A few years ago I was having coffee with an LP who wanted to understand what we did differently. He ran a family office and had looked at plenty of seed funds. Midway through the second cup, he asked the question every investor eventually asks a high-volume GP. "How many companies are you going to invest in?"

I said 150 over the fund's three-year deployment window.

He paused, then, polite but pointed, said the thing I have now heard dozens of times. "Isn't that spray and pray?"

That conversation is why this piece exists, and why we built the Power Law Simulator, an interactive tool that lets anyone test the math themselves. It's free, no login. You can open it on your phone, drag a slider, and see the argument we have been making for years play out in real time. It is also why we have historically invested in volume, fund after fund, and this piece lays out the reasoning in full.

Here is the argument in one sentence. Venture outcomes follow an extreme power law, and power laws reward more at-bats with better-quality returns, not fewer at-bats with bigger bets.

That is not a slogan. It is a consequence of how the returns distribute. The shape of the distribution is not a matter of opinion. It comes from the largest publicly available dataset of venture outcomes (Correlation Ventures, 21,000 financings), from Horsley Bridge's empirical analysis of what separates top-tier VC funds from the rest, and from the Kelly criterion, the math gamblers and quantitative investors use to decide how much to bet when the odds are asymmetric. Three lenses on the same underlying data. They all point in the same direction.

Let me show you what the math looks like. Then I will explain why more companies produces more value per dollar, not less.

The odds of a good fund lift with volume

Imagine you are running a seed fund with industry-average luck, writing checks at a blended $15 million entry price. Fifty-three percent of your companies go to zero. Thirteen percent give you your money back. Ten percent land small exits worth around 3x, twelve percent exit near $125 million for roughly 6x, six percent reach several hundred million for 15x, and about two and a half percent land just under the billion-dollar line. Two percent, cumulatively, become unicorns or better, which after dilution returns about 55x on your entry check. Roughly one in a thousand reaches $10 billion or more, one in five thousand crosses $100 billion, and about one in fifty thousand becomes an Apple, Nvidia, or SpaceX. (The probabilities are calibrated to Correlation Ventures' 21,000-financings study and Carta's cohort data; the multiples reflect Carta's per-round dilution medians. You can edit any of them in the simulator if you think they're wrong.)

Most seed funds hold around 30 companies. At that size, with reserves deployed sensibly into the winners, the math says a fund clears 3x gross only two times in three, and the median outcome is about 3.7x gross. After fees, that is a fund most LPs would politely call "fine," delivered with a coin-flip's worth of confidence.

Now invest in 400 companies with the same luck and the same per-company budget.

The probability of clearing 3x climbs from 66 percent to 99 percent. A good fund stops being a two-in-three shot and becomes a near-certainty. The median rises from 3.7x to 4.6x gross, and the range of likely outcomes tightens to roughly a quarter of its former spread, which is the difference between a return you hope for and a return you can plan around. Run the same comparison on the simulator's 50/50 blend of the industry-average and YC-calibrated distributions and the numbers get better still: a 6.8x median gross and a 93 percent chance of clearing 5x at 400 positions.

Nothing in the outcome distribution changed. We did not improve at picking. We did not forecast winners. The quality of returns got better because we took more shots at a distribution where almost all the value lives in the tail. At 30 positions, you need to catch a unicorn to look good. At 400, the math hands you several unicorns in expectation, and the fund's outcome stops depending on any single bet.

The intuition is the one every poker player already has. If one suit in the deck pays a million and everything else pays nothing, and the dealer is about to turn over cards, you would rather see more of the deck. The concentrated fund is making a smaller number of draws from a distribution where the payoff is wildly asymmetric. Sometimes it works. More often, the unicorn doesn't show up in those 30 cards.

Same math, two strategies

The chart below is the inside of the simulator. The two panels show 30 positions and 400 positions running against the same simulated outcome distribution (the 50/50 blend) with the same random seed, so the draws are mathematically identical. The only thing that differs is how many positions the fund owns.

Read the numbers in pairs. The concentrated side has higher variance, which is the technical way of saying the distribution of possible fund outcomes is wide. Sometimes a 30-position fund catches a unicorn and looks like a genius. More often it doesn't, and the whole fund underperforms. The distributed side compresses that variance. Fewer funds land in the tails at all, and more of them land somewhere an investor can live with. On the blend, the probability of a 3x fund rises from 86 percent to essentially 100, and the probability of a 5x fund climbs from 55 percent to 93.

If you are thinking "the concentrated fund's upside is still real, so maybe it's worth the variance," that is the honest version of the counter-argument. A concentrated fund in 1999 that caught Google returned every dollar you put in several hundred times over. The trouble is that you are not the person who caught Google, and you don't know who is. The 30-position fund that doesn't catch Google is the typical outcome, not the exceptional one. For an average GP picking 30 companies, the distribution of fund outcomes is dominated by the versions where the unicorn never comes.

Why this isn't obvious

If the math is this clear, the reasonable question is why every allocator isn't already buying it. The answer is that they already are. They just take it through more expensive wrappers.

A $500 million commitment to a fund of funds lands in 20 to 40 underlying managers, each of whom is running a portfolio of 30 to 80 companies. The endowment, pension, or sovereign writing that check ends up with effective exposure across 800 to 3,000 startups. Mega funds like Sequoia Capital Global Growth and Tiger Global hold hundreds of positions across their strategies for the same reason. The biggest pools of institutional capital have already decided, with their allocations, that venture exposure wants to be held at hundreds or thousands of underlying positions. That is a settled practice at scale.

What is newer is that a single fund with the right deal flow can produce position counts that rival a fund of funds, without the double layer of fees. That requires being able to see, evaluate, and invest in enough high-quality companies per year to make the math work. Which brings us to the part that took us over five years to build.

The binding constraint

Here is what the simulator cannot show you. A high-volume portfolio requires access to hundreds of investable companies across a deployment window. Our funds have historically run 150 or more positions over three-year deployment periods, which means sourcing at a pace most seed funds never reach. Their deal flow runs out. Most concentrated funds are not concentrated by choice. They are concentrated by necessity.

We are not. Y Combinator produces roughly 800 pre-vetted companies per year across four batches. That is the raw input, and our network now sources a comparable share of the portfolio from the broader early-stage market at similar quality, something that wasn't true when we started. But raw input isn't enough on its own. You also need the relationships to get in at the terms you want, the intelligence to tell the signal from the noise, and the operating muscle to support hundreds of founders without becoming the investor who doesn't pick up the phone. Volume without the rest of the stack is the spray and pray caricature. Volume with the stack is something else entirely.

The stack is where this gets interesting, and it is the part that compounds.

Every investment we make adds to the network. Every founder who works with us can become a scout, a reference, or a customer for another portfolio company. Every VC we co-invest with becomes a co-invest partner for the next deal. Every conversation on the Ignite Podcast becomes a relationship with a founder, VC, LP, or operator who now knows us. Every post brings in another intelligent reader who wants to understand private markets better.

Said out loud, that sounds like marketing. On the ground it is a set of very specific loops. More companies means more referrals. More referrals means better deal flow. Better deal flow means better companies invested in. Better companies means better outcomes. Better outcomes means more founders want to work with you. More founders want to work with you means more referrals. Run that loop for half a decade and the deal flow problem that constrains most seed funds stops being the binding constraint.

The same loop runs through the other sides of the business. Three thousand VC relationships means we can route a founder toward the right Series A lead in the right week. Thousands community members means when a founder needs ten customer conversations, they get ten. Half a million monthly readers across social means a portfolio company launch moves when we push it. None of this is free. But the marginal cost of adding one more position, given that the network is already in place, is almost nothing. And the marginal value of that position, both for the company and the network, is real and compounds geometrically.

That is how volume becomes a strategy instead of a prayer.

The honest caveats

If I only tell you the good part, I have failed you. The math has real limitations and it is worth being explicit about them.

First, the simulation assumes the outcome distribution holds across every position. In reality, if you invest in every YC batch without discrimination, you are also catching the bottom of the batch. Selection still matters. The simulator has a skill slider that adjusts for this, and at every level of skill the volume argument still holds, but the magnitudes change. A below-average picker investing in 400 companies still does meaningfully better than the same picker investing in 30. A great picker investing in 400 does materially better than a great picker investing in 30. Skill and volume compound.

Second, the follow-on math in the simulator assumes threshold-based reserve deployment rather than conviction-based. Real GPs make judgment calls on which winners deserve more capital. The simulator handles this by scaling follow-on outcomes with the skill parameter, but the abstraction is imperfect. This is one of the places I genuinely want feedback. If you have seen reserve deployment fail for a reason the model doesn't capture, tell me.

Third, the three frameworks in the simulator (Horsley Bridge, Kelly, and Monte Carlo) are not independent. They share distributional inputs. Change the assumption about how often a unicorn appears and all three move together. The simulator's Overview tab shows an expected-value breakdown that makes this explicit. Look at it. The rare outcomes do most of the work. If you disagree with the rare-outcome probabilities, your disagreement moves the whole result. The simulator lets you edit the distribution yourself.

Fourth, and most importantly, this is our strategy. It is not a universal prescription. A great investor running Benchmark-style conviction at 10 positions can beat us if their picking is great and their access is deep. Historical concentrated funds that worked produced extraordinary returns. The math in this simulator is the defense of our particular approach, not a claim that everyone else is wrong.

Fifth, the numbers in this piece come from a specific set of defaults. If you change the defaults, you get different numbers. Some of those differences will challenge my framing. Good. That is the point.

Try it yourself

The Power Law Simulator is at https://www.teamignite.vc/resources/power-law-simulator. It's free and takes no login. You can move a slider from 5 to 1,000 positions and watch what happens. You can flip between the industry-average distribution, the YC-calibrated distribution, and a 50/50 blend of the two. You can edit the underlying probabilities, the entry valuation, and the per-tier dilution assumptions if you think ours are too generous or too conservative. You can change the skill parameter. You can compare concentrated and distributed strategies head to head with the same random seed so the comparison is fair.

If you allocate to venture, I would particularly encourage you to play with the head-to-head tab. Pick a GP pitch you have heard recently, estimate the position count they are running, and see what the math says the distribution of outcomes looks like for their strategy. You do not have to believe our numbers. Edit them until you believe them. The shape of the answer will probably surprise you.

We shipped this fast so we could get feedback early. If something is confusing, wrong, or missing, please say so. I read every response.

We have run this strategy because we believe the math. We built the simulator because we think anyone evaluating venture deserves to see the math for themselves, without having to take any GP's word for it, including ours. Venture capital has too many people telling you what to believe and not enough people showing you why.

Here is the math. Here is the code. Run it. Push back. If you come to a different conclusion, I want to hear what it is.

If you want to see the companies we have backed, check out our portfolio here. If you want to listen to founders and fund managers describe how all of this plays out in practice, the Ignite Podcast is three hundred-plus episodes deep at this point.

More than anything, the best thing you can do is open the simulator and spend five minutes with it. The numbers make the argument better than I can.

FAQ: the pushback I have already gotten

Before publishing I sent this piece to a friend who runs another VC firm and has a computational CS background. He was skeptical. Specifically, he pushed back on the math. He pushed back hard on every assumption in the simulator, and some of his pushback was right. What follows is a set of questions that came up in that conversation and in early reviews, with honest answers rather than defensive ones.

Aren't you treating each investment as independent when they clearly aren't?

Yes. That is a real limitation of the model and it deserves a direct answer.

A Monte Carlo simulation assumes each draw is independent and identically distributed, meaning each company in the portfolio is treated as a fresh roll from the same distribution. In reality, the 150th investment in a fund can differ from the 1st. For a fund with a fixed deal pool and declining access, the 150th pick is worse than the 1st. For a fund like ours, drawing from YC batches that refresh with a consistent quality bar, the pool is replenished continuously. The 150th pick across a three-year deployment is drawn from a universe of thousands of pre-filtered companies, not from the same batch with the best ones already taken. Whether the distribution degrades, stays flat, or improves with position count is a function of the specific fund's sourcing model. The simulator assumes it stays flat, which is closer to our reality than the degrading-distribution case. Not to mention our deal flow leads to more deal flow.

The simulator is a thought experiment that says, "if position count were the only variable that changed, what would the math predict?" It's useful for building intuition about why power-law distributions reward more at-bats. It is not a fund model. A real fund has correlated outcomes across sectors and vintages, a selection function that interacts with portfolio size, and network dynamics that evolve over time. The simulator captures none of those.

What it does capture is directionally correct for a picker with stable deal flow: more positions produce better expected outcomes in a power-law world. What it cannot show is the exact tradeoff between position count and per-position quality, which is the question every real GP has to answer with judgment rather than math.

Isn't a smaller portfolio better because it lets you focus on the best companies?

Only if you can reliably tell which companies will be the best in advance. The research on selection says most GPs cannot.

Consider the arithmetic. A 20-position fund investing out of a YC batch of 200 is making an implicit bet that its 20 picks will include the batch's outliers. If its selection is no better than random, the expected number of outliers captured is one-tenth of the batch's total outliers. A 150-position fund investing out of the same batch captures three-quarters of whatever outliers that batch produces, regardless of picking ability. For the concentrated fund's smaller portfolio to be the better choice, the GP has to be confident their picking meaningfully beats the batch's own filter, which is already one of the most selective in venture (roughly a 1% acceptance rate).

That is a higher bar than most GPs can clear. Including, to be clear, us. We do not claim to out-pick the YC admissions process. We claim to run a disciplined investment framework on top of an already-selected population, and to capture more of that population's tail than a concentrated fund can. The concentration bet and the volume bet make different assumptions about selection ability. Volume is the honest choice for a GP who does not believe they have a reliable crystal ball.

Where does your outcome distribution come from, and why do you assume it applies to every position in a high-volume portfolio?

The distributions come from published datasets: Correlation Ventures' 21,000-financings study, Horsley Bridge's published outlier rates, Cambridge Associates' industry averages, and Carta's cohort and dilution data. Sources are cited in this piece below and in the simulator footer.

The fair critique is that those distributions are averages across the whole venture industry, and no single fund's outcomes match the industry average. A top-tier fund has a better distribution. A bottom-tier fund has a worse one. Applying a single distribution uniformly to hundreds of positions assumes you can hold quality constant as you scale, which is a defensible assumption for some sourcing models and not others.

That's why the simulator has a skill slider. Move it up and the distribution shifts toward better outcomes. Move it down and it shifts the other way. The volume argument holds at every skill level, but the magnitudes change. A below-average picker still does better with 400 positions than with 30. A top-tier picker does even better with 400 than with 30. Skill and volume compound. Neither replaces the other.

What the simulator doesn't do yet, and what I'm adding to a future version, is let users configure how the distribution changes across position count. The current assumption of "stays flat" is our view for a fund with a compounding network drawing on a continuously refreshed pool. Others will disagree, and the simulator should let them test their own assumptions.

How can diversification reduce your risk if sector risk and vintage risk don't go away?

It can't. And it doesn't.

Diversification across hundreds of positions reduces idiosyncratic risk, which is the risk tied to any specific company. It does nothing about systematic risk. If the entire AI category collapses, or interest rates wreck the IPO market, or a macro event shuts down exits for three years, a 400-position portfolio suffers the same way a 30-position portfolio does.

The claim in the piece is narrower than "diversification reduces risk." The claim is that for the specific risk of a fund underperforming its target return because none of its companies became outliers, more positions reduce that specific risk. That is a real benefit. It is not a claim that all risks are reduced.

Anyone considering this strategy still has to think about sector concentration (we're heavily AI and B2B SaaS, so that is real risk), vintage concentration (our prior funds deployed into overlapping market conditions), and macro exposure (we can't diversify away a 2008 or a 2022). The volume argument is specifically about power-law capture, not a general claim of risk reduction.

Standard deviation of fund outcomes isn't observable the way it is in public markets. Isn't the "spread tightens" claim fake?

Partially yes. And this is a good catch.

Public companies are marked to market every day, so standard deviation of returns is a measured quantity. Private companies are valued quarterly at best, usually lag the real value by a year or more, and only resolve to actual dollars on exit. So "standard deviation of fund MOIC" isn't a number you can point at in reality. It's a number that falls out of a simulation.

What the simulator shows is that within the model, simulated fund outcomes cluster more tightly as position count grows. That statement is mathematically true given the independence assumption. Whether it translates to real-world funds is a separate question.

The better way to read the tightening is as "distribution of simulated outcomes," not "distribution of real fund outcomes." Real funds have additional sources of variance the simulator doesn't model. So the absolute number (a standard deviation around 6x at 400 positions, versus roughly 21x at 30) is not a prediction for any actual fund. It's a statement about what the math does when you hold everything else constant.

Doesn't the efficient frontier say there's no free lunch? How can you lower risk without lowering return?

You can't, and the piece doesn't claim you can.

Look at the head-to-head comparison carefully. The 400-position portfolio has a lower probability of extreme upside than the 30-position portfolio does. The concentrated fund has a wider distribution on both sides. Sometimes it produces an outlier outcome the distributed fund might not reach, because the distributed fund is pulling toward its average by holding many positions.

What the distributed fund trades is the tail risk on both ends. It gives up some ceiling in exchange for a higher floor and a higher median. For an investor trying to plan a portfolio around a predictable target return, that trade is attractive. For an investor explicitly seeking maximum upside and willing to accept higher variance, it isn't. The volume strategy is not Pareto-superior to concentration. It is optimized for a different objective function.

The efficient frontier still holds. We are picking a different point on it than a concentrated fund picks, and we think that point is the right one for most allocators. That is a judgment call, not a math proof.

If the math says more positions is better, why doesn't the perfect investor just invest in everything?

They don't, and that's the right intuition pump to stress-test the model.

A perfect investor with oracle-level foresight would invest in 3 companies: the three terracorns (trillion dollar outcomes) of the next decade. Their portfolio would return tens of thousands of x. That's the theoretical optimum, and a high-volume strategy would have trouble beating it.

The reason the volume strategy makes sense is that nobody has oracle-level foresight. Among real investors, the spectrum runs from "well above average" to "well below average," and for the vast majority of the distribution, adding positions improves expected outcomes because the cost of missing a winner is higher than the cost of adding a loser.

The implication is not "volume is always right." The implication is "volume is right unless you have genuine oracle-level insight, in which case concentrate." We think most investors, including most professional GPs, are closer to average than oracle. We count ourselves in that group, not above it.

Peter Thiel says concentration wins. Aren't you arguing against him?

Thiel made Facebook returns on a $500,000 check because he had genuine contrarian insight and the access to act on it. That worked. I'm not arguing it didn't.

The argument is that Thiel's strategy is right for Thiel and wrong for almost everyone else. The counterexample to his framing is the hundreds of concentrated-conviction funds that didn't catch Facebook and underperformed as a result. We only remember the ones that got the home run. Survivorship bias makes concentrated strategies look better than they have performed in aggregate.

If you genuinely believe you are Peter Thiel, concentrate. If you believe you are a disciplined, well-networked, hard-working, thoughtful investor who is not literally Peter Thiel, the math says spread out a bit.

The simulator compares the same distribution at different portfolio sizes. What happens if I change the distribution as I scale?

Fair question, and it's the one I owe the most detailed answer to. The current simulator holds the outcome distribution constant as position count grows. That assumption is closer to our reality than to most funds' reality, for reasons explained in the first answer. But reasonable people disagree about it, and the simulator should let them test their own view.

A user could model distribution-degrading scale by splitting the portfolio into tiers. For example: the first 30 positions draw from the YC-calibrated distribution (6% unicorn rate), the next 60 draw from a slightly worse distribution (4%), the rest draw from a worse one still (2%). That would model a fund whose deal flow quality declines as it extends its reach.

I'm adding that configurability to a future version as a user-adjustable toggle. The user will be able to set whether the distribution improves, stays flat, or degrades with position count, and see how the result changes. Our own view is that for a fund with an active, compounding network drawing on a continuously replenished batch system, the distribution stays flat or improves slightly over the deployment window. We think that because our sourcing draws from pools that refresh every quarter, because our network compounds with each portfolio addition, and because our selection tooling has gotten meaningfully better with scale, not worse. But I want you to test your own assumptions, not mine.

If you want to stress-test this today, the skill slider is the closest proxy. Drop the skill to below-average and see what portfolio size maximizes expected return. That's a rough approximation of what happens when deal flow quality degrades.

What is this simulator for, then?

It is a pedagogical tool, not a fund model. Its job is to show why power-law distributions make volume-based strategies mathematically coherent, not to predict the outcomes of our funds or anyone else's.

If you are trying to decide whether a volume strategy is defensible, the simulator is useful. If you are trying to figure out whether any specific fund will hit a specific multiple, the simulator tells you nothing you can use. The real question is whether a manager's deal flow, selection, network, and operating stack are good enough to capture the volume strategy's theoretical upside in practice. That is an empirical question about the manager, not a question the simulator can answer.

The reason to publish the simulator anyway is that most people evaluating venture have never seen the underlying math laid out in a form they can poke at. The argument for a volume strategy is usually hand-waved with phrases like "more shots on goal" and "power law," which is not persuasive. Showing the math, and explicitly flagging where it breaks down, gives readers something to push back on. The conversations I've had since publishing it are better than the conversations I was having before. That was the goal.

So is the strategy defensible or not?

The strategy is defensible. The specific numbers in the simulator are directionally useful and depend on assumptions a reasonable person might or might not share. Those are two different statements and both are true.

A fund running hundreds of positions with average deal flow and average selection will in expectation beat a fund running 30 positions with the same deal flow and selection. That claim holds up under every reasonable modification of the model I have tested, including versions that let the distribution degrade modestly with scale. What shifts with better modeling is the magnitude of the advantage and the optimal position count. What doesn't shift is the direction.

If you think we have below-average deal flow and selection, the strategy still works but the numbers are worse. If you think we have above-average deal flow and selection, the strategy still works and the numbers are better. If you think you have top-tier deal flow and selection yourself, concentrate, and you disagree with your own assessment of your ability to tell the difference in advance.

The simulator is one piece of evidence. The rest of the evidence is our track record, our network, our operating platform, and the specific portfolio companies we have backed. Anyone evaluating a volume strategy should weigh the math alongside those, not instead of them.

Sources and further reading

The probability distributions in the simulator are calibrated from published research. Every claim in the piece and the simulator footer maps to one of these:

Outcome distributions

• Correlation Ventures' 21,000-financings study (covering 2004-2013), originally published by Seth Levine at sethlevine.com, with a 2019 update here. The source of the "65% of financings return below 1x" and "4% return 10x or more" figures.
• Carta's Class of 2018 cohort data on seed-funded startup outcomes, summarized by SaaStr here. The source of the "30-35% of seed-funded startups shut down within 7 years" and "1.3% become unicorns" figures.
• Carta's dilution-by-round medians (Series A ~19%, B ~15%, C ~11%, D ~9%), used to convert raw exit multiples into post-dilution returns: carta.com/data/founder-ownership-2026
• Carta's seed and pre-seed entry valuations, used for the $15M blended entry price: carta.com/data/state-of-private-markets-q3-2025

Outlier rates by fund tier

• Horsley Bridge outlier-rate data via Ulu Ventures, at uluventures.com/picking-winners-is-a-myth and their longer Portfolio Construction essay. The source of the "top-tier VC 4.5% outlier rate" and "average VC 2% outlier rate" figures. Cambridge Associates' 2.5% industry average is cited inside both of these pieces.
• The "superstar" end of the simulator's skill slider (roughly double the market unicorn rate) is a construct from Ulu Ventures' analysis of what separates top-decile managers from the average. It is not a separate empirical distribution, and the simulator labels it as an adjustable assumption, not a fact.

YC-specific numbers

• YC unicorn rate, Garry Tan on X, August 2025: x.com/garrytan/status/1953069914132238775. Tan's stated range is 6% to 12% for the last ten years of batches. Our default of 6% sits at the low end.
• PitchBook's 2023 analysis of YC cohorts, via Yahoo Finance. The source of the "4.5% of YC startups since 2010 become unicorns, 5.4% for 2010-2015 cohorts" figures.
• YC batch size is common knowledge, but does change every few years.

Fund economics

• Exit timing (the simulator's standardized 10-year hold from first check to exit): Crunchbase's analysis of startup exit timelines and SaaStr's data on the median years-to-exit for $1B+ SaaS acquisitions.
• 2-and-20 (2% annual management fee, 20% carry on profits above 1x) is an industry convention, not a single-source citation.

Team Ignite

• Our portfolio, listing the companies we have backed.
• The Ignite Podcast, three hundred-plus episodes deep.
• The simulator source code: github.com/BrianBeezy/power-law-simulator

Brian Bell is the founder and managing partner of Team Ignite Ventures, an early-stage firm investing in YC and broader early-stage B2B SaaS, AI, fintech, and marketplace companies. He hosts the Ignite Podcast and writes the Ignite Insights newsletter. The Power Law Simulator was built to let anyone stress-test venture math for themselves. Feedback welcomed.