Allocator One | The Great Divergence

What happens to edge and judgment when analysis is nearly free. The venture market has separated into two poles, the largest funds and the specialists, and the middle is emptying.

The market is splitting in two, and the gap widens every quarter.

At one end sit a handful of firms with more capital than anyone has ever deployed at this stage. At the other end sit specialists. People who spent twenty years inside one field and now invest in it, close enough to the work to be in the room before there is a round to price at all.

Those two ends are not doing the same job. The large funds commit once a company is visible, at prices that only work if the outcome is enormous. Their edge is scale and distribution. They can move a market and fund an outcome into existence.

The specialists work with a different edge, and it is not foresight. At the point they commit, they usually know more about the field than anyone else in the room, because they spent twenty years inside it. And they can actually get into the deal, which is a separate thing from wanting to. Those are calculated bets, made on better information than anyone else has at that stage, at the only moment when the price is still low.

Almost everything that will matter in ten years is being decided at that second end, by people most allocators have never heard of.

That is the divergence. It does not stop at fund managers. The same split is running through the companies they back and through the industries those companies are rebuilding.

Figure 01 · Two different sources of edge
Group	Items
Specialists	Early access, Domain knowledge, Entry ownership
Mega funds	Scale, Distribution, Capacity to fund outcomes

Figure 01 · Two different sources of edge

Conceptual comparison of the author’s thesis. Positioning is qualitative; this is not a return comparison.

And it is fast. An industry used to take a decade to turn over, which was long enough for an incumbent to see it coming and do something about it. That window has gone. Categories are now repriced inside a single planning cycle.

Which is why this should concern you whether or not you invest in any of it:

Your next competitor is being funded right now, by someone you have never heard of.
The returns of the next decade are being priced this year, at entry, and you cannot buy that price back later.
If you work in government, the tax base of 2040 is being decided by where these people choose to build.

None of it waits.

Where this comes from

Around 800 emerging managers set out to raise a venture fund each year. They do not all close in the same year, and many never close at all. We see roughly 80% of them.

Since 2023 we have analysed more than 2,500, interviewed over 500 in depth, and taken more than 40 positions, in 85% of cases as the largest investor. Alongside that I have had more than 500 conversations with investors across more than 35 countries, and over 100 of them are now invested in what we do.

This is not a market outlook. It is what I have learned from inside a funnel almost nobody else sees, written for the people who have to decide where capital, companies and countries go next.

Source: Allocator One, author-supplied figures as at September 2026. Cumulative activity since 2023; thresholds are not exact counts or conversion rates ^[1].

Figure 02 · The manager selection pipeline

2,500+Managers analysed since 2023 ^[1]

500+In-depth interviews ^[1]

40+Positions taken, 85% as the largest investor ^[1]

What the divergence actually is

Take any industry and plot the returns of the companies inside it. Take any profession and plot what the people in it earn. For most of modern economic history you got the same shape both times. A dense middle, thin edges, and an average that described something real.

Nobody was ever trying to own the middle. Everybody hunted outliers. The difference is that in that world, if you aimed at outliers and missed, you still landed somewhere reasonable. Average was the consolation prize and it was survivable.

Missing the outlier no longer leaves you with average. It leaves you with nothing that matters.

The economy is starting to behave the way venture capital always has. A small number of outcomes account for nearly all of the value, and the long tail is not slightly behind, it is structurally irrelevant.

The reason is simple. Being good at analysis used to separate people, and now it separates nobody, because everyone has it. Any competent person can read every document, build every comparison and produce a credible plan for the price of a coffee. Competence is not an advantage any more. It is the floor, and everyone stands on it at the same moment.

What still separates people is what a machine cannot produce. Knowing which founder ships and which one presents. Getting the call before the round is open. Deciding before the evidence is complete.

Those things do not get cheaper as software gets cheaper. They get rarer, because everything around them is now free and they are not. That is the whole mechanism.

Figure 03 · From a dense middle to a heavy tail

Figure 03 · From a dense middle to a heavy tail
Panel	Shape	Note
A clustered distribution	Bell-shaped: most outcomes cluster around the middle
A heavy right tail	Heavy right tail: most outcomes are small, a few are very large	Rare, large outcomes

Illustrative distributions, not measured economic data. The curves express the author’s thesis; they do not establish that every industry follows a power law.

Where the name comes from.

Around 1800, China and Western Europe were at roughly the same level. Comparable cities, comparable technology, comparable standards of living. Within a century one had industrialised and the other had not, and the gap took two hundred years to close.

Historians call that the Great Divergence, and the explanation was never that one side was cleverer than the other. Britain had coal sitting close to its industrial centres, the machines to burn it, and colonies supplying both raw material and demand. China had comparable ingenuity and its coal a thousand kilometres from where its people lived. Steam then compounded the advantage every decade, because each gain funded the next one.

The gap did not open because one side thought harder. It opened because one side had access first, and the advantage was self reinforcing. That is the argument of everything below.

Bigger than the internet, and mostly good news

People reach for the internet and the smartphone as the comparison. Both changed how we live. Neither is the right analogy.

The first difference is what gets automated. The earlier waves automated transactions and coordination. They made it cheaper to reach a customer, move a payment, book a journey, publish a page, and a lot of work disappeared along the way. But a person still had to decide what the work was. Every step had to be specified by someone before a machine could carry it out.

What is different now is that the specification itself is being produced. The judgment layer, the part we all assumed was permanently ours, is the part being automated. Anyone using the last two waves as their reference point is underestimating this one.

The second difference decides the timing. The internet had to be delivered before it could change anything. Cables, servers, browsers, then handsets, networks, app stores. Two decades passed before it reached an ordinary small business in an ordinary town.

This wave needs no delivery. The build is enormous, but it is happening upstream in compute and power, out of sight of the people it affects. At the point of use it arrives through devices and software that everyone already owns, on the day it ships. That is why the bakery is exposed in the same year as the software company, and why the twenty year absorption curve most institutions still plan around does not apply.

There will be corrections on the way, and probably a bad one. That is not an argument against any of this, and web 1.0 is the reason why. The dot com bubble burst and took most of the companies with it, but the fibre, the servers and the payment rails stayed. The value did not stay where the money was first spent. It moved off the infrastructure layer and onto the applications and aggregators built on top of it. Almost nobody in 1999 was buying the layer that eventually captured the returns.

The same shift is coming here. Today the money and the attention sit at the model and compute layer, which is where the capital intensity is. As models mature and commoditise, the value pool moves up: to applications built into specific workflows, to businesses with proprietary data a general model cannot reach, and to whoever owns the customer relationship in a particular industry. Healthcare, defence, industrials, logistics, financial operations.

Those are not markets a generalist covers well. They are markets you have to know, which is the whole argument for specialists, and the reason a correction would change the timing of this and not the direction.

Figure 04 · Where the author expects value to move
Step	Detail
1. Compute and models	Capital intensity
2. Industry workflows	Proprietary data
3. Customer relationship	Distribution and trust

Figure 04 · Where the author expects value to move

Conceptual sequence from the essay. This is a forecast, not an observed allocation of profits.

The third difference is the one investors feel. The last two waves were mostly additive. They built new companies alongside the old ones, and the old ones had years to respond. This one is substitutive. It reprices what already exists, and it reaches your portfolio before it reaches your strategy deck.

And still, the speed is not a bad thing.

It is a challenge for all of us, including me, but the direction has been the same for decades. My parents' generation could not see the faces of people they loved on another continent. Now everyone can, for free. Travel used to be uncomfortable, expensive and dangerous. Illnesses that killed people fifty years ago are managed with a prescription today. Every one of those arrived through a technology somebody at the time found alarming.

What is coming will do the same thing again, faster. Diagnostics available to people who currently have no doctor within a hundred kilometres. Education that adapts to a child instead of the other way round. Work that is less repetitive for a lot of people who never had a choice about it.

I am optimistic about the destination. My argument in this paper is only about the journey, and about who ends up close enough to it to have a say.

What this does to investing

Everything our industry uses was designed for the first distribution:

Diversification assumes that spreading capital lowers your risk without costing you the upside.
Benchmarks assume the average describes something.
Mean reversion assumes what falls comes back.
Risk models assume volatility around a centre.

All of it is technology for a bell curve world, and in one it works beautifully.

In a power law those same instruments start to mislead. The average return of a venture portfolio describes no fund in it. The median tells you nothing about the outcome you will actually get. Spreading capital more thinly stops reducing risk and starts raising the probability that you miss the only positions that mattered. Diversification quietly turns from protection into dilution of your one chance at the tail.

The second change is where the risk sits. In a clustered world the danger is a bad decision. In a dispersed one it is the absence of a decision. If nearly all of the return comes from a small number of outcomes, then not being there when they happen costs more than any single mistake you could make.

Sitting out a vintage feels prudent, and is not.

Third, edge changes character. An analytical edge now decays, because whatever you worked out this quarter is available to everyone next quarter for nothing. Structural edges compound: proprietary access, a reputation that gets you the call, a community that tells you things before they are written down, the willingness to commit before the evidence is complete. Anything you can buy, everyone can buy.

Which leaves two defensible positions:

Own the index at the top, through the largest funds, and accept that you are buying the market.
Go early and concentrated, with managers you have actually selected rather than managers who happened to reach you.

Moderate exposure to moderately good managers is where most institutional capital sits today, and it is the one place the maths no longer works.

All of this is happening while institutional capital opens up in a more structured way than it has in a decade. Evergreen and semi-liquid structures are proliferating ^[2]^[3], ELTIF 2.0 has pulled European wealth and smaller institutions into private markets ^[4], and in the United States a March 2026 proposal would open defined contribution assets to the asset class for the first time ^[5]. Most of that flow is heading into credit and buyout rather than venture. The structures are arriving before the allocation logic has caught up, which is exactly the window in which early decisions matter most.

Seed is where you see it first

Here is the claim I would most like people to take away.

If you want to know what is about to happen to your industry, your region or your country, the place to look is pre-seed and seed. Not because that is where the returns are, although they are, but because that is where the information is. It is the earliest legible signal of what is being built, by whom, and against which incumbent.

By Series B a company is already a fact. The ownership is distributed, the price is set, the information advantage is gone, and the industry it is disrupting has usually not yet noticed. Waiting for that point is not caution, it is arriving after the interesting part.

The clock has moved. Bessemer's data on the fastest AI companies has them reaching around 100 million dollars of recurring revenue in roughly eighteen months ^[6], against the seven year benchmark that governed software for two decades ^[7]. That number describes the outliers, not the average company. Most businesses are still slow. But outliers are the entire subject of this industry.

Figure 05 · Growth speed depends on the cohort
Item	Years to reach $100 million ARR
AI Supernovas	1.5 years
AI companies in Cloud 100	5.7 years
All Cloud 100 companies	7.5 years

Figure 05 · Growth speed depends on the cohort

Bessemer 2025 benchmarks. Supernovas are a selected breakout group; Cloud 100 cohorts use different selection criteria. This is not an average for all AI or software companies. Sources: Bessemer Venture Partners, State of AI 2025 and Cloud 100 Benchmarks 2025 ^[6]^[7].

Hamilton Lane's 2026 market overview finds AI companies reaching real revenue faster than in any earlier technology cycle ^[8]. In our own portfolio, through DQ Ventures, Thinking Machines went from pre-seed to a 40 billion dollar valuation in under 24 months ^[1].

We see it in our own book. Since we began deploying in 2023 we hold more than 15 unicorns in the portfolio ^[1]. That is not a claim about our skill, it is a description of how fast the underlying market is moving underneath us.

Marks are not returns, and I would rather be judged on cash.

Our flagship fund began deploying in 2024 and is at roughly 0.15 DPI ^[1]. For context, PitchBook puts the 2021 vintage at 0.05x DPI after five years, the weakest since 1997 ^[9]. Ours is a small number on a young fund and I am not presenting it as proof of anything, except that the model returns capital rather than only reporting it.

Something else has changed that I did not expect, and it may matter more than the speed. In a growing number of the specialised funds we have backed, portfolio companies reach profitability within a round or two and then grow out of their own cash flow. Some have decided not to raise again at all.

That quietly breaks an assumption the whole industry rests on. Venture maths has always assumed serial dilution. You enter early, you are diluted through successive rounds, and the later capital funds the growth that eventually produces your return.

If a company is profitable after seed and never comes back to the market, the earliest cheque is the only cheque. Ownership at entry stops being a negotiating position and becomes the whole outcome. It also means a large amount of capital has been raised on the assumption it will be needed at Series B and beyond, and some of it will find there is nothing there to fund.

So a few scouts in San Francisco and a calendar of conferences is not an information strategy any more. It was a reasonable answer in 2015. It is not one now, whether you run a company, a bank or a ministry.

The world is not one market

The other thing you cannot see from a single geography is how differently regions are developing. What follows is a read rather than a finding. It is what I perceive from 35 countries of conversations and from what reaches our platform.

What we can measure

In the current application cycle we received 1,700+ applications from managers raising a fund. 198 of them came from Asian markets, across 12 countries, and 70% of those were from India. Median target fund size 20 million dollars, and roughly three quarters of the applicants had prior venture experience. That is who is actually trying to raise.

Source: Allocator One, current application cycle as at September 2026. The 70% figure uses the 198 Asian applications as its denominator. This is an applicant pool, not a regional market survey ^[1].

Figure 06 · The Asian manager application pool

1,700+Global applications this cycle ^[1]

198From Asian markets, across 12 countries ^[1]

70%Of Asian applications came from India ^[1]

India is moving fastest. Not just in capital but in the density of people building, the quality of the manager layer forming underneath it, and the domestic pull that lets companies scale at home first. If I had to name one region where the base rate has genuinely changed in the last three years, it is this one ^[10].

The Gulf is harder than the headline suggests. Regional funding hit a record in 2025, driven by Saudi Arabia and the UAE ^[11], but the sovereign capital is increasingly buying into global platforms and then pulling those companies into the region physically. That is a legitimate strategy and it may work. It is not the same as building a local manager ecosystem, and the manager layer is where I see stagnation rather than growth.

Africa is coming, but not there yet. The talent and the problems are both real, and funding has held up better than in most emerging regions. My honest estimate is three to four years before the manager layer is deep enough to support serious allocation.

Southeast Asia is in a genuine contraction. Venture deal value fell around 34% in 2025 ^[12], and in the first half of 2026 just one dedicated regional fund reached a final close ^[13]. Capital is rotating into pan-Asian vehicles that are not obliged to deploy there. Founders should plan accordingly.

Asia as a whole is dynamic, heterogeneous and, on the manager layer, genuinely encouraging. We see rising momentum in new manager formation in China and Japan, and Singapore building steadily as a hub for the wider region. The contraction in Southeast Asian deal value and the growth in people setting up to invest there are happening at the same time, which is usually what the beginning of a cycle looks like vis-à-vis the end of one. The one thing the whole region has in common is that everyone watches China closely and learns from it, which is not true anywhere else.

Valuations and exits remain American, and I expect that to hold for years. That is where the technology fortunes sit, where the buyers are, and where the public market has the depth. This does not mean managers must be American, and it does not mean a region loses when its companies exit elsewhere. A founder who builds at home and lists abroad still comes back with capital, with experience and with a network, and that is how talent clusters form. The ecosystem is built by the people, not by the venue of the listing.

Europe is specialising quickly. Among new funds launched through VC Lab, one of the largest fund-formation platforms, the share of generalist strategies has fallen from 22% in 2020 to around 5% by early 2026 ^[14], and the shift is visible in almost every European fund we see. I would not claim Europe leads the world on this, only that the direction is unmistakable here. Europe's problem was never talent at the early stage. It is what happens at growth.

I hold a handful of dated predictions on all of this, including that by 2030 as many new institutional managers will form in Asia each year as in the United States, and that Europe keeps producing seed talent while losing its winners at growth. We publish them with their dates and grade ourselves against them in public every year.

The ordinary economy is next

Start with a bakery. Ordering, pricing, rostering, the bookkeeping, the accountant who does the year end, the delivery app, the person who answers the phone. Every one of those layers is being rebuilt right now, and the owner did not ask for it and did not vote on it.

Multiply that by every traditional business in every economy and you have the actual story. This is not AI as a sector. It is AI applied to the ordinary economy.

Figure 07 · The ordinary business as an exposure map
Part	Of
Ordering	Bakery
Pricing	Bakery
Rostering	Bakery
Bookkeeping	Bakery
Delivery	Bakery
Customer calls	Bakery

Figure 07 · The ordinary business as an exposure map

Conceptual map of functions named in the essay. Lines show exposure to change, not measured automation or job losses.

Private equity is the most concentrated and most leveraged claim on exactly those businesses, which is why it registers the shock first.

In February 2026 public software lost about 1 trillion dollars of market value in a single week, as investors repriced what agents do to recurring revenue ^[15].
Private equity technology deal value fell roughly 70% between the fourth quarter of 2025 and the first quarter of 2026 ^[16].
On Bain's numbers, around 32,000 companies worth roughly 3.8 trillion dollars now sit unsold ^[17].
Distributions have stayed below 15% of net asset value for four years running, an industry record ^[17].

Figure 08 · The private equity exit backlog

32,000Unsold companies ^[17]

$3.8tnUnrealised value ^[17]

~7 yearsBuyout holding periods at exit ^[17]

Small and mid cap private equity is now going through what venture went through, one cycle later. On PitchBook's 2026 data, funds under one billion dollars have taken under 17% of all capital raised this year, and only 23 first time funds reached a close in the first half, against an average of 181 a year between 2021 and 2023 ^[18]. Same shape, same sequence.

My own prediction, which we publish with a date and grade ourselves against annually, is that this process is largely complete within three years. Anyone sitting in traditional private equity today should assume they are already inside it, not approaching it.

What it now takes to be a GP

The job description changed, and it changed at the very beginning of the funnel rather than at the end.

It used to be enough to have a network, a thesis and a plausible track record. The expectations are now higher on every dimension at once:

Real data insight, not a market map.
DPI discipline, not a story about marks.
Co-investment capability, because LPs want the exposure without the second layer.
A community, because founders choose on more than money.

Being a GP is one of the hardest jobs I know. It is closer to being a doctor than to being an investor. You are on call permanently, there is no version where you say no to the thing that needs you at eleven at night, and you carry other people's outcomes.

The compensation for the ones who do it well is genuinely attractive, in venture financially as well as emotionally, and I think that is fair given what it costs. I would say the same about doctors and the people who work alongside them in most countries, who are paid badly for work of the same intensity, in a system about to go through its own transformation.

One more thing about the people doing this job. The word emerging has done real damage. It suggests someone doing this for the first time, which is almost never what we are looking at. What we see is an operator with two decades of proven judgment in one domain, carrying a brand that is three months old. The fund is new. The person is not.

Both facts matter, and the combination is the point. The experience is old and the exposure is new. A partner in fund seven has a reputation that survives a bad vintage. A first fund has nothing to survive on. Twenty years of standing is attached to a name nobody recognises yet, and they know it every day. So they work harder on one position than a large firm works on ten.

Selection is a craft, not a search

Here is where I want to be most direct, because it is the part most people get wrong.

Analysis may be free. Selecting a GP is not.

It is a distinct skill and it cannot be done in a meeting, or in ten minutes, or by reading a data room carefully. It requires proprietary data, constant presence in the market, and the ability to compare, rank and vet a manager against hundreds of others you have seen at the same stage of their journey. You cannot benchmark a first time manager against anything if you only see twenty a year.

We have selected more than 40 positions out of that pipeline, and the portfolio has outperformed the top quartile consistently since 2023 ^[1]. Early rankings are computed off marks and should be treated with care, which is why I put more weight on the cash.

I am not saying we do this perfectly. We learn something every week, we get things wrong, and a lot of what we are building for our managers is still ahead of us. What I am confident about is the discipline itself. It is slow, repetitive work, and it is the only part of this business that cannot be shortcut.

What we say no for is more instructive than what we say yes to. Across the 1,700+ applications in this cycle, three reasons dominate ^[1].

1. No investment experience. Not operator experience, investment experience. Have they put their own money in, in amounts that meant something to them? Have they made a significant number of decisions themselves inside another fund? An operator background on its own is not a track record.

2. No commercial drive. We are not looking for the philanthropic or the purpose driven manager. We want people who can produce solid returns and have the data to show that outcome is likely. Speed, rigorous execution, the ability to sell, to win access to a round that is already competitive, and to make a founder's next quarter measurably easier.

3. No honesty about their own process, usually a lucky angel track record presented as a method. Ask a manager to name a deal they lost and why. You learn more in ninety seconds than in a full data room.

The test of whether any of that works is not our marks. It is whether the managers stand up on their own afterwards. Of those we anchored in 2023, around 70% are back in market with a second fund ^[1], in a period when first time fund formation fell 78% from its peak on PitchBook-NVCA numbers ^[19].

It also needs technology now. When more than 1,700 managers apply in a single cycle, you cannot rank them on memory and meetings. You need the data to be coherent, comparable and fast, or you are simply reacting to whoever was most persuasive last week.

So anyone who wants this exposure has an honest choice to make:

Do it properly, which means years of it, constant presence, a team and the tooling.
Do it inside a community of people who also do it, so your benchmarks are not only your own.
Or give it to someone whose sole occupation this is, whoever that turns out to be.

What does not work, and what a great many institutions are currently doing, is doing it occasionally, from a distance, on the basis of whichever funds happen to reach them. That is not selection. It is sampling, and in a market this dispersed sampling is hard to distinguish from guessing.

Why specialists, and not only mega funds

This is the part of the market everyone is talking about, and mostly getting wrong. The argument is not that mega funds are bad investors. It is that they are playing a different game, and that the game they are not playing is where most of the value will be created.

Hundreds of thousands of companies are going through disruption right now, across every industry and every region. No global firm can cover that, and the mega funds are not trying to. They operate as market makers with distribution power, concentrating on the model layer and on a small number of names they can fund into existence. On Crunchbase's first half 2026 data, two companies absorbed 217 billion dollars, around 43% of all venture funding raised worldwide ^[20]. That is index construction with a balance sheet, and it is entirely rational at their size.

Figure 09 · Two companies captured 43% of funding
Segment	Value
OpenAI and Anthropic	$217bn · 43%
All other companies	$293bn · 57%

Figure 09 · Two companies captured 43% of funding

Other companies: $293bn, calculated as $510bn minus $217bn. Shares rounded. Funding volume does not measure investment returns. Source: Crunchbase, 2 July 2026 ^[20].

The other pole is thinning fast, and this is why I say we need fewer funds rather than more. On PitchBook-NVCA data the number of US venture firms fell in 2025 for the first time on record, fund closings dropped from 1,793 in 2022 to 626 in 2025, and first time funds fell 78%, from 478 to 106 ^[19]^[21]. That is not a downturn passing through. It is the middle of the market being cleared out, and most of what is going was capitalised by conditions rather than selected by anyone.

What the largest firms cannot do is see early. Several of our managers reached global winners before they did, because they were closer to the people building them. Nordic Web was into Lovable at the start ^[1]. Lovable raised a Series C in August 2026 at 13.3 billion dollars, roughly seven times its Series A mark thirteen months earlier, with revenue tracking toward 600 million dollars annualised ^[22]^[23]^[24]. None of that came from a bigger balance sheet or a better model. It came from being close enough to recognise it first.

Specialists do not compete with mega funds for the mega rounds. They arrive earlier and cheaper, and they help in ways a balance sheet cannot.

The other side of this is real.

Emerging managers do not simply outperform. Their distribution is wider at both ends, and the median tells you almost nothing. If you cannot select, that spread is an argument for staying out, not for going in. Which is the point. This is a selection business, not an allocation business.

The people holding the capital

Those 500 conversations are mostly family offices, entrepreneurs who have exited, and private investors. I have spoken with pension funds and other institutions as well, far fewer of them, so I do not claim a full picture on that side. What I have seen there points the same way the surveys do.

Private Equity International's LP Perspectives 2026 found that most institutional LPs have no plans to cut their venture commitments, even though half of them say their venture holdings are behind benchmark ^[25]. That matches my own institutional conversations, and the explanation is not that institutions are cleverer. They are simply more consistent by construction. They have a policy allocation, a committee and a pacing model, so they keep going through a bad stretch because the process makes them.

Private capital has none of that, and nobody is surveying it. It decides year by year, based on how the last few years felt.

Most of them treat venture as a gamble. Something you try with a slice of the portfolio, and stop when it disappoints. That is the misunderstanding underneath almost everything below.

Venture done properly is a structured allocation, and the return is not the only thing it produces. It gives you market intelligence you cannot buy anywhere else, and it gives you access: to growth equity positions, to secondaries, to partnerships, and to companies you might eventually want to own.

And in a period of change this fast, it gives you something harder to price. Clarity. When everything else is fog, this is the one asset class that shows you what is actually being built and by whom, well before it is visible anywhere else. The secondary effects run across the whole portfolio: what you hold in public markets, what your operating businesses should be preparing for, what you buy next. Treated that way it is infrastructure for the rest of your capital, not a lottery ticket held beside it.

Those investors sort into roughly five groups of similar size. The groups overlap at the edges, but the pattern has held for three years ^[1].

One in five has understood that and is acting on it. They deploy through the cycle rather than around it.

Two in five tried it, got burned, and stopped. This is the group I think about most. One partner disappointed you, so no more partners, ever. Nobody would accept that logic anywhere else in their life.

In public markets everyone has accepted that you do not time the market. You stay invested, because the days that carry the return are few and impossible to identify in advance, and missing them costs far more than the drawdown you avoided. Private markets work the same way, only harder, because the good years are concentrated in a handful of positions and you cannot buy your way in afterwards at any price.

Nobody sends you a statement for the vintage you skipped.

Some vintages produce less. That is how the asset class works, and the answer is to be in all of them with discipline and to back only the top of the market. The bill for sitting out never arrives, which is exactly why it is the most expensive position available.

One in five is in secondaries and mega fund exposure. A fair strategy, and I understand it. It buys scale and some liquidity, and it accepts market pricing to get them.

Some of that is conviction and some of it is career safety. Nobody was ever fired for hiring McKinsey, and nobody was ever fired for committing to the best known fund in the market. A brand name protects the person who made the decision even in the years it does not protect the return. That is a real consideration for anyone who has to explain a portfolio to a family, a board or a committee, and it is worth naming it.

One in five believes direct investing is better and that they can do it themselves. Some have found a genuine sweet spot for their single family office. A sector they know from operating in it, a deal flow that comes to them, a size that suits them. That makes complete sense and I would not argue with it.

What it does not give you is the index, the insight or the outlier. You are seeing the deals that reach you, not the market, so you are not participating in the breadth that produces the intelligence, and you are unlikely to be in the positions that carry the returns.

The gap is usually not skill, it is a reference point. In a period changing this fast you need a structured benchmark group and a community around you, so that what you decide is measured against what several hundred other people are seeing rather than against your own last few deals.

There are two other kinds of capital owners in this, and they are making the same decision under different names.

Corporates. I spent close to a decade building and investing alongside multinationals including Allianz, Riyadh Bank and Vattenfall. Building new businesses inside a large company does work. I have seen it work, and I would still do it under the right conditions.

What has changed is the conditions. The operating model a corporate would have to put in place today is not fit for the speed this moment demands, and the capital required to be genuinely competitive has risen to a level most corporate budgets are not built for.

Which is why LP investing, partnership and acquisition are now the smarter route into the same outcome. You see the market properly, you work alongside the companies already ahead, and you buy when the thesis is proven rather than funding your way to it. There is one exception. When the new business grows out of a capability you already hold at world class level, Amazon building AWS on infrastructure it had built for itself, then building is right.

What I would not attempt is starting at the growth stage. By then the competition is fierce, the price reflects it, and you are buying into a race that others began years earlier.

Governments. This one is urgent and under-discussed. Countries are competing for the same small population of exceptional builders and the managers who back them. A nation that wants its industries transformed rather than displaced has to attract the GPs, not only the companies, because the managers arrive first and the ecosystem forms around them.

Figure 10 · The ecosystem formation sequence
Step	Detail
1. Managers	Local presence
2. Companies	Formation and growth
3. Ecosystem	Talent and capital

Figure 10 · The ecosystem formation sequence

Conceptual sequence in the author’s argument. Arrows show the proposed mechanism, not a measured causal effect.

That means working closely with market participants instead of around them. And it means being honest about the practical things that decide where people go: what it costs to set up, how quickly a capital round can be done, and how the whole thing is taxed.

I am in favour of a strong welfare state. But the money to fund one has to be earned somewhere, and if the structures here are slower and more expensive than elsewhere, the talent and the capital will simply go where they are not. The earliest stage is where national outcomes are decided, several years before any of it shows up in an economic statistic.

At a glance

The venture market has separated into two poles: a handful of firms with more capital than anyone has ever deployed at this stage, and specialists who know their field better than anyone else in the room. The middle is emptying: US fund closings fell from 1,793 in 2022 to 626 in 2025, and first-time funds fell 78%.
Being good at analysis used to separate people. Now everyone has it. What sits above it is judgment, access, and the willingness to commit before the evidence is complete.
In a market this dispersed, diversification costs upside, and the danger is the absence of a decision rather than a bad one. Sitting out a vintage feels prudent, and is not.
Two positions remain defensible: own the index at the top, or go early and concentrated, with managers you have actually selected. Moderate exposure to moderately good managers is the one place the maths no longer works.
Seed is the earliest legible signal of what is being built. The fastest AI companies reach about 100 million dollars of revenue in about eighteen months, and where a company never raises again, ownership at entry becomes the entire outcome.
This is a selection problem, not an allocation problem. We say no to roughly 97% of the managers who reach us.

Why I do this

I want to end with something that is not an argument.

I am extremely grateful that I get to do this every day, and I learn from the people in this community constantly. It was never the intention to build a fund. The mission is to make it transparent who is actually out there, to give access to the outliers, to support the people doing the difficult work of transformation, and to build the infrastructure that makes market intelligence, co-investment and running a fund work the way they should by 2030 rather than the way they did in 2010.

We need fewer funds, not more. But we need the right ones. At the moment too many of the right people never get seen, while a lot of capital sits with people who were funded by conditions, not selected by anyone.

The people closest to a field see it first. That is the whole argument of this paper, and if it is right, then over time the capital ends up with them. By 2040 I believe the people who understand a field will be the ones funding it. That is what all of this is for.

The divergence is not coming. You are already inside it. The question is whether you are close enough to see it.