a feature, revealing weaknesses in scoring functions and improving the system over time.

Data Points: Episode date: March 25th, 2026 - Opening of the podcast Subnet number for Metanova: 68 - Metanova is introduced as subnet 68 Subnet number for BitCast: 93 - BitCast is introduced as subnet 93 Subnet number for Score: 44 - Score is introduced as subnet 44 Drug development cost estimate: $2.6 billion - Metanova cites average drug development cost Drug development timeline estimate: 10 years - Metanova cites average time to bring a drug to market Chemical search space: ~65 billion possibilities - Metanova says a billion-molecule dataset plus five combinatorial reactions expands the search space Initial molecule dataset: 1 billion molecules - Metanova’s starting dataset Combinatorial reactions layered on top: 5 - Metanova’s search space expansion method AI drug assets in late clinical trials: A few assets - Metanova says some AI-developed assets are already in late-stage trials Expected timeline for interesting AI drug results: 3 to 5 years - Metanova’s estimate for seeing meaningful outcomes BitCast creator network size: 2 million subscribers - BitCast says its creator network has grown to this level BitCast creator count: 50 YouTube creators - BitCast says it currently works with 50 creators BitCast growth rate: 40% to 50% per month - BitCast says creator network growth is accelerating BitCast watch time and views growth: 50% to 60% month on month - BitCast reports recent performance growth Creator economy size: $250 billion worldwide - BitCast cites this as the current scale of the creator economy Score model size reduction example: 3.4 GB to 50 MB - Score compares a large model like SAM 3 to a task-specific expert model Current vision validation throughput: 50 to 100 videos/day - BitCast says its AI currently checks this many videos against briefs Target vision validation scale: 100,000 videos/day - BitCast says the system needs to scale to this level AI bubble burst market probability: 24% - Polymarket odds that the AI bubble burst condition occurs by end of 2026 NVIDIA drawdown threshold in market: 50% from all-time high - One condition in the Polymarket AI bubble contract SOXX drawdown threshold in market: 40% from all-time high - One condition in the Polymarket AI bubble contract H100 rental price threshold in market: $1 or lower for five days straight - One condition in the Polymarket AI bubble contract H100 rental price today: About $750/hour - Host cites the data source used for the prediction market BitCast early 2026 creator growth: 60% hours watched, 56% views - Host references BitCast’s 2025 year-end growth figures BitCast current validation track: 50 to 100 videos a day - Score/validation scale mentioned during BitCast discussion

Pivotal Quotes: "It is a marketplace for intelligence production." — Pedro Penna: Pedro summarizes BitTensor’s core concept during the Metanova segment "We believe in vision vibe coding." — Max Septi: Max describes Score/Manico’s go-to-market philosophy for computer vision "The true challenge becomes, can you program their behavior and align them in a way that generates valuable inputs?" — Michaela Bazo: Metanova explains how adversarial miners can be turned into a productive feature

Implications: The episode suggests decentralized AI subnets may become practical infrastructure for drug discovery, creator marketing, and vision deployment. If these systems keep improving, they could lower costs, broaden participation, and make AI outputs more commercially usable.

🔓 Sign Up for Unlimited Episode Search

About This Week in Startups

Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.

View all episodes from This Week in Startups