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

Bg2 Pod

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

1h 20mJun 11, 2026
Key Themes
SpaceX IPOAI computeStarlink growthlaunch economicsfrontier modelsdata center capexmarket dispersionmodel benchmarking
Summary

SpaceX’s IPO, AI compute ambitions, and the scale of the coming AI buildout dominate a bullish, infrastructure-heavy market discussion.

This episode centers on SpaceX’s expected IPO and the surprising breadth of its business story: launch economics, Starlink growth, and a possible role as a large-scale AI compute provider. The conversation broadens into frontier-model competition, the economics of data centers and orbital compute, and whether the current wave of AI capital expenditure can be justified by revenue growth. The hosts also step back for a market check, noting that semiconductors remain strong while parts of software and internet have lagged. Overall, the tone is highly constructive on long-term AI and space infrastructure, while acknowledging near-term IPO volatility and broader market caution.

1
Reusability is the key economic lever in space launch

The episode repeatedly ties SpaceX’s valuation story to lower launch costs, higher cadence, and rapid reusability. That combination is presented as the foundation for everything from Starlink expansion to more speculative opportunities like orbital compute.

2
SpaceX’s story is bigger than rockets

The discussion frames SpaceX as a business with multiple growth vectors, including Starlink broadband, direct-to-cell service, and potentially large-scale AI compute. That makes the company’s future less about a single product and more about infrastructure and platform economics.

+1 more
3
Frontier-model quality may be undercounted by simple benchmarks

The hosts argue that long-running tasks, sustained reasoning, and real-world workflows are more revealing than short benchmark snapshots. That means the market may underestimate the gap between model generations if it focuses only on static tests.

4
Compute demand remains the core bottleneck and opportunity

Across the episode, the speakers return to the same idea: AI demand is rising faster than supply, and willingness to pay for compute is improving. Whether the hardware sits on Earth or, eventually, in orbit, the scarce resource is still high-quality compute capacity.

5
AI monetization appears to be scaling faster than skeptics expected

The panel challenges the idea that AI revenue remains too small to support the current buildout. They point to stronger-than-expected revenue growth, better margins, and early-stage adoption as reasons the capex math may work.

6
Market leadership is narrow, but the AI trade is still broadening

The closing market check suggests semiconductors have led, while parts of internet and software have lagged. That contrast implies a market that is still sorting winners, even as the underlying AI investment cycle continues to expand.

7
Long-duration reasoning may become a more important measure of AI progress

The discussion around Fable 5 and similar models suggests that the next phase of AI evaluation will need to account for tasks that unfold over time. If that view is right, both product design and benchmark design may need to evolve.

Select any chapter text to Deep Dive with AI
01SpaceX IPO and Starlink Growth Drivers

The episode opens with a framework for the major discussion points and then focuses on SpaceX’s impending IPO, launch economics, and Starlink’s growth runway. The hosts emphasize that reusability and higher launch cadence are central to SpaceX’s valuation story, while Starlink remains early in penetration and could expand through both broadband and direct-to-cell services.

The hosts frame the episode’s main themes: SpaceX’s IPO, AI competition, and a Taiwan/GPU market check.
SpaceX’s valuation is tied to launch cadence, rapid reusability, and operating leverage in the launch business.
Starship reusability is presented as a key economic unlock for lower launch costs.
Starlink is described as still early, with substantial room to grow in broadband and direct-to-cell connectivity.
The chapter also introduces the idea that AI deals and compute economics may influence the broader competitive landscape.
02SpaceX's AI Compute Push: Terrestrial Data Centers and Orbital Optionality

The conversation shifts to SpaceX as an emerging AI compute provider, with major compute deals changing how the company is viewed. The hosts argue that data-center execution, engineering speed, and premium monetization can make the terrestrial business compelling, while orbital compute remains an optional upside rather than a requirement for the IPO case.

SpaceX is reframed as a possible hyperscaler-scale AI compute provider.
The panel stresses first-principles engineering and rapid deployment as competitive advantages.
Large compute deals help validate the terrestrial AI infrastructure story.
Orbital compute is treated as a call option, not the core IPO thesis.
Launch reusability and reliability are presented as the main enablers of future space-based compute.
03Cursor acquisition, xAI model upside, and SpaceX IPO drawdown risk

The discussion examines how Cursor’s acquisition could strengthen xAI through better talent, data, and model-building capability. The speakers then debate the SpaceX IPO bull and bear cases, balancing strong long-term optionality against the risk of post-IPO volatility, lock-up dynamics, and the challenge of sustaining rapid revenue growth.

Cursor is presented as a valuable coding-agent company with strong data and talent.
The acquisition is framed as potentially improving xAI’s frontier-model capability.
SpaceX is compared to an AWS-style capacity-first, monetization-later strategy.
The bull case combines core business strength with AI compute and model upside.
The bear case centers on valuation, revenue growth expectations, and typical IPO drawdown risk.
Lock-up structure and Elon’s ownership are discussed as limiting near-term supply pressure.
04Fable 5, Mythos, frontier models, and AI capex

This chapter focuses on frontier-model performance, especially the idea that long-running intelligence is difficult to capture with benchmarks alone. The hosts argue that frontier models are still capturing most economic value even as open-source models gain usage, and they connect model progress to rising compute demand and infrastructure spending.

Benchmarks may understate real-world model capability.
Frontier models are still capturing most of the revenue.
Open-source models may take a large share of usage for lower-value tasks.
Model routing and multi-agent orchestration are becoming more important.
Better models can increase token usage and compute demand.
The chapter links these trends to rising AI capex.
05AI CapEx vs Revenue Math Check and the Next Trillion

The closing chapter weighs whether the current wave of AI infrastructure investment can be justified by revenue growth and margins. The panel argues that demand is accelerating, monetization is improving, and the long-term market opportunity remains very large, even as near-term conditions look mixed across sectors and macro indicators.

The key question is whether AI capex can be justified by current and projected revenue.
The speakers argue revenue and margins may be strong enough to support the buildout.
Demand for compute appears to be outpacing supply.
Only a small fraction of users are said to be engaging with AI agentically today.
The market check notes strong semiconductors and weaker internet/software performance.
The episode closes with a very bullish long-term view on revenue growth and GDP impact.