AI Is Great At Analyzing The Past. Venture Capital Bets On The Future

Roughly three-quarters of venture capital firms now use AI in deal evaluation, but the article argues this can bias investors toward incremental ideas and away from true outliers. It highlights AI's strengths in diligence while warning that pattern-based models may miss breakthrough startups like Airbnb or emerging opportunities such as small modular nuclear reactors. The piece is a strategic commentary on VC decision-making rather than a market-moving event.

Analysis

The second-order issue is not that AI is bad at diligence; it is that it systematically compresses the distribution of venture outcomes toward the median. If LPs and partners lean on the same model-generated heuristics, they will converge on the same “obvious” sectors, which raises entry prices in familiar adjacency bets and lowers the hit rate on true category creators. That should favor founders with unusual technical or behavioral wedges, but only if investors retain a human override to tolerate ambiguity longer than a model would recommend.

The market implication is a misallocation of capital, not just a screening error. In private markets, when a tool makes consensus better at identifying incremental winners, capital floods to the safest growth paths and starves the weird, policy-dependent, or infrastructure-heavy opportunities that need patience. That creates a wider spread between “AI-legible” software names and frontier physical-world bets, where outcomes may be farther out but option value is materially higher. The energy angle is the clearest example: anything that shortens the power lead time for data-center-driven load growth can become a beneficiary even if the historical precedent is poor.

For the public comps named in the piece, the likely near-term winner is not the most obvious “AI winner” but the infrastructure layer that makes the new demand possible. Over a 12-36 month horizon, utilities, grid equipment, uranium supply chain, and modular nuclear exposure should see valuation support if AI capex keeps translating into firm power demand. By contrast, consumer platform names can suffer a subtle multiple headwind if investors start discounting “next big thing” narratives more aggressively because the funnel is biased toward incrementalism.

The contrarian read is that the article underestimates how quickly humans can use AI against itself. The best firms will use models to screen the obvious, then deliberately hunt for model blind spots — regulatory arbitrage, behavior change, and capital intensity inflections — exactly where alpha lives. So the real trade is not anti-AI; it is to own the enablement layer and finance the non-obvious outliers before consensus data catches up.

Key Decisions for Investors

• Overweight NVDA / MSFT / AMZN vs UBER and ABNB for the next 3-6 months: if AI capital spending remains strong, the infrastructure beneficiaries should re-rate faster than consumer platform names; use a long NVDA, short ABNB relative-value pair to express the view that capital will favor visible monetization over narrative optionality.
• Initiate a 12-24 month basket long in energy-enablers tied to AI power demand: uranium, grid equipment, and modular nuclear exposure. The risk/reward is asymmetric because the downside is time-to-revenue, while the upside is a step-function re-rating if data-center load forecasts force faster procurement cycles.
• Avoid shorting venture-facing software outright; instead, fade overcrowded private-market comparables only on financing windows. The better trade is to reduce exposure to late-stage consumer and horizontal SaaS that screens well to models but lacks a defensible data or distribution moat.
• Buy long-dated call spreads on a public modular nuclear or uranium proxy if available, with 12-18 month tenor. The thesis needs regulatory and execution patience, but the convexity is attractive if AI-driven power demand shifts the market from “story” to “capacity shortage.”
• If ABNB underperforms on multiple compression, pair it against MSFT or NVDA rather than broad market shorts; the relative trade isolates the article’s key point that consensus tools favor known-pattern businesses while underpricing true platform shifts.