Ep. 13: Sarah Guo: Open Source Is Already Won. Also, Compute Independence Needs Nuclear Power. (2026)

Episode guide

Sarah Guo’s latest Invest Like the Best conversation carries two claims that sound opposed until you separate the software layer from the physical layer. She argues that open-source AI is already widespread, and that restricting it would mainly slow lawful American businesses. In the same interview, she argues that the real strategic bottleneck is physical compute: power, data centers, nuclear approvals, and semiconductor supply-chain redundancy.

Alex and Priya use that tension to ask what Guo is defending, what Conviction’s public strategy makes visible, and what the argument signals about where capital may flow next. The episode treats the incentive reading as analysis, not proof of private motive or portfolio causality. Public sources do not disclose Conviction’s LP composition, ownership percentages, or carry exposure.

Statement Timeline

All times below use UTC+08:00, the channel’s listener display timezone. X engagement figures are captured platform snapshots, not historical counters from the moment of posting.

September 1, 2026, 22:00:00: Patrick O’Shaughnessy announced the new Invest Like the Best episode and published its chapter map, including “The Case for Open Source” and “America’s Compute Independence.” The post showed 326,837 views, 549 likes, and 47 reposts when captured. 1
September 1, 2026: The canonical Invest Like the Best video page was published. Its description identifies the discussion areas as frontier AI researchers, compute, open-source AI, compute independence, robotics, biology, and Sarah Guo’s investing framework. 2
September 2, 2026, 00:46:05: Guo announced that she had sat down with Patrick O’Shaughnessy to discuss Conviction and the AI revolution. The post showed 35,262 views, 192 likes, and 11 reposts when captured. 3
September 2, 2026, 05:00:02: O’Shaughnessy posted the clip containing Guo’s open-source claim: “If you restrict use of open source models in the United States, you'd basically just restrict law-abiding American businesses and slow them down.” The clip showed 40,701 views, 90 likes, and 13 reposts when captured. 4
September 2, 2026, 08:21:12: Guo amplified that clip from her own account. The two claims in this episode came from the same interview, not from positions that evolved over time or a documented flip-flop. 5
September 1–2, 2026: Secondary coverage summarized Guo’s physical-infrastructure argument as a focus on data-center siting, nuclear approvals, natural-gas supply, advanced packaging, and redundancy rather than total autarky. That summary is used as corroboration, while the original interview and Guo’s own posts remain the primary evidence. 6

What the episode decodes

The apparent contradiction is a layer distinction. Guo’s open-source argument concerns who can use and distribute models. Her compute-independence argument concerns who can reliably supply the electricity, facilities, chips, and manufacturing capacity that make large-scale AI possible.

That distinction matters for Conviction’s public positioning. Guo’s site describes Conviction as an AI-native investing firm founded in 2022. A 2026 profile by Colossus describes a firm with three funds and nearly $1 billion in total capital, and identifies investments including Baseten, Harvey, Sierra, Cognition, Mistral, Thinking Machines, and OpenEvidence. Those facts establish exposure to infrastructure, applications, and model-layer companies; they do not prove that any one statement was made to benefit a particular holding. 7 8

The strategic reading is narrower and more useful. If open models become widely available, value can move toward the infrastructure and application layers that make those models usable. If physical compute remains scarce and geographically concentrated, then power, data-center development, advanced packaging, and resilient supply chains become strategic bottlenecks. Guo’s two claims together narrate a market in which software access broadens while the rails underneath it remain capital-intensive.

For founders, that creates a test: is a business merely benefiting from cheaper or more open models, or does it control a scarce workflow, deployment layer, physical input, or source of proprietary demand? For LP-side allocators, the parallel test is whether a fund’s public thesis maps to observable exposure and repeatable access, or whether the thesis is mostly a persuasive description of a market it hopes will form.

The episode does not resolve the question for the listener. It makes the layers, incentives, and unknowns easier to inspect.