The discussion highlights the critical importance and scarcity of Nvidia compute resources for AI startups, with firms like Conviction proactively securing hardware to overcome supply constraints amid soaring demand driven by enterprise AI applications. It also emphasizes the evolving AI ecosystem where strategic control over infrastructure, models, and applications is key to maximizing value and capitalizing on the vast market opportunities.
The discussion centers around the critical role of compute resources, particularly Nvidia chips, in the AI startup ecosystem. Sarah Guo, founder of the AI-native venture firm Conviction, highlights that many portfolio companies face compute as their biggest constraint. These startups often start by accessing Nvidia hardware through cloud providers but sometimes acquire direct control over clusters depending on their needs. Early-stage companies typically experiment with the latest Nvidia chips to innovate, while more mature companies focus on training smaller models to optimize costs and improve user experience.
Guo explains that her firm took the unusual step of purchasing Nvidia compute resources upfront for their portfolio companies to mitigate timing risks and ensure access amid ongoing supply shortages. The market has experienced significant constraints, with demand far outstripping supply, making it difficult for startups to secure on-demand, small-scale compute resources. This shortage has led to unprecedented scenarios where companies attempt to buy hundreds of millions of dollars worth of compute capacity with multi-year commitments, underscoring the intense competition for Nvidia hardware.
The conversation also touches on Nvidia CEO Jensen Huang’s perspective, who consistently emphasizes the parabolic growth in demand for Nvidia chips. Despite the supply challenges, Huang remains optimistic about the company’s ability to meet this demand, though many in the industry find it hard to keep pace with the rapid growth. Guo agrees with Huang’s assessment of demand, noting that the explosion in cloud code revenue is just the beginning, driven by the rise of long-horizon AI agents that enhance productivity across various knowledge economy functions beyond just coding.
The discussion shifts to the broader AI market opportunity, referencing SpaceX’s S-1 filing, which highlights a $26.5 trillion total addressable market focused heavily on enterprise applications. Guo emphasizes that much of the AI investment is enterprise-centric, with automation being a key driver—AI models are increasingly used to automate existing tasks, improving efficiency and productivity. The commitment by companies like SpaceX to enter the enterprise AI space signals a significant strategic focus on leveraging AI to transform business operations.
Finally, the conversation explores the debate over where the most value lies in the AI stack—whether in infrastructure, models, or applications. Guo points out that owning the infrastructure and the capability to build models is highly valuable, as seen in deals involving companies like Anthropic and Cursor. However, there remains a question about whether companies also need to control the model and application layers to maximize value. Overall, the ecosystem is evolving rapidly, with compute access and strategic positioning across the AI stack being critical factors for success.