{"version":"1.0","type":"agent_native_article","locale":"en","slug":"why-ai-analyses-past-venture-capital-bets-future-mpvxmepn","title":"Why AI Analyses the Past Well but Venture Capital Bets on the Future","primary_category":"startups","author":{"name":"Tomás Rivera","slug":"tomas-rivera"},"published_at":"2026-06-02T00:03:00.415Z","total_votes":88,"comment_count":0,"has_map":true,"urls":{"human":"https://sustainabl.net/en/articulo/why-ai-analyses-past-venture-capital-bets-future-mpvxmepn","agent":"https://sustainabl.net/agent-native/en/articulo/why-ai-analyses-past-venture-capital-bets-future-mpvxmepn"},"summary":{"one_line":"AI tools are structurally biased toward historical patterns, which makes them useful for due diligence but dangerous as gatekeepers in venture capital, where the highest returns come from bets that have no precedent.","core_question":"Can AI-driven investment analysis coexist with the fundamental VC mandate of identifying discontinuities that historical data cannot predict?","main_thesis":"Language models are pattern-recognition engines trained on the past; venture capital's edge comes from identifying futures that break from past patterns. Firms that let AI filter deal flow will systematically exclude the highest-upside opportunities, creating a structural blind spot that deepens as more firms adopt the same tools."},"content_markdown":"## Why AI Analyzes the Past Well but Venture Capital Bets on the Future\n\nThree quarters of venture capital firms already use artificial intelligence to evaluate investment opportunities. That figure alone sounds like inevitable modernization. But there is a structural tension that this percentage does not capture: language models are extraordinarily good at doing exactly what venture capital cannot afford to do too often, which is looking backward.\n\nVenture capital is, in its most basic mechanics, a bet on discontinuities. Not on markets that expand in a predictable manner, but on moments when a technology or a behavior breaks away from what previous data suggested was probable. Introducing tools trained on historical patterns into that process is useful until it stops being so, and the boundary between both states is narrower than most firms are acknowledging out loud.\n\n## The Most Sophisticated Confirmation Bias That Has Ever Existed\n\nLarge language models generate responses by identifying patterns in massive text corpora. This makes them extraordinarily capable for analytical tasks with well-defined contours: mapping competitors, identifying regulatory obstacles, summarizing technical literature, flagging risks in a known market. What they cannot do, by construction, is recognize the moment when those contours are about to redraw the entire map.\n\nThe history of venture capital is filled with examples where the correct analysis of the present was the very reason the future was missed. When Airbnb raised its first rounds in 2008, the thesis that strangers would pay to sleep in other people's homes was not just counterintuitive; it was directly inconsistent with the data available at the time about consumer behavior. The sentiment analysis of that era pointed in the opposite direction. The same was true in the early stages of the social web: the dominant surveys from the early 2000s showed that the main barrier to internet usage was fear of privacy. Facebook was built, in part, by ignoring that reading.\n\nA well-calibrated system would have flagged both proposals as high risk. And it would have been right, from the perspective of the past. The problem is not that the analysis was incorrect; it is that it was the wrong analysis for that specific decision.\n\nThis is where the bias becomes difficult to detect within firms that have adopted AI as a standard part of the due diligence process. It does not manifest as an obvious error. It manifests as a series of very well-documented analyses that systematically favor bets that have precedent and disfavor those that do not. In the short term, that produces more orderly portfolios. In the long term, it produces portfolios that do not generate the returns that justify the asset class.\n\n## What the Flow of Capital into AI in 2025 Reveals\n\nThe concentration of global venture capital in 2025 illustrates that pattern with precision. Global venture capital funding reached approximately **141 billion dollars** in the fourth quarter, an increase of 12% compared to the previous quarter, making 2025 the most active year since 2021. Artificial intelligence represented **more than 25% of global venture capital** that year, up from 15% in 2024 and 7% in 2023. In the enterprise segment, spending on generative AI grew from **11.5 billion dollars in 2024 to 37 billion in 2025**, according to data from Menlo Ventures.\n\nThose numbers describe an industry that, in part, is betting on the future with genuine conviction. But they also describe an industry that, in part, is following the most legible pattern available. AI is today the sector with the most recent historical validation, with the most citations in research papers, with the highest volume of news flow. It is, in practical terms, the market on which an AI tool can most easily produce analysis. The result is a feedback loop that concentrates capital where the signal is clearest, which is exactly the type of concentration that produces the highest median returns but not necessarily the returns in the upper-right tail.\n\nThe distribution within the AI segment also deserves attention. Menlo Ventures breaks down 2025 spending between **19 billion in the application layer** and **18 billion in infrastructure**. Within applications, horizontal tools captured 8.4 billion, departmental solutions 7.3 billion, and specialized verticals 3.5 billion. That level of granularity suggests that the bet is no longer about whether AI matters as a category, but about which layer of the value chain will capture sustainable margins. That is a much more refined question, and it is precisely the type of question where well-executed analysis, with or without AI, can provide differential value.\n\nWhat historical analysis cannot resolve is identifying which categories that do not appear in any dataset today will capture the next wave. Modular nuclear energy is the clearest example at this moment.\n\n## When a History of Failures Conceals a Real Discontinuity\n\nThe records on nuclear energy are full of warnings. Three Mile Island, Chernobyl, Fukushima. Decades of failed commercialization attempts. Construction timelines that stretched from years to decades. Structural cost overruns. An analytical system trained on that corpus would produce, in a completely reasonable manner, a high-risk assessment for any startup proposing small modular reactors as an energy solution.\n\nThe problem is that small modular reactors are technically and economically distinct from the large-scale nuclear plants that generated that track record. They are designed for serial manufacturing and standardization, not for bespoke construction at each site. And the demand context has changed in a structural way: AI data centers require volumes of continuous and predictable energy that intermittent sources cannot satisfy in an economically efficient manner at scale. Companies such as Microsoft, Google, and Amazon have already begun signing agreements and making investments linked to nuclear generation, which indicates that the demand signal exists and is being formalized in contracts, not merely in statements of intent.\n\nA model trained on the nuclear past would likely see accumulated risk. An analyst who understands what has changed in the economics of energy demand can see a technology reaching the market at the very moment the market finally needs it. The difference between those two readings is not optimism versus pessimism. It is the capacity to identify when an external variable has reconfigured the space of possibilities for a technology that was previously unviable.\n\nThat capacity cannot be delegated to a system that does not have access to information about what has not yet occurred.\n\n## Imagination Is Not an Analytical Luxury — It Is the Variable the Model Cannot Import\n\nWhat venture capital has historically purchased is not analysis of the existing market. It has purchased the capacity to imagine markets that do not yet exist and to identify the teams that can create them. That capacity has components that are not delegable to systems that analyze historical patterns: the reading of weak signals, the ability to recognize when an emerging behavior is about to become mainstream, the distinction between a market that failed due to poor execution and one that failed because the timing was wrong.\n\nNone of this means that firms should reduce their use of AI in the investment process. Current tools are genuinely valuable for accelerating due diligence on known markets, stress-testing business model assumptions, and structuring competitive analysis. Used well, they make more rigorous the work that analysts were already doing.\n\nThe risk is not in using AI. It is in building processes where the absence of historical precedent automatically becomes a signal for rejection. That institutional design expels from the portfolio exactly the opportunities that venture capital should be best positioned to capture.\n\nThe capital that flows to where models can already measure well is competing against every firm that uses the same models. The capital that flows to where models still cannot measure well is competing against far less. That asymmetry does not disappear because the tool becomes more sophisticated. It deepens.","article_map":{"title":"Why AI Analyses the Past Well but Venture Capital Bets on the Future","entities":[{"name":"Airbnb","type":"company","role_in_article":"Historical example of a bet that contradicted available consumer behavior data at the time of its early funding rounds."},{"name":"Facebook","type":"company","role_in_article":"Historical example of a company built by ignoring dominant survey data showing privacy fears as the main barrier to internet adoption."},{"name":"Microsoft","type":"company","role_in_article":"Named as an early signatory of agreements linked to nuclear energy generation, validating structural demand."},{"name":"Google","type":"company","role_in_article":"Named as an early signatory of agreements linked to nuclear energy generation, validating structural demand."},{"name":"Amazon","type":"company","role_in_article":"Named as an early signatory of agreements linked to nuclear energy generation, validating structural demand."},{"name":"Menlo Ventures","type":"institution","role_in_article":"Source of granular data on enterprise generative AI spending breakdown by layer in 2025."},{"name":"Small Modular Reactors","type":"technology","role_in_article":"Primary case study illustrating how historical failure records can obscure genuine technological discontinuities."},{"name":"Large Language Models","type":"technology","role_in_article":"The AI technology whose architectural constraints are analyzed as a structural mismatch with VC decision-making."},{"name":"Venture Capital","type":"market","role_in_article":"The investment context in which AI adoption is analyzed and critiqued."},{"name":"Tomás Rivera","type":"person","role_in_article":"Author of the article."}],"tradeoffs":["AI in VC: faster, more rigorous due diligence on known markets vs. systematic exclusion of high-upside opportunities without precedent.","Capital concentration in AI sector: higher median returns vs. compressed upper-tail returns that define VC performance.","Institutional AI adoption: reduced analyst error on legible bets vs. increased competition with every firm using the same tools on the same opportunities.","Historical pattern recognition: accurate risk assessment from past data vs. inability to detect when external variables have reconfigured the viability of a technology.","Portfolio orderliness: well-documented, defensible investment decisions vs. long-term underperformance relative to the asset class mandate."],"key_claims":[{"claim":"Three quarters of venture capital firms already use AI to evaluate investment opportunities.","confidence":"high","support_type":"reported_fact"},{"claim":"Global VC funding reached approximately 141 billion dollars in Q4 2025, a 12% increase quarter-over-quarter, making 2025 the most active year since 2021.","confidence":"high","support_type":"reported_fact"},{"claim":"AI represented more than 25% of global VC in 2025, up from 15% in 2024 and 7% in 2023.","confidence":"high","support_type":"reported_fact"},{"claim":"Enterprise spending on generative AI grew from 11.5 billion dollars in 2024 to 37 billion in 2025.","confidence":"high","support_type":"reported_fact"},{"claim":"Menlo Ventures breaks 2025 AI spending into 19 billion at the application layer and 18 billion in infrastructure.","confidence":"high","support_type":"reported_fact"},{"claim":"Microsoft, Google, and Amazon have begun signing agreements and investments linked to nuclear generation.","confidence":"high","support_type":"reported_fact"},{"claim":"AI-assisted due diligence systematically disfavors opportunities without historical precedent, producing structurally conservative portfolios.","confidence":"medium","support_type":"inference"},{"claim":"The concentration of VC in AI is partly driven by the fact that AI tools can most easily analyze AI markets, creating a feedback loop.","confidence":"medium","support_type":"inference"}],"main_thesis":"Language models are pattern-recognition engines trained on the past; venture capital's edge comes from identifying futures that break from past patterns. Firms that let AI filter deal flow will systematically exclude the highest-upside opportunities, creating a structural blind spot that deepens as more firms adopt the same tools.","core_question":"Can AI-driven investment analysis coexist with the fundamental VC mandate of identifying discontinuities that historical data cannot predict?","core_tensions":["AI's core strength (pattern recognition on historical data) is structurally misaligned with VC's core mandate (identifying discontinuities that break from historical patterns).","Using AI to reduce investment risk may increase portfolio-level risk by excluding the high-variance bets that generate VC-class returns.","The more sophisticated AI tools become, the more firms adopt them, and the more the asymmetric opportunity shifts to where those tools cannot yet operate.","Rigorous analysis and imaginative judgment are both necessary in VC, but optimizing for the former through AI may crowd out the latter institutionally."],"open_questions":["At what point in the investment process does AI add value without introducing structural bias — and where exactly does that boundary lie for each firm?","How should VC firms design institutional processes that use AI for acceleration without allowing it to function as a precedent-based filter?","Which current sectors, like nuclear energy in this analysis, have historical failure records that conceal genuine discontinuities driven by changed external conditions?","Will the upper-tail return asymmetry between AI-legible and AI-illegible opportunities persist as models gain access to more diverse and forward-looking data sources?","How do LPs evaluate and price the risk that a VC firm's AI-assisted process is systematically excluding the highest-upside opportunities?"],"training_value":{"recommended_for":["Venture capital analysts and partners evaluating AI tool adoption in deal sourcing and due diligence.","Corporate strategy teams assessing AI integration in investment or innovation scouting functions.","AI product teams building tools for financial decision-making who need to understand structural limitations.","Founders in sectors with negative historical records seeking to frame their investment thesis against legacy data.","Business agents trained on investment reasoning, technology evaluation, and institutional decision-making design."],"when_this_article_is_useful":["When evaluating whether to integrate AI tools into investment or strategic decision-making processes.","When assessing opportunities in sectors with negative historical records but changed external conditions.","When designing due diligence workflows that need to balance analytical rigor with openness to discontinuous opportunities.","When reasoning about competitive differentiation in markets where most players use the same analytical infrastructure.","When advising on AI adoption strategy in contexts where the highest-value decisions are inherently forward-looking."],"what_a_business_agent_can_learn":["How to identify when an analytical tool's strengths are structurally misaligned with the decision type it is being applied to.","How to distinguish between a market that failed due to poor execution and one that failed due to wrong timing or wrong external conditions.","How to reason about feedback loops between analytical infrastructure and capital allocation, and why they compress returns over time.","How to evaluate the asymmetric competitive value of operating in spaces where dominant analytical tools cannot yet produce clear signals.","How to design institutional processes that use AI as an accelerant without allowing it to function as a gatekeeping filter based on historical precedent."]},"argument_outline":[{"label":"1. The structural mismatch","point":"LLMs generate outputs by identifying patterns in historical corpora. Venture capital generates alpha by betting on discontinuities that contradict those patterns.","why_it_matters":"This is not a limitation that better models will fix; it is a design constraint. The mismatch is architectural, not a calibration problem."},{"label":"2. Confirmation bias at institutional scale","point":"AI-assisted due diligence systematically favors opportunities with historical precedent and flags those without it as high-risk, producing well-documented but structurally conservative portfolios.","why_it_matters":"The bias is invisible in the short term because it produces orderly portfolios. It becomes visible only when long-term returns fail to justify the asset class."},{"label":"3. The 2025 capital concentration as evidence","point":"AI represented over 25% of global VC in 2025, up from 7% in 2023. Capital is concentrating in the sector where AI tools can produce the clearest analysis.","why_it_matters":"This feedback loop maximizes median returns but compresses the upper-tail returns that define VC as an asset class."},{"label":"4. The nuclear energy case study","point":"Small modular reactors carry a historical record of failure, but are technically and economically distinct from legacy nuclear. AI trained on that record would flag them as high-risk at the exact moment structural demand from AI data centers makes them viable.","why_it_matters":"It illustrates how an external variable can reconfigure the viability of a previously failed technology in a way that historical analysis cannot detect."},{"label":"5. The asymmetry argument","point":"Capital flowing where models can measure well competes against every firm using the same models. Capital flowing where models cannot yet measure competes against far fewer.","why_it_matters":"This asymmetry does not shrink as AI improves; it deepens, because more firms adopt the same tools and converge on the same legible opportunities."},{"label":"6. The non-delegable VC function","point":"Reading weak signals, distinguishing bad timing from bad ideas, and imagining markets that do not yet exist are functions that cannot be delegated to systems without access to information about what has not yet occurred.","why_it_matters":"Firms that delegate these functions to AI are not augmenting their judgment; they are replacing it with a structurally backward-looking filter."}],"one_line_summary":"AI tools are structurally biased toward historical patterns, which makes them useful for due diligence but dangerous as gatekeepers in venture capital, where the highest returns come from bets that have no precedent.","related_articles":[{"reason":"Directly complementary: examines the blind spots in corporate AI adoption that official reports do not capture, paralleling this article's argument that AI-assisted VC processes create invisible structural biases.","article_id":13274},{"reason":"Addresses the human-in-the-loop requirement for enterprise AI, supporting the article's argument that non-delegable human judgment remains essential even as AI accelerates analytical work.","article_id":13161},{"reason":"Case study of a deep-tech startup (programmable biology) in a sector where historical data is sparse and the investment thesis requires imagining markets that do not yet exist — the exact scenario this article describes as AI's blind spot.","article_id":13106},{"reason":"Covers a hardware startup solving a problem for AI data centers, illustrating the structural demand shift that this article uses to argue for nuclear energy's renewed viability.","article_id":13217}],"business_patterns":["Feedback loop between analytical tools and capital allocation: tools that analyze a sector well attract more capital to that sector, which generates more data, which makes the tools better at analyzing it, reinforcing concentration.","Technology timing mismatch: technologies that failed in prior cycles due to wrong market conditions are systematically undervalued when conditions change, creating asymmetric opportunities for investors who can distinguish timing failure from fundamental failure.","Institutional mimicry risk: when most firms adopt the same analytical infrastructure, differentiated returns require differentiated inputs, not better execution of the same process.","Legibility premium: markets that are easy to analyze attract more capital than their fundamentals justify; markets that are hard to analyze are systematically underpriced."],"business_decisions":["Whether to use AI as a filter in deal sourcing versus only as an accelerant in due diligence on pre-selected opportunities.","How to design institutional processes so that absence of historical precedent does not automatically trigger rejection.","Whether to allocate capital to sectors where AI tools produce clear analysis versus sectors where they cannot yet measure well.","How to evaluate technologies with negative historical records when external demand conditions have structurally changed.","Whether to treat AI-generated risk flags as inputs to human judgment or as decision gates."]}}