Venture capital firms investing in AI have increasingly diverged into two distinct strategic approaches that require different underlying skills and risk tolerance, and firms trying to pursue both simultaneously often find themselves under-resourced for at least one of them.
The infrastructure-scale strategy
One approach concentrates capital in a small number of very large, capital-intensive bets on foundational infrastructure-model development, compute, core platform layers-where the required check sizes are large, the competitive field is narrowing to a small number of well-capitalized leaders, and the firm's strategy depends on picking correctly among a small set of likely long-term winners rather than diversifying broadly.
The application-niche strategy
The other approach spreads capital across a larger number of smaller bets on application-layer startups targeting specific, defensible workflow niches, where the required check sizes are smaller, the competitive field is broader and more fragmented, and the firm's strategy depends on portfolio diversification and the recognition that most individual bets will underperform while a smaller number of larger successes power overall fund returns.
Why these require genuinely different skills
Evaluating infrastructure-scale bets well requires deep technical understanding of compute economics and a high tolerance for concentrated risk on a small number of large positions. Evaluating application-niche bets well requires closer, more numerous engagement with individual founders and markets, and comfort with a portfolio construction model built around a higher failure rate across more individual investments. A firm's investment team, network, and typical check size are generally better suited to one approach than the other.
What this means for founders raising capital
Founders are increasingly well served by understanding which strategic camp a prospective investor actually falls into before pitching, since a firm oriented toward large infrastructure bets is likely to evaluate a narrower application-layer startup through an ill-fitting lens, and vice versa. Matching the pitch to the investor's actual strategy, rather than treating all AI-focused venture capital as a single undifferentiated pool, improves the odds of a productive conversation.