Sarah Guo - Funding the Frontier - [Invest Like the Best, EP.489]
What this covers
The conversation is with Sarah Guo, founder of Conviction, a venture firm that backs AI-native companies from the earliest stage. It moves between what the people building AI currently believe and how an investor should act on that. Guo is skeptical of forecasting which layer of the stack the frontier labs will win. She would rather judge each effort on its merits and learn from founders who build things that surprise her. She cites missing Suno, having underestimated demand for expression and entertainment tools. She also describes raising Conviction's first fund on a reputation for spotting extraordinary people rather than a layered strategy, and she warns that investors who lean on founder pedigree and who else invested are making a dangerous, less fundamental kind of decision.
The range is wide. On infrastructure, she calls compute the key constraint for the next five to ten years and likens compute independence to energy independence. She locates the American bottleneck in regulation and public sentiment rather than capital. She defends the open-source model as the route by which capability reaches the broader economy, and argues that restricting it would hamper law-abiding businesses more than adversaries. Sector by sector, she covers robotics, where Sunday Robotics expects semi-humanoid robots in homes in beta by year's end, and biology, where software platforms like Chai Discovery could bypass the biobucks deal model. She also covers legal work as a natural fit for language models, and the contested claim that venture-backed semiconductors are now a different bet. She hopes Jevons paradox will show up across domains as agents absorb mundane work.
Guo argues that AI capability is diffusing too broadly to be owned by a few frontier labs, so investors should back founders through the technology transition by taking early, conviction-based bets on domains where models create new value, while protecting their judgment from narrative and pedigree noise.
- Open-source models are already competitive and widely used, so capability diffusion is effectively irreversible
- Physical-world constraints such as compute, energy, and supply chains determine who can build the frontier, so investing there matters
- Early-stage investing rewards asymmetric, first-principles understanding of markets that others have mispriced, such as biology, robotics, and law
Investment decisions should rest on builder and customer evidence, not grand strategic frameworks.
- Guo argues investors should learn from founders who are building surprising products or believe something the investor does not, rather than reasoning about grand strategic frameworks for which AI layer will win, because such frameworks are not useful for decision-making.
“Coming up with some grand strategic framework for what layer is going to win here, I think is not useful to me.”
- Guo says Conviction's debates typically start from aligned beliefs about the direction of technology and then ask whether the market is friendly to a venture-backed company, and whether the distribution of outcomes is worth betting on, a question that applies to solar, batteries, nuclear, turbines, robotics, and biology.
“Space is not a friendly market to a venture-backed company, but is it possible? Is the distribution of outcomes worth betting on?”
- Guo says that evidence of an AI biology company's value comes most reliably from talking to the scientist or tool user at the customer, since the customer knows if the product is valuable, rather than from waiting for public proof.
“The other piece is actually you just talk to the scientist at the customer or somebody who leads a— ... Well, the customer knows if it's valuable or not.”
- Guo says the appeal of investing in Harvey was the ambition of what seemed possible at the time and the technical logic of why it would work, which she believes is a different decision framework from how others approached it then.
“The thing that appealed to me in that moment was the ambition of what was possible then and the technical logic of why it would work.”
- Guo argues that a fund's differentiated positioning can't be known before contact with the market, so raising money early and experimenting is better than pre-building a story.
“I think you don't know until you make contact with reality and think about it and you're in the market.”
- Guo says Sunday Robotics' founders were impressive because in under two years they went from a basement prototype to a full-stack system through hundreds of hardware iterations, arguing that nothing is true until it is shipped.
“Nothing is true until it is shipped.”
Firm success depends on setting high standards for collaborators and learning across markets.
- Guo argues that the firm's success depends more on the standard it sets for the people it works with than on raw effort, and that Conviction started with preexisting relationships that helped.
“It's more about what your bar is for the people you work with.”
- Guo says that the firm's early success came from a relatively uncompetitive generational transition, a massive technology shift, and a focused bet on understanding the technology and community, combined with preexisting relationships she and Mike already had.
“My read of the environment was the growth of firms and generational transitions that were happening meant that it wasn't the most competitive landscape in early that it has been.”
- Guo argues that a small, ecosystem-oriented firm learns more by spending time with investors across asset classes and with public-markets people who think differently, because early-stage investors in a large firm can become myopic.
“I've learned so much from getting to know investors who think differently than I do, including different asset classes.”
AI reshapes pharma software economics while physical buildout lags behind software progress.
- Guo argues that AI changes the economics of pharma software: unlike the conventional biotech model of making drugs and seeking biobucks deals, a computational platform like Chai Discovery can sell to top pharma companies, which she believes will produce a large acceleration in cures.
“You can create and capture enormous value with models in biology, and there could be different AI software in biology.”
- Guo argues that the next five to ten years of AI scale depend on building energy, data centers, and industrial capacity, and that the physical rollout is slower than software progress.
“The version of us having enough compute over the next five to ten years.”