TikTok Ban, Data Centers in Space… and What's a Vector? | E2073

Episode Summary

Executive Summary: The episode opens with major tech news and then features two deep-dive interviews: Weaviate’s Bob van Luijt explains why vectors, vector databases, RAG, and agents are foundational to modern AI infrastructure; Lumen Orbit’s Philip Johnston argues space-based data centers can become economically viable as launch costs fall and compute demand rises. Both segments frame 2025 as a pivotal year for new infrastructure layers.

Main Topics: Breaking startup and policy news (Priority: 5/5): Alex covers the Supreme Court’s TikTok divestiture/ban ruling, the looming inauguration, and key venture/crypto funding headlines to set the tone for a consequential week ahead. Weaviate and the vector database stack (Priority: 5/5): Bob van Luijt walks through vectors, embeddings, vector indexing, and how vector databases store and search unstructured data for AI applications. RAG, transformers, and the rise of AI infrastructure (Priority: 5/5): The interview explains how deep learning and transformers made embeddings practical at scale, enabling retrieval-augmented generation and modern AI app development. Weaviate’s business model and enterprise growth (Priority: 4/5): The discussion covers open source distribution, serverless offerings, BYOC, cloud partnerships, and how enterprise adoption is driving revenue growth. Agents and feedback loops (Priority: 4/5): Bob defines agents as systems that act on data rather than merely retrieve it, framing them as real technical architecture and also a market term. Lumen Orbit and space-based data centers (Priority: 5/5): Philip Johnston outlines a vision for moving compute into orbit using abundant solar energy, passive cooling, and falling launch costs. Economics, launch costs, and infrastructure scale (Priority: 5/5): The Lumen segment emphasizes that the decisive variable is launch cost, with space data centers potentially outperforming terrestrial ones on total cost of ownership at scale.

Key Arguments: Vector databases are a critical AI infrastructure layer because embeddings let machines organize unstructured data by semantic distance, not just keywords. Deep learning transformed embeddings from an academic concept into a practical tool by making distance prediction scalable rather than brute-force. Weaviate’s value is not just search; it enables similarity search, hybrid search, RAG, and now agentic feedback loops that make AI applications more useful. Open source plus managed/cloud deployment has helped Weaviate capture startup adoption first and then enterprise demand, which typically pays more and grows larger bills. Agents are real: they are systems that do something with data, not just retrieve it, and can create feedback loops that improve data quality. Lumen Orbit’s thesis is that if launch costs fall enough, it becomes cheaper to power and cool compute in space than to build more terrestrial data centers. Space data centers benefit from free solar power, passive radiative cooling, and large modular scaling, which may let them expand faster than Earth-based infrastructure. The biggest constraint for orbital compute is not physics but launch economics and reliable heavy-lift rockets such as Starship and New Glenn.

Data Points: Bench liabilities: $65.4 million - Alex cites TechCrunch reporting on former accounting startup Bench’s liabilities in its collapse. Phantom raise: $150 million - Crypto wallet Phantom raised at a $3 billion valuation amid renewed crypto optimism. Phantom valuation: $3 billion - The company’s latest funding round implied multi-unicorn status. Insight Partners fundraising: $12.5 billion - Insight raised multiple new funds, including a flagship fund and a dedicated buyout co-invest fund. Weaviate customer growth milestone: first $1 million in revenue in an instant - Bob describes rapid uptake after launching the serverless offering and enterprise interest. Weaviate headcount: just crossed 100 people - Bob notes the company has scaled its team significantly. Weaviate customer adoption: 5 lines of Python code - Bob says developers can now build these AI apps very quickly by connecting models and Weaviate. Weaviate enterprise ecosystem: AWS, GCP, Azure - Alex references Weaviate’s cloud partnerships and deployment options. Lumen Orbit demo satellite mass: 50 kilograms - Philip describes the first demonstrator satellite. Lumen Orbit demo satellite power: 1 kilowatt - Philip says the first demonstrator will be about one kilowatt. On-orbit compute comparison: 100x more powerful GPU computer than has ever been flown in space - Philip says the demo will use state-of-the-art terrestrial NVIDIA chips. Solar panel output in space: 200 watts per square meter - Philip uses this to explain the energy available for orbital data centers. Radiator dissipation: 800 watts per square meter - Philip explains the cooling math for space-based compute. Target solar array: 4 km by 4 km - Philip describes the scale needed for a roughly five-gigawatt data center. Target radiator size: 1 km by 1 km - Philip says the radiator area needed is about a quarter of the solar panel surface area. Space data center power target: 5 gigawatts - Philip uses this as the scale for the long-term vision. Launch cost threshold: 10x to 100x lower in 5 years - Philip says the investment thesis depends on launch costs falling dramatically. Ground data center electricity cost example: $140 million - He compares four years of power cost for a 40-megawatt terrestrial data center at 10 cents/kWh. Launch cost example: $10 million or more - Philip contrasts launch cost with terrestrial electricity expense for the same compute over four years. Data center deployment time: running in a month - Philip claims multiple space modules can be assembled into a 200-megawatt data center quickly. Orbital altitude tradeoff: below 400 km / around 1,200 km - Philip explains debris risk vs. radiation tradeoffs across very low and higher orbits.

Pivotal Quotes: "What we can do is we can count words in a sentence... we can work with it and we do that by distance calculations." — Bob van Luijt: Bob explains the core intuition behind vectors and embeddings. "What is an agent? An agent does something with your data." — Bob van Luijt: Bob defines agents as active systems that transform data rather than merely retrieving it. "If you don't believe that the launch cost is going to come down by 10x in the next five years, we are not a good investment." — Philip Johnston: Philip states the key venture condition behind Lumen Orbit’s space-data-center thesis.

Implications: The episode signals that AI infrastructure is moving beyond models into databases, orchestration, and new physical compute layers. If these theses hold, 2025 could reward startups building developer tools, data quality systems, and off-Earth infrastructure.