Descriptions:
In this Bloomberg Technology interview, Sarah Guo—founder of Conviction, an AI-native venture firm—offers a rare insider account of how GPU compute scarcity is shaping the AI startup ecosystem in 2026. Speaking from direct experience, Guo describes her firm’s decision to pre-purchase H100 nodes on behalf of portfolio companies to hedge against supply timing risk, and recounts personally attempting to negotiate $100 million-scale compute commitments with major cloud providers only to be turned away—a scenario she calls unprecedented.
Guo corroborates Jensen Huang’s repeated assertions that AI chip demand is parabolic, noting roughly two quarters of worsening supply stress for startups. She observes that on-demand, small-scale compute is particularly difficult to access—a critical gap for early-stage companies that need to experiment before committing to large contracts. Startups across all layers (infrastructure, model, and application) currently default to current-generation Nvidia hardware for frontier performance, shifting to smaller, cost-optimized models only as they mature.
The conversation extends to how long-horizon coding agents are driving a new wave of token-heavy consumption—visible in metrics like Claude Code revenue—and why Guo expects this pattern to expand across all knowledge-economy functions. The segment also addresses SpaceX’s S-1 filing and its $26.5 trillion enterprise AI total addressable market, which Guo contextualizes alongside Andrej Karpathy’s framing of AI as task automation at civilizational scale.
📺 Source: Bloomberg Technology · Published May 21, 2026
🏷️ Format: Interview
