The SpaceX IPO, Fable 5, AI Capex Update & Market Check w/ Gavin Baker, Andrew Fox & Clark Tang | BG2

Episode Summary

Executive Summary: The episode argues that SpaceX’s upcoming IPO is underpinned by three overlapping growth engines: launch/Starlink, terrestrial AI compute, and a likely upside option in frontier models via Cursor/XAI integration. The speakers contend that AI demand is still outrunning supply, frontier models are capturing most revenue, and SpaceX’s engineering speed could make it a dominant AI infrastructure player as well as a space company.

Main Topics: SpaceX IPO valuation and core growth levers (Priority: 5/5): The panel breaks down the IPO by separating launch, Starlink connectivity, and AI compute, arguing investors should evaluate each business from first principles rather than focus only on headline valuation. Terrestrial AI compute as a major revenue driver (Priority: 5/5): Speakers emphasize SpaceX’s ability to stand up data centers quickly, monetize compute at high margins, and potentially become a top hyperscaler-style player through deals with firms like Google and Anthropic. Starlink, direct-to-cell, and broadband expansion (Priority: 4/5): They argue Starlink remains early in penetration, with room to scale across households, mobility, aviation, and direct-to-cell use cases, especially if Starship reusability lowers launch costs. Orbital compute economics (Priority: 4/5): The discussion presents space data centers as a call option that could materially lower capex versus terrestrial builds, but not as a requirement for the IPO thesis. Frontier models, long-running agents, and superintelligence race (Priority: 5/5): The speakers debate Fable/Mythos and the idea that long-running test-time compute has expanded model capability and demand, keeping frontier labs like Anthropic and OpenAI economically dominant. Open source vs frontier model economics (Priority: 4/5): The panel argues open source may win on token share while frontier models keep most economic value, implying continued growth in total compute demand and continued benefits for infrastructure providers. NVIDIA, ASICs, and compute bottlenecks (Priority: 3/5): The conversation covers why NVIDIA remains strong despite ASIC competition, with more customized accelerators emerging but demand still constrained by power, supply, and deployment speed.

Key Arguments: SpaceX is not just a launch company; it is evolving into a multi-engine platform spanning connectivity, compute, and frontier AI models. Rapid reusability of Starship is the key gating variable for lowering launch costs and unlocking both high-volume Starlink expansion and orbital compute economics. SpaceX can stand up data centers faster than competitors, and speed is itself a financial advantage because idle time on power, land, and equipment destroys returns. The Google and Anthropic deals imply materially higher monetization per gigawatt than prior market assumptions, making terrestrial AI compute highly profitable even before orbital compute. Frontier models are currently capturing the majority of AI economic value because long-running agents and test-time compute have made them materially more useful. Open source may take a larger share of token volume, but that does not necessarily reduce demand for compute; it may actually increase it by expanding total usage. NVIDIA remains the dominant hardware provider because customers value performance per watt and deployment reliability, even as custom ASICs gain relevance for certain workloads. The IPO should be viewed as a long-term ownership opportunity; short-term volatility is expected and not central to the thesis.

Data Points: SpaceX IPO price: $135/share - Referenced as the IPO price in the discussion. Implied IPO valuation: $1.77 trillion - Valuation cited for the upcoming SpaceX IPO. Reported 2028 revenue forecast: $160 billion - Wall Street Journal / Goldman forecast referenced for SpaceX revenue by 2028. Current trailing revenue: $18 billion - Used by bears as the comparison point against the IPO forecast. Revenue multiple implied by current forecast: 39x trailing revenue - Stated after including recently signed AI deals, down from roughly 100x trailing revenue. AI monetization per gigawatt: $14 billion per gigawatt per year - Implied monetization rate from the leaked $160 billion revenue number. Anthropic deal monetization: $22–23 billion per gigawatt per year - Cited as SpaceX/Google-Anthropic style monetization benchmark. Google deal monetization: $50 billion per gigawatt per year - Cited as an especially high monetization example. Colossus One IRR: 55% IRR - Referenced as a prior return calculation on AI compute deployment. Data center build speed: 122 days - Jensen/Elon claim for how quickly data centers can be stood up. Starship cadence last year: 160–165 launches - Baseline launch volume used to frame future growth. Expected launch cadence: High hundreds, then thousands of launches - Speakers project a rapid scale-up over the next several years. Starlink global household penetration: Less than 1% - Used to argue the broadband business is still early-stage. Direct-to-cell / connectivity revenue expectation: $10 billion to $50 billion by 2028 - Range discussed from bank models and forecasts. Cost per kilogram on Falcon: About $1,500/kg - Baseline launch cost cited for comparison with Starship. Cost per kilogram target on Starship: About $250/kg or lower - Expected with two-stage and rapid reusability. Orbital AI satellite payload: About 5 megawatts per Starship launch - Derived from satellite mass and launch capacity discussion. Orbital compute capex: About $5 billion per gigawatt - Estimated space-based compute build cost before losses/failures. Terrestrial data center capex: About $20–25 billion per gigawatt - Comparable cost for land, shell, power, and cooling on Earth. Total ground gigawatt cost: About $60 billion per gigawatt - Broken into roughly $35B GPUs/silicon and $25B infrastructure. AI compute market share reference: Number four hyperscaler - Claimed SpaceX would become the #4 hyperscaler after the Google deal. Frontier model frontier count: 4 companies - XAI/SpaceXAI, Google, Anthropic, and OpenAI were described as on the Pareto frontier. Enterprise AI adoption: Less than 0.2% of people on Earth using AI agentically - Used to argue that adoption is still in an early phase. Projected AI lab revenue next year: Around $300 billion - Used in the CapEx vs revenue discussion. Morgan Stanley 2027 CapEx forecast: $1.1 trillion - Updated from a prior forecast of $950 billion. Altimeter position sizing change: Large to medium-small - Portfolio stance mentioned after prices ran up. Frontier revenue share: ~90%+ - Claim that frontier models are capturing the majority of AI economic value. OpenAI / SpaceX customer relationship: Jalapeno chip discussed - OpenAI’s custom chip was referenced as a successful ASIC example.

Pivotal Quotes: "If you're AI pilled, that means we got to build a lot more compute than the world thinks." — Speaker 1: Opening thesis connecting AI belief to infrastructure spending and SpaceX’s opportunity. "Speed is literally cost because every day you're paying electricians and plumbers, that's cost." — Gavin Baker: Explains why rapid data center deployment is a competitive and financial advantage. "I don't know another entrepreneur or another business that's a better bet on the future than SpaceX." — Speaker 1: Summarizes the bullish IPO view and long-term conviction in the company.

Implications: If the speakers are right, SpaceX becomes a core AI infrastructure owner as well as a space leader, with demand for compute and connectivity still early. Investors should expect IPO volatility, but the strategic thesis is long-duration and potentially transformative.