Conference Presentation, Panel, Fireside Chat, Keynote
Sam Altman, Arthur Mensch and more discuss:Which Startups Are Threatened vs Enabled by OpenAI?|E1156
20VC with Harry StebbingsSam Altman, Arthur Mensch, Brad Lightcap, Des Traynor, Tom Hulme, Tomasz Tunguz, Sarah Tavel, Harry Stebbings, Emad Mostaque, Tom Blomfield, Miles Grimshaw
- Industry consolidation is expected to result in a small number of providers dominating big-scale model training within three to five years, with Meta predicted to hold 14% of the world's H100s by year-end following an additional $100 billion investment and a total of 350,000 H100s, while Nvidia, Google, Microsoft, OpenAI, and Apple are identified as the primary trainers.
- Long-term differentiation will shift from base models to highly personalized models incorporating whole life context, yet technology is commoditizing rapidly with competitors expected to replicate GPU power stations within a few months, causing asset depreciation for foundation models to occur in minutes rather than years.
- Financial returns on foundation model investments are projected to be difficult unless the firm maintains momentum, with cloud providers eventually becoming the cash cow that buys model companies like Google, Amazon, or Microsoft, potentially leading to Amazon acquiring Anthropic, while investors in OpenAI face a "monstrous" fund with inherent dilution.
- The application layer is predicted to drive most enduring value and represent a lower-risk opportunity due to greater diversity of needs, with consumer-facing products expected to generate around 50% of revenue but lacking stickiness, whereas startups building generic co-pilots are unlikely to be defensible compared to incumbents leveraging existing distribution and data.
- Business models are expected to shift from per-seat software pricing to selling full work products or outcomes, while AI will act as a sustaining innovation to lower costs rather than causing creative destruction, with every computer user anticipated to have an AI co-pilot assistant within two to three years.
- Market dynamics will favor those who assume models will continue improving at OpenAI's pace, with 95% of the world advised to bet on this trajectory, whereas companies assuming static intelligence or relying on weekend hackathon builds face being steamrolled by incumbents or OpenAI's "good enough" minimum viable standard.
- Price per intelligence unit will decrease due to compression and efficiency improvements, yet the opportunity for startups lies in being orthogonal to incumbents rather than competing in the co-pilot space, as OpenAI is not expected to deeply integrate into specific sectors like wealth management or banking.