Deep Investment Research15 min read
AI Took 86% of Venture Dollars—Now Venture Capital Looks Like an Illiquid Concentrated Index
U.S. venture investment broke records in the first half of 2026, but deal counts did not share the boom. PitchBook says AI absorbed 86% of capital and rounds of $100 million or more captured 87.5%. The Information describes the pressure on Benchmark and its peers: accept higher prices and smaller stakes or miss the handful of companies that may swallow the industry’s returns. What looks like the venture power law may be mutating into something more dangerous—multiple funds, corporate investors and limited partners repeatedly owning the same private giants while calling the result diversification.
One Record, Two Completely Different Markets
Read only the headlines and 2026 looks like the greatest venture boom ever. PitchBook-NVCA counted $412.7 billion in U.S. venture deals during the first half, nearly 30% more than all of 2025. AI companies received $355.9 billion, or 86%. Crunchbase, using U.S. and Canadian coverage and different deal classifications, reported $392 billion. The totals differ; the message does not. Capital exploded. Breadth did not.
PitchBook says rounds of at least $100 million absorbed 87.5% of all dollars. Crunchbase says second-quarter North American investment reached $137.2 billion, the second-highest quarterly total ever, even as deal count remained far below recent peaks. “Risk appetite returned” is the wrong diagnosis. Capital became willing to take unprecedented risk on a very short list while remaining skeptical of the median startup.
This is not a rising tide. It is a taller needle. Companies on the frontier-AI list can absorb billions in weeks. Those outside it still face fewer meetings, slower decisions and stricter terms. An average that combines those worlds no longer describes either one.
| 2026 capital signal | Latest figure | What it really says |
|---|---|---|
| U.S. H1 venture | $412.7B | Above full-year 2025, driven by megadeals |
| AI share | 86% | Portfolios load on one technology factor |
| $100M+ round share | 87.5% | Dollar volume separates from market breadth |
| North America Q2 seed | $4.9B | Down 15% QoQ and 27% YoY |
Benchmark’s Choice: Pay More, Own Less—or Miss the Winner
The Information recently framed Benchmark’s AI pressure test around high prices, smaller stakes and a poached star investor. That captures the central problem for elite venture firms. The classic early-stage model used a low entry price and meaningful ownership to earn tens or hundreds of times invested capital on a rare winner. When a coveted company reaches a multibillion-dollar valuation before its product economics mature, the same check buys far less of it.
Consider a simplified case. A fund invests $50 million at a $1 billion valuation and owns roughly 5%. A $10 billion exit produces about a tenfold gross return before dilution. Competition pushes the entry price to $5 billion and the same exit produces about two times. The company is still a winner. The investment is merely adequate. To restore venture-level returns, the firm must write a larger check, concentrate the fund, or assume an end value that would once have looked implausible.
Fear of missing AI therefore raises prices, and those prices require a still larger terminal story. Each financing validates the last one until public markets or cash flow decline to continue the relay. Venture capital shifts from discovering mispriced uncertainty to auctioning certainty that everybody already recognizes. The label has not changed. The behavior increasingly resembles momentum investing.
The Power Law Is Not Broken. It Is Becoming More Brutal
Venture returns have never followed a bell curve. A small number of successes cover many failures. What changed is the slope. Commonfund analyzed PitchBook exits and found that the top 1% of venture-backed companies generated 80% of exit value since 2023. Remove SpaceX’s $1.77 trillion IPO and the figure is still 45%. It was 17% in 2005–2010 and 34% in 2017–2022.
Funding is concentrating in parallel. Russell Investments found that OpenAI, Anthropic, xAI, Waymo and Nscale represented 78% of first-quarter unicorn deal value; the top ten represented 81%. A fund can own 30 AI names and remain exposed to the same compute price, model-progress curve, enterprise adoption cycle and exit window. Counting companies is not the same as counting independent risks.
The power law is usually invoked to justify boldness: if the top 1% creates most value, you must own the winner. That leaves out price. A commercial winner becomes a fund winner only when the investor owns enough, enters with enough upside and survives dilution. Buying a celebrity company near its terminal value separates business success from investment success.
The Hidden Risk: LPs Think They Diversified but Bought the Same AI Factor
Pensions, endowments and family offices allocate across multiple venture managers to diversify manager risk. But when Sequoia, Andreessen Horowitz, Thrive, corporate venture arms, growth funds and secondary vehicles repeatedly own the same frontier labs, diversity of fund names stops meaning diversity of underlying assets. An LP can own one company ten times through ten partnerships.
A public index at least publishes a daily price, weight and liquidity. A private portfolio does not. The latest financing price may be used to mark an entire capitalization table even though different shares carry different liquidation preferences, anti-dilution rights and strategic commitments. Common stock, employee options and protected preferred shares collapse into one headline “valuation.” Lower reported volatility is not lower economic risk. It is slower price discovery.
That leads to this article’s central judgment: the leading venture portfolio of 2026 is becoming an AI index without a live price, common disclosure standard or redemption mechanism. It is harder to inspect than the Nasdaq and harder to value than a single stock, yet it is still sold as alternative-asset diversification. The danger is not only that one company fails. It is that one assumption reprices every holding together.
How Four Master Investors Would Pull the Story Apart
Howard Marks would watch the pendulum. When consensus moves from “Can AI make money?” to “The greatest risk is not owning AI,” risk has not disappeared. It has migrated from business risk to price risk. Second-level thinking asks what remains unpriced after every sophisticated investor agrees on the winner. Valuefocus’s AI bubble framework separates being right about a technology from being right about a return.
Seth Klarman would begin with the downside. Preferred protections can reduce loss severity without creating a good fund return. At a $965 billion valuation, a company may need to multiply again to produce venture-style gains. Margin of safety comes from price, rights, capital requirements and exit depth—not from the company being admired. The margin of safety calculator makes the optimistic story survive a valuation haircut.
Buffett and Munger would test the circle of competence and incentives. A software investor may not be able to forecast model commoditization, GPU depreciation, electricity constraints, regulation and enterprise budgets together. Fee incentives add another layer: a larger fund needs larger checks, and larger checks naturally migrate toward higher-valued companies. Munger’s inversion asks how to manufacture a portfolio that looks diversified but is highly correlated. The current market supplies nearly every ingredient.
Stop Watching Dollars Raised. Watch Ownership Efficiency
Five measures matter more than the next funding headline: fully diluted ownership acquired per dollar; revenue growth relative to valuation growth; capital consumed for each new dollar of annual recurring revenue; dilution required by the next round; and the exit market’s capacity to absorb trillion-dollar supply. A financing total is not proof of customer demand. It may be evidence that the business requires extraordinary fuel.
Seed data is the second warning. Crunchbase counted about $4.9 billion of North American seed and angel investment in Q2, down 15% from Q1 and 27% year over year. Yet at least five companies raised seed rounds of $100 million or more. Capital did not leave “early stage.” It redefined early stage around a handful of celebrity founders, compute contracts and enormous narratives. The future innovation pipeline may become narrower, not wider.
TikHub’s live X results compress the debate into two slogans: “AI takes 77%–86% of global venture and winners take all,” or “this bubble will break the market.” A value investor does not need either slogan. Ask what price was paid, what is actually owned, what exit is required and whether the return survives if the technology succeeds but the valuation multiple falls by half. AI may be the century’s most important technology. That is precisely why no price should automatically be called cheap.
Recent posts
%22%2F%3E%3Cpath%20d%3D%22M0%20610%20370%20370l300%20160%20400-330%20530%20360%22%20fill%3D%22none%22%20stroke%3D%22%23c99a62%22%20stroke-width%3D%2218%22%20opacity%3D%22.65%22%2F%3E%3Ccircle%20cx%3D%221070%22%20cy%3D%22200%22%20r%3D%2246%22%20fill%3D%22%23c99a62%22%2F%3E%3C%2Fsvg%3E)
%20-1.jpg?width=1600)
Frequently asked questions
What share of venture funding went to AI in 2026?
PitchBook-NVCA says AI represented 86% of U.S. venture dollars in H1 2026. Crunchbase, using different regional and deal classifications, estimated about 80% in Q2. Both indicate historic concentration.
Why can an AI company succeed while its VC investors earn weak returns?
Returns depend on entry price, ownership, dilution and exit value. If investors enter near the eventual value, strong company growth can still produce an inadequate multiple.
What is the venture capital power law?
A small number of outliers produce most portfolio returns. Since 2023, the top 1% of companies have generated roughly 80% of exit value, or 45% excluding SpaceX.
How should investors evaluate private AI valuations?
Track fully diluted ownership, revenue growth versus valuation growth, capital efficiency, liquidation rights, future dilution and exit capacity. Stress-test returns under lower valuation multiples.
A technology winner is not automatically an investment winner. Use Valuefocus to record entry price, implied growth, dilution and exit assumptions—and keep a margin of safety when consensus is strongest.
Download Valuefocus