The $60 billion resource hiding in space, and the startup trying to mine it | E2268

Key Concepts

• Asteroid Mining: The commercial extraction of Platinum Group Metals (PGMs) from near-Earth asteroids.
• Decentralized AI Training: Using distributed networks (BitTensor subnets) to train large language models (LLMs) by incentivizing miners with tokens.
• Incentive Engineering: Designing game-theoretic rules to force self-interested actors (miners) to contribute compute power toward a collective objective.
• Sovereign AI: The ability for nations or organizations to train their own AI models independently of major tech monopolies.
• Real-time AI Agents: Software that listens to live conversations to provide context, fact-checking, and research insights.

1. AstroForge: Mining Near-Earth Asteroids

Matt Gallich, CEO of AstroForge, discussed the company’s mission to extract Platinum Group Metals (PGMs) from asteroids.

• Technical Approach: Unlike sci-fi depictions, landing is a delicate docking procedure. The company targets "metal asteroids" (approx. 70% iron/nickel) and uses magnets to secure the spacecraft.
• Extraction Methodology: They utilize directed energy (lasers) to remove material, which is then sorted via magnetism.
• Economic Viability: A single mission costs ~$10.4 million and aims to return up to 1,000 kg of material, valued at ~$105 million. Gallich argues this is more economical than terrestrial mining, which faces high costs and environmental destruction.
• Mission Status: The company is preparing for "Deep Space 3" (Vestri) in November, following the "Odin" mission, which experienced a power deployment failure.

2. Templar: Decentralized AI Training

Sam Dair, founder of the Templar project on the BitTensor network, explained how they trained a 72-billion parameter model using decentralized compute.

• Incentive Structure: Templar operates as a "magic factory" where miners compete to reduce loss in training epochs. The best-performing miners receive the most rewards (in subnet tokens).
• Technical Breakthrough: They developed "Heterogeneous Sparse LoCo," allowing training to occur across diverse hardware (from data centers to consumer GPUs), making the network indifferent to specific network topology.
• Cost Efficiency: The 72B model was trained for approximately $2–3 million, significantly cheaper than traditional frontier model training.
• Mission: The goal is to provide "optionality" and "sovereign AI," allowing entities that lack $10M+ budgets to train their own models.

3. OpenOats: Real-time AI Notetaking

Yazin Ali Raheem demonstrated "OpenOats," an open-source, local-first AI agent for meetings.

• Functionality: It listens to live conversations, performs speaker diarization, and surfaces real-time insights or fact-checks based on a knowledge base.
• Privacy: By running locally (using models like Parakeet or Olama), it prevents sensitive meeting data from being siphoned to third-party cloud providers.
• Automation: The project is maintained by an AI agent ("AutoMaintainer") that researches bugs, opens pull requests, and merges fixes, allowing the developer to manage the project with minimal manual intervention.

4. Media & Entertainment Economics

The hosts analyzed the shift in media consumption and IP monetization:

• Lord of the Rings: Peter Jackson is producing new films (e.g., The Hunt for Gollum) to fill gaps in the lore. The hosts noted that Warner Bros. is limited to adapting only The Hobbit and The Lord of the Rings due to rights restrictions.
• Harry Potter Series: HBO is producing a series adaptation of the books, with an estimated budget of $100M per episode. The hosts argued the math "maths" because the series will drive long-term subscriptions and theme park revenue, targeting a new generation of "super fans."
• The "True Fan" Thesis: Referencing Kevin Kelly, the hosts emphasized that a small base of "super fans" (approx. 5 million) can sustain multi-billion dollar franchises through lifetime value (merch, theme parks, subscriptions).

5. Notable Quotes

• Matt Gallich (AstroForge): "There’s no other way to have our way of life unless we go out in the cosmos and mine it for its resources."
• Sam Dair (Templar): "The mission is optionality. The mission is choice... we can’t make the same mistake [as Bitcoin] of thinking we were going to replace JP Morgan."
• Jason Calacanis: "Everything gets replaced at some point... a better version of almost everything comes out."

Synthesis/Conclusion

The episode highlights a convergence of "brutal capitalism" and technological innovation. Whether it is mining asteroids for rare metals, using decentralized incentives to train AI, or building open-source agents to automate productivity, the common thread is the lowering of costs to unlock new markets. The hosts conclude that the current era of tech is defined by founders who can build "better, faster, cheaper" products that either educate, entertain, or fundamentally improve human efficiency.