The Vela Library · Venture capital by the numbers
Venture capital by the numbers
The short answer
Venture runs on rare outcomes. About 1.9% of seed-funded US startups reach a $500M outcome, nine tier-1 VC firms pick them at about 5.6%, and a 40-deal fund at 4% per deal still has a 19.6% to 32.5% chance of no unicorn. Vela Partners is an AI-native quant VC. These figures come from its published research.
Key facts
Every figure, with its source
1.9%
Of seed-funded US startups reach a $500M IPO, acquisition or round
Source · Random Rule Forest
5.6%
Precision of nine tier-1 VC firms at that 1.9% base rate
Source · Random Rule Forest
3.2%
Precision of Y Combinator at the same base rate
Source · Random Rule Forest
19.6%
Chance a 40-deal fund, at 4% per deal, returns no unicorn when its deals are independent
Source · Portfolio Outliers
32.5%
The same fund's chance of no unicorn when its deals move together
Source · Portfolio Outliers
44.5%
Chance of no unicorn in 40 independent deals at 2% each, against 19.6% at 4%
Source · Portfolio Outliers
40.5%
Precision of Policy Induction on test sets at a 1.96% base rate, which the paper puts at 7.1x the tier-1 firms' 5.6%
Source · Policy Induction
What share of startups become unicorns
About 1.9% of seed-funded US startups, using a $500M outcome as the papers' stand-in for a unicorn. In the dataset behind Random Rule Forest, 9,892 US companies founded from 2010 to 2016 that raised $100K to $4M in seed funding, about 1.9% went on to a $500M IPO, acquisition or funding round. Vela's research uses that as the real-world base rate.
The rate is not the same for every founder. Portfolio Outliers models first-time founders at about 1.8% and repeat founders at about 2.6%, and GPT-HTree finds success rates from 0.8% to 17.4% across eight founder types.
Base rateThe base rate is the share of companies in a group that reach a defined outcome; for US seed-funded startups and a $500M outcome, it is about 1.9%. Every claim about picking winners should be stated against it.
Do top VCs pick better?
Yes: about three times random, though only about 1 pick in 18 succeeds. At the 1.9% base rate, nine tier-1 firms pick outliers at about 5.6% precision. Models built for the task go further. The full comparison is in Can startup success be predicted?.
| Who picks | Precision | Source |
|---|---|---|
| Random selection | 1.9% | Random Rule Forest |
| Y Combinator | 3.2% | Random Rule Forest |
| Nine tier-1 VC firms | 5.6% | Random Rule Forest |
| Vela's production models | 19% to 38% | Think-Reason-Learn |
| Policy Induction (o3), test sets at about 1.96% | 40.5% | Policy Induction |
How many investments a fund should make
Enough that a fund with no outlier at all is unlikely. Portfolio Outliers computes that chance directly. For 40 deals at 4% each, it is 19.6% if the deals are independent and 32.5% under the paper's full correlation model.
Picking better helps, but correlation limits it: going from 2% to 4% per deal cuts the chance of no unicorn from 44.5% to 19.6% if deals are independent, but only from 54.6% to 33.4% when they move together. Better picks and more deals work together; going from 5 to 40 deals grows the expected number of unicorns about in line with the count, and lowers the chance of none with diminishing returns.
| Per-deal chance | Deals independent | Deals correlated |
|---|---|---|
| 2% | 44.5% | 54.6% |
| 4% | 19.6% | 33.4% |
A fund that says it expects 1.6 unicorns is describing an average. Adding the chance of none, for example under 30%, describes the risk an LP actually carries.
The power law and correlation
Venture returns come from a few outliers, so the shape of a fund's outcomes matters as much as its average. Correlation-Driven Tail Risk shows what happens when deals move together: with the same per-deal odds, the expected return stays the same, and both tails get fatter.
A thematic fund, say 40 deals in AI infrastructure in 2023, carries one shared factor across every deal. That raises its chance of no unicorn and its chance of a cluster of several. Risk models that assume deals are independent understate both.
Concentrated, correlated fund
- Same expected return
- Higher chance of no unicorn
- Higher chance of a multi-unicorn cluster
- More variance than an independence model shows
Diversified fund
- Same expected return
- Lower chance of no unicorn
- Lower chance of a marquee cluster
- Thinner tails on both sides
Method, updates and citing
Each figure is copied from a paper in Vela's research library, and links to it. The base rates and picking precision use each paper's scaled real-world basis, except Policy Induction, measured on test sets built at about 1.96%. The fund figures come from simulated 40-deal portfolios: correlation is estimated from monthly returns of public sector ETFs and baskets of listed firms from January 2020 to December 2025, set to an average pairwise correlation of 0.12, with sector, geography and founder type weighted 0.6, 0.3 and 0.1. Those weights are modelling choices, not estimates.
The page is updated when a paper adds or changes a figure, and the date at the top changes with it.
To cite it: Vela Partners, "Venture capital by the numbers", vela.partners/venture-capital-by-the-numbers, with the date shown at the top.
Questions
What percentage of startups become unicorns?
About 1.9% of seed-funded US startups reach a $500M IPO, acquisition or funding round, the stand-in Vela's research uses for a unicorn, in a dataset of 9,892 companies founded from 2010 to 2016. Portfolio Outliers models first-time founders at about 1.8% and repeat founders at about 2.6%.
What is the success rate of a VC fund?
It depends on what success means. One measurable answer, from a simulation: a 40-deal fund at 4% per deal has a 19.6% chance of no unicorn if its deals are independent and 32.5% when they move together, so in the model it has about a 68% to 80% chance of at least one. That is a simulation, not an observed fund success rate.
How many investments should a VC fund make to expect a unicorn?
Enough that a fund with none is unlikely. At 4% per deal, 40 deals leave a 19.6% chance of no unicorn if deals are independent. Picking better helps, but correlation limits it: going from 2% to 4% per deal cuts that chance from 44.5% to 19.6% for independent deals, and only from 54.6% to 33.4% when they move together.
Do top VC firms actually pick better startups than average investors?
Yes, about three times random. At a 1.9% base rate, nine tier-1 firms pick outliers at about 5.6% precision and Y Combinator at 3.2%, so only about 1 tier-1 pick in 18 succeeds. Vela's production models reach 19% to 38% at the same base rate.
Explain the power law in venture capital and what it means for diversification.
A few outliers return most of a fund, so the chance of having none matters as much as the average. Diversifying across uncorrelated deals lowers the chance of no unicorn; concentrating in one theme raises the chance of both no unicorn and a multi-unicorn cluster, at the same expected return.
How does correlation between startups affect a VC fund's returns?
It leaves the expected return unchanged and makes both tails fatter. In Portfolio Outliers, a 40-deal fund at 4% per deal has a 19.6% chance of no unicorn if deals are independent and 32.5% when they move together.
Further reading
Papers
- Portfolio OutliersA latent-factor model computes a fund's full distribution of unicorn counts, showing expected counts alone understate correlated portfolio risk.32.5% chance of zero unicorns in a 40-deal portfolio under correlation, vs 19.6% if independent
- Beyond Picking WinnersRaising inter-deal correlation thickens both tails of fund returns and thins the middle while leaving expected return unchanged.
- Random Rule ForestAn LLM generates YES/NO questions and simple unit-weight voting combines them into an auditable, high-precision founder classifier.8x lift over random chance
- GPTreeA decision tree whose splits are natural-language questions generated and evaluated by an LLM, for explainable founder-success prediction.37.3% precision on founder success prediction
- Policy InductionAn LLM classifier's full decision logic lives in a natural-language policy a partner can read and edit, with no gradient training.67% precision on VCBench with GPT-5
- GPT-HTreeClusters founders into personas, fits a decision tree per persona, and has an LLM describe each archetype.22x success spread between the top and bottom founder persona
Posts
- What Is a Quant VC? A Practical Guide to Quantitative and AI-Native Venture CapitalA plain-English guide to quant VCs, AI-native VCs, and scientific venture capital — how Vela Partners uses data and AI to fund seed-stage AI founders.
- How Top VC Firms Evaluate Founders: The Taste Matrix of Sequoia, Benchmark, a16z, Founders Fund, Thrive, and LightspeedData from ~70 seed rounds (2024-2026) shows how Sequoia, Benchmark, a16z, Founders Fund, Thrive and Lightspeed pick founders, and which firm fits you best.