A new benchmark from the University of Oxford, VC-Bench, found that AI models like DeepSeek Chat can predict founder success (defined as a >$500M exit or raise) with 80% accuracy based on anonymized profiles. This starkly contrasts with the 23% accuracy of human VCs, questioning the notion that venture investing is an inimitable human art.
Related Insights
Redpoint Ventures' Erica Brescia describes a shift in their investment thesis for the AI era. They are now more likely to back young, "high-velocity" founders who "run through walls to win" over those with traditional domain expertise. Sheer speed, storytelling, and determination are becoming more critical selection criteria.
Sequoia quantifies its search for 'outlier founders' in statistical terms. An exceptional founder is three standard deviations above the mean in a key trait, but a true outlier is four. This statistical lens explains their high bar, reviewing around 1,000 companies for every single investment.
Low-cost AI tools create a new paradigm for entrepreneurship. Instead of the traditional "supervised learning" model where VCs provide a playbook, we see a "reinforcement learning" approach. Countless solo founders act as "agents," rapidly testing ideas without capital, allowing the market to reward what works and disrupting the VC value proposition.
Founders are consistently and universally wrong about their financial projections, particularly cash runway. AI tools can provide an objective, data-driven forecast based on trailing growth, correcting for inherent founder optimism and preventing critical miscalculations.
OpenAI's new GDPVal framework evaluates AI on real-world knowledge work. It found frontier models produce work rated equal to or better than human experts nearly 50% of the time, while being 100 times faster and cheaper. This provides a direct measure of impending economic transformation.
The venture capital return model has shifted so dramatically that even some multi-billion-dollar exits are insufficient. This forces VCs to screen for 'immortal' founders capable of building $10B+ companies from inception, making traditionally solid businesses run by 'mortal founders' increasingly uninvestable by top funds.
AI-powered VC introduction platforms are not just connectors; they are stringent gatekeepers reflecting the high bar of the current market. By assigning a "grade" and only facilitating introductions for high-scoring decks, these systems programmatically enforce VC standards at scale.
The CEO of Numeral notes that in the current fundraising climate, startups must heavily feature AI in their pitch to secure investor meetings. Furthermore, landing a major AI lab as a customer has become a key signal for VCs, leading to valuation multiples as high as 100-200x revenue for some companies.
OpenAI's new GDP-val benchmark evaluates models on complex, real-world knowledge work tasks, not abstract IQ tests. This pivot signifies that the true measure of AI progress is now its ability to perform economically valuable human jobs, making performance metrics directly comparable to professional output.
AI startups' explosive growth ($1M to $100M ARR in 2 years) will make venture's power law even more extreme. LPs may need a new evaluation model, underwriting VCs across "bundles of three funds" where they expect two modest performers (e.g., 1.5x) and one massive outlier (10x) to drive overall returns.
