This Week in Startups | Episode 2268: The $60 Billion Resource Hiding in Space—and the Startups Mining It
Air Date: March 27, 2026
Host: Jason Calacanis (A)
Co-Host: Lon Harris (B)
Guests:
- Matt Gialich, CEO/co-founder, Astroforge (C)
- Sam Dare, Templar/Bittensor (D)
- Yazin Ali Rahim, Open Oats (E)
Episode Overview
This episode centers on high-stakes frontier technology: from mining asteroids for precious metals to radically decentralized AI training and democratized, privacy-first AI note-taking tools. Jason and Lon host in-depth interviews with top founders from Astroforge (asteroid mining), Templar (distributed AI), and Open Oats (open-source, privacy-focused meeting summarizer). The off-duty segment wraps up with sharp banter about IP-driven streaming economics, culture, and tech recommendations.
1. Space Mining Gold Rush: Astroforge and the $60B Asteroid Opportunity
Introduction to Astroforge and Guest
[03:18]
- Matt Gialich joins to talk about Astroforge, a startup developing spacecraft to mine platinum group metals (PGMs) from near-Earth asteroids.
- Jason: “Literally science fiction is reality. Matt, welcome.”
Where Is Astroforge Today?
[03:58 – 07:01]
- Astroforge is building "Deep Space 2," a 200kg craft launching later this year on Falcon 9.
- Target: near-Earth metal asteroids (~10 million miles away; 200m diameter).
- Craft sticks to the asteroid via magnetic landing – “we land with magnets.” (C, [05:27])
- The plan: deploy not one giant vessel but many small, scalable miners.
- Environmental motivation: “Mining is one of the most destructive things we do on this planet…we got one habitable planet…we got to protect it.” (C, [06:11])
Tech Breakdown: How the Space Miner Works
[08:29]
- Uses directed energy (lasers) to vaporize material, sorts metals via magnetism (PGMs are non-magnetic, iron/nickel are).
- Each mission aims to recover up to 1,000kg of value (~$105M per mission at current platinum prices).
Quote:
“We use directed energy i.e. lasers to remove material from the asteroid. And then we sort it out with magnetism... The spacecraft we have is designed to bring back up to 1,000kg of material, which in today's dollars... is worth about $105M per mission.”
— Matt Gialich, [08:29]
Is the Model a Sure Bet?
[11:11]
- Mission cost: $10.4M (launch included).
- Founder’s honest take: “I’d be lying to you if I said I was 100% confident we can do this. …As soon as I’m 100% confident, that means I’m not paying attention to all the things that can go wrong.” (C, [11:11])
Live Streams & Transparency
[12:25]
- No live streaming from asteroid due to bandwidth, but entire control room is live-streamed for public transparency.
- “When ODIN failed, we wrote a ton of articles… on exactly what happened and how we plan on fixing it. Because at the end of the day, that’s how you learn and get better at this.” (C, [12:25])
Past Missions and Investment Realities
[13:18 – 16:17]
- “Odin” (Feb 2025): first commercial FCC-licensed deep space mission, but solar panels didn’t deploy—a learning failure, not catastrophic loss.
- Investor communication is key: “It is all about incremental progress as you go here. I mean, guys, Elon failed three times with Falcon 1 before he got there…” (C, [15:39])
- Investors include Initialized, Y Combinator, and 776.
Market Truths: Are Asteroids Unlike Earth?
[16:27 – 20:17]
- Focus is on economic advantage, not unique materials per se.
- PGMs are essential to many technologies: electronics, chips, displays, cancer drugs, car emissions—“things VCs invest in.”
- “Best platinum group metal mine in the world in South Africa has 14% gross margins. I just told you a story, we could have 90% gross margins.” (C, [20:01])
Next Major Launch?
[20:29]
- November this year, Deep Space 2 launches on Falcon 9.
2. Decentralizing AI Development: Templar and Bittensor’s Ambitious Model
Templar & the New Incentive Layer
[27:59 – 33:25]
- Guest: Sam Dare, founder at Templar/Bittensor subnet
- Tao/Bittensor philosophy: stripped-down blockchain focused strictly on mechanizing trust—“the only thing that matters is incentives.”
- Templar's innovation: trains language models by rewarding miners who improve model loss the most each round—"it's a game."
Quote
“Bitcoin…evolved into many different things…now with Tao…let’s strip out everything from blockchains, all the noise and let’s take their core…incentives and let’s blow it up a thousand feet…Bittensor tries to solve: how do you strip away all the noise?”
— Sam Dare, [27:59]
Technical and Economic Progress
[34:06 – 38:57]
- Any GPU can participate; validation is the key, not hardware specs.
- Last run: trained a 72B parameter model (Llama2-class) distributed, for ~$2–3M payout to miners.
- “It’s not a frontier model…but we are moving so quickly, from 1.2B to 72B parameters in 9 months.” (D, [38:57])
Social and Global Implications
[40:58]
- “The mission is optionality…the mission is choice…There are people who want to train [AI], but can’t.”
- Decentralized compute democratizes pre-training and fine-tuning, especially for “sovereign AI” in developing countries.
What’s Next for Templar?
[43:30]
- Next algorithm will harness any available GPU (not just top-tier H100s), increasing network capacity.
- “Wherever there is an Internet connection and a GPU that is free, we will be able to put that in our runs.” (D, [45:04])
- Expect next run “in a couple of weeks”; next model will focus on utility rather than sheer size.
Investment and Structure
[43:30 – 45:51]
- Corporation mainly to pay staff; Templar exposure is via Tao tokens.
3. Open Oats: Open-Source, Privacy-First Meeting Note-Taker
The “AI Producer” for Conversations
[53:43]
- Guest: Yazin Ali Rahim, creator of Open Oats (formerly Open Granola).
- Built in response to closed-source note-taker drama—emphasis on open code and local processing.
- “The vision behind OpenNotes is not just to create an open version of granola…but local LLMs—so you’d be able not to siphon off your meeting transcripts out to some random third party.” (E, [55:37])
Features Demoed
[58:08 – 61:56]
- Local, ultra-fast transcription (Parakeets model); integrates with any LLM via API/key (Claude, Ollama, etc).
- Real-time, in-context “insight panel” suggests research, checks claims (e.g., when a startup says “no competitors,” it finds them live).
- Highly customizable: control how noisy or quiet it is with feedback.
- MIT licensed, already ported to Windows/iOS by community, designed for privacy and extensibility.
Quote
“It actually shows you other companies based in your knowledge base that also operate in a similar space. So as the founder is…speaking, I thought this might be a case you’d relate to.”
— Yazin, [60:38]
Open Source Power & Auto-Maintaining
[63:24 – 65:54]
- Yazin delegated much of the GitHub management to Claude-based bots; codebase is maintained, bugs triaged, and minor PRs handled automatically.
On the future of open source productivity:
“Everybody can build everything…and then it’s a matter of who stops maintaining. …But now, it’s like when the bar gets so low to build software, everybody’s going to make everything. …And will the Oatmeal project be better than Granola?” (A, [65:54])
4. Off Duty: IP Economics, Streaming, & Stuff We Love
Streaming Economics: Harry Potter, Lord of the Rings, and the Future of Franchise
[74:13 – 79:56]
- HBO’s Harry Potter reboot: $100M/episode, could cost $5.6B across 56 episodes. (B, [75:03])
- Jason’s math: With only a 4% subscriber bump (5 million users x $1,000 LTV), economics “start to math out.”
- New adaptations offer global subscriber growth and reignite high-value franchise merchandise (theme parks, wands, etc).
Music, Culture, and Tech Tips
- Whisper Flow: Jason’s must-have AI tool for dictation, better than Siri (“Whisper Flow is like perfect Siri…this is the must-have device of the year for me.” — A, [87:21])
- Music: Tedeschi Trucks Band, Lily Allen’s “West End Girl”—the benefits of concept albums as performance.
- TV: “Paradise” (S2), “American Sweatshop,” and “The Screener.”
Jason: “I love a good post-apocalyptic thriller.”
Notable Banter
- Open source vs SaaS: “Will the open source be able to keep up with Granola, or will many hands make light work and will the Oatmeal project be better than Granola?” (A, [65:54])
- Discussion of Disney adults, theme park economics, and franchise fanaticism—for humor fans.
Notable Quotes
-
On transparency & iteration:
“I believe in transparency up and down…When ODIN failed, we wrote a ton of articles…because at the end of the day, that’s how you learn and get better.”
— Matt, [12:25] -
On scaling a new industry:
“Can you make hundreds, if not thousands of [small craft], that are going to go out to these asteroids, mine them and bring it back to Earth…”
— Matt, [06:11] -
On decentralized AI’s mission:
“The mission is optionality…There are a lot of people that want to train but they can’t…It’s not even about competing with [big AI labs]; it’s creating a market that never existed.”
— Sam, [40:58] -
On open source productivity explosion:
“Everybody can build everything…and then it’s just going to be a matter of who stops maintaining. …If will the Oatmeal project be better than Granola?”
— Jason, [65:54]
Key Timestamps
- Astroforge intro: [03:18]
- Asteroid mining process: [08:29]
- Risk honesty: [11:11]
- Transparency/culture: [12:25]
- Market impact of asteroid platinum: [18:53]
- Next Astroforge mission: [20:29]
- Templar decentralized AI: [27:59]
- Training 72B model, cost: [38:21]
- Global, political benefits: [40:58]
- Open Oats demo: [58:08]
- Open source/AI-automated project management: [63:24]
- Streaming economics - Harry Potter/LOTR: [75:03]
- Off Duty recommendations (Whisper Flow, music, TV): [85:26–94:10]
Final Thoughts
This episode is a masterclass in frontier tech—from the mind-bending logistics and necessary humility of asteroid mining, to the bold new economics and politics of distributed AI, to the explosive creative power of open-source developer tools. Jason and Lon’s blend of deep questions, wit, and practical business thinking makes this essential listening for anyone invested in the future of tech, science, and entertainment.
