Allocator One | The Vanishing Middle Round

Why startups now raise once, or raise billions. The ladder that defined startup financing for thirty years is breaking in the middle, and that changes who makes money in venture.

Figure 01 · The market in three numbers

87.5%of US venture dollars went to rounds of $100M+ in H1 2026 ^[1]

15–17%of 2022 seed companies raised a Series A within two years ^[2]

22% → 5%generalist share of new funds launched via VC Lab, 2020–2026 ^[3]

Two numbers from the same summer

Every quarter I read through the reports of the fund managers we back. This year, two kinds of stories kept showing up side by side. On one page, a company that raised a single seed round, turned profitable and has no plans to raise again. A few pages later, a company raising more in one round than most venture funds will raise in their lifetime.

The headlines tell the same story at scale. This summer, Lovable raised $400 million at a $13.3 billion valuation, and Mistral raised €3 billion, the largest equity round ever by a European tech company ^[4]. Judging by the headlines, venture capital has never been more abundant.

Now look one level down. Rounds of $100 million or more took 87.5% of the $412.7 billion invested in US startups in the first half of 2026 ^[1]. Everything below that line, which is most deals by number, shared $51.4 billion. And only about 15 to 17% of companies that raised a seed round in 2022 went on to raise a Series A within two years, against about 30% for the 2018 cohort ^[2].

Figure 02 · Where the money went, US, H1 2026
Segment	Value
Rounds of $100M+	$361.3B · 87.5%
All other rounds	$51.4B · 12.5%

Figure 02 · Where the money went, US, H1 2026

US venture investment by round size, H1 2026. Rounds of $100 million or more absorbed $361.3 billion of the $412.7 billion invested. Source: PitchBook-NVCA Venture Monitor, Q2 2026 ^[1].

The usual reading is a two-speed market: one for AI, one for everything else. I think something more structural is happening. The ladder that defined startup financing for thirty years, seed, then A, then B, then C, is breaking in the middle. Startups increasingly raise once, or they raise billions. The round in between, the Series B that built much of the venture industry, is hollowing out.

If that is right, it changes who makes money in venture. That is the question I spend most of my time on: which managers, and which models, still work when the rules of the game change.

The headcount ladder

The staircase of rounds was never a law of nature. It existed because building a software company meant hiring people, and people cost money long before revenue arrived. Each round bought roughly 18 to 24 months of headcount: engineers first, then salespeople, then managers to manage them. The Series B existed because a company that had found product-market fit needed fifty more people to prove it could scale.

I call this the headcount ladder: a financing structure whose real job was to pay for people. Every rung was sized to fund the next wave of hiring, and every valuation step was a bet that more people would produce more revenue.

I saw this up close earlier in my career, working on US growth at a venture-backed company on its way to becoming a unicorn. Every round came with a hiring plan attached. The money was, quite literally, a budget for people.

Take headcount out of the bottleneck, and the headcount ladder loses its reason to exist. That is exactly what AI does: it decouples output from headcount. Support, code, design and analysis, work that once required a hire, now requires a subscription.

The Series B paid for scale. When scale no longer means people, many companies simply skip it.

Force one: AI collapses the cost of reaching revenue

Start with team sizes. The median seed-stage company now has about four employees besides the founders, down from six or seven a few years ago. The median Series A company has 15 to 17, down from roughly 45. Solo founders made up 17% of new startups in 2017 and 36% by 2024 ^[2].

Revenue per employee makes the gap starker. The median private SaaS company generates about $130,000 per employee ^[5]. The fastest-growing AI-native companies average around $1.13 million ^[6]. That is a deliberately extreme comparison: the best AI-native companies against the typical SaaS company. But the frontier is where the next cohort starts, and it has moved by an order of magnitude.

Figure 03 · What $10M in revenue takes

Figure 03 · What $10M in revenue takes
Model	Summary	Note
Classic SaaS ($130K revenue per employee)	~77 people · $11.5M payroll	Needs a Series B
AI native ($1.1M revenue per employee)	~9 people · $1.4M payroll	Fits inside a $4M seed round

Illustrative comparison at median revenue per employee. Payroll assumes $150,000 fully loaded per head; the order of magnitude is the point, not the exact figure. Sources: SaaS Capital; Bessemer Venture Partners ^[5]^[6].

In our own network, companies are already living this. One synthetic biology company went from pre-product at pre-seed to $12 million in annual revenue and profitability in about two years. Across one deeptech manager's portfolio, 56% of companies already generate revenue, and they have won $48 million in non-dilutive grants alongside $89 million in follow-on rounds. For some companies, revenue and grants now do the job a Series B used to do.

Some founders now plan for this from day one. Founder Henry Shi, who popularised the term "seed-strapping", argues that a modest seed round plus early profits leaves almost no downside ^[7]. Sam Altman has said he and other tech CEOs have a betting pool on the year the first one-person billion-dollar company appears ^[8].

Force two: capital concentrates on the few

At the other end, the opposite is happening. Worldwide, AI startups raised more than $407 billion in the first half of 2026, more than in all of 2025, and OpenAI and Anthropic alone took about $217 billion ^[9]. Frontier AI is compute-hungry, winner-take-most, and priced accordingly. Of the $274.2 billion in venture-growth capital deployed through May, 86.4% went into just four rounds by three foundation-model companies ^[16].

I feel this in my own conversations. This year, one of the most common requests I got from investors was not for a new fund. It was for access to a handful of AI megarounds.

The capital behind those rounds is concentrating too. Three firms, Andreessen Horowitz, Thrive Capital and Founders Fund, took 48.1% of all US venture capital raised in H1 2026. First-time fund formation is on pace for its lowest year since 2016 ^[1].

The same polarisation even runs inside seed rounds. In Q1 2026, the number of seed deals fell about 30% while seed dollars rose 31%, to $12 billion ^[10]. In other words, fewer companies get funded, and the ones that do get far more.

The squeezed middle

Figure 04 · Half as many seeds grow up
Item	Share raising a Series A within two years
2018 seed cohort	About 30%
2022 seed cohort	15–17%

Figure 04 · Half as many seeds grow up

Share of seed-funded companies raising a Series A within two years: 30.6% for the 2018 cohort, about 15 to 17% for the 2022 cohort. Source: Carta ^[2].

Between these two forces, the middle is thinning. Of companies that raised a seed round of $1 million or more through 2020, typically 55% or more went on to raise again or exit, a broader measure than Series A graduation, since any new round counts. For the 2023 cohort, it is 24% so far. For 2024, 16%. Nearly 40% of seed financings in 2024 were bridge rounds ^[11]. The next step is slowing too: the gap between a Series A and a Series B is now almost twice as long as it was three years ago ^[2].

This is where the data stops being useful, and where I think most commentary gets it wrong. "Did not raise again" now describes two opposite outcomes. Some companies are profitable, growing and happily seed-strapped: $3 to 5 million in revenue, a team under 15, founders paying themselves and compounding quietly. Others are zombies, alive on a bridge extension, with flat revenue, waiting for a round that will never come. From the outside, on a Crunchbase profile, they look identical.

For a long time, a Series A was treated as the market's signal of product-market fit. That signal is breaking. The better signal now is unit economics: healthy gross margins even after compute costs, net revenue retention above 100%, and a burn multiple below 1.5, or no burn at all ^[12]. A seed-strapped company that shows those numbers does not need a Series A to survive. A zombie that doesn't won't be saved by one. Investors will need to underwrite the business, not the round. It's also where my conversations with managers start now, well before we get to markups.

The barbell on the cap table

You see the same split in who writes the cheques. Because we see most new fund managers before they close, we tend to notice how the investor side of the market is reorganising before it shows up in the reports. When I joined, a generalist pitch was still a normal part of the week. Today it stands out. Four patterns explain why.

Where this comes from

Around 1,000 emerging managers set out to raise a venture fund each year, and Allocator One sees roughly 80% of them. Since 2023 we have analysed more than 2,500 managers. Portfolio examples in this section come from the funds we back and are anonymised.

Source: Allocator One, author-supplied figures as at September 2026 ^[13].

1. Specialists are replacing generalists at the start. Among new funds launched through VC Lab, one of the largest fund-formation platforms, the generalist share fell from 22% in 2020 to around 5% by early 2026 ^[3]. Applications to our platform rose 33% year on year in Q1 2026, and the growth came overwhelmingly from specialist theses: AI infrastructure, robotics, defence and vertical software. Even funds that launched as generalists are converting. One Central and Eastern European manager in our network moved from a generalist regional strategy to a focus on European strategic industries, and every company in its portfolio now generates revenue.

2. Specialists are winning access to the best seed rounds. Domain expertise has become a way into competitive rounds rather than a niche. Founders now seek out one manager in our network, a fund dedicated entirely to type 1 diabetes, as their lead investor precisely because of its depth in the field. When money is abundant for the best companies, expertise is what makes a cheque stand out.

3. The giants use specialists as their filter. When those companies raise again, the lead is now often a large multistage firm arriving earlier and bigger than before: $20 million seed rounds led by firms like Accel, and early Series A rounds led by firms like Atomico ^[14]. And the giants lean on specialists to get there. Two examples from our network: one university-spinout manager is trusted by later-stage investors such as General Catalyst for its technical due diligence, and a synthetic biology specialist regularly co-invests alongside a16z. The specialist finds and de-risks; the giant scales.

4. Capital for new managers is shrinking, even as their role grows. Established firms took 90.9% of all US venture capital raised in Q1 2026, the highest share PitchBook has ever recorded ^[15]. By midyear, first-time managers accounted for less than 10% of the capital raised ^[16]. The market relies on specialists to find the next generation of companies, and gives them less money to do it.

So the investor side now looks exactly like the round side: specialists at the very beginning, a handful of giants at the end, and the mid-sized generalist, historically the Series B leader, squeezed out of both. The generalists I meet are not worse investors than they were five years ago. They have simply lost the job only they could do.

What this does to venture maths

When I walk investors through how a seed fund actually makes money, this is the part that surprises them most. The short version: in a raise-once world, the maths tilts towards small, specialised funds. It takes three steps to see why.

Step 1: you keep more of the company. Every new round issues new shares, so earlier investors end up with a smaller slice. After a typical Series A, B and C, a seed investor who started with 10% owns about 5.4% (10% × 0.8 × 0.8 × 0.85). If the company never raises again, they keep the full 10%.

Step 2: so you need a smaller exit. For one company to pay back an entire fund, the fund's stake multiplied by the sale price has to equal the fund size. A $100 million fund that keeps its 10% needs a $1 billion sale. Diluted to 5.4%, it needs about $1.85 billion. Keeping your stake almost halves the exit you need.

Step 3: and small funds need the smallest exits of all. The same formula punishes size. A $20 million fund owning 10% returns itself with a $200 million sale. A $1 billion fund needs $10 billion, an outcome 50 times larger, and outcomes that big are rare, even in a record year. A $200 million exit can make a seed fund's year. For a billion-dollar fund, it barely moves the needle. Smaller funds need far smaller wins, and those come along far more often.

Figure 05 · A small fund needs a 50x smaller exit
Item	Exit needed to return the fund at 10% ownership
$20M fund	$200M
$1B fund	$10B

Figure 05 · A small fund needs a 50x smaller exit

Exit needed for a single company to return the whole fund, assuming 10% ownership at exit. Illustrative: $20 million ÷ 10% = $200 million; $1 billion ÷ 10% = $10 billion.

Put the three steps together and the conclusion is simple. A company that raises once is a gift to the small fund that got in first, and a problem for the mid-sized fund that was waiting to lead its Series B.

This has two consequences that most of the industry has not priced in yet.

LP reporting breaks. No new rounds means no markups. A profitable seed-strapped company stays at its entry valuation on paper, while a peer that raised an inflated Series A shows a threefold markup. The best raise-once portfolios will look worst on paper for years, and LPs who judge managers only on interim marks will back the wrong funds. The fix is to look through the marks: portfolio revenue, profitability, and the share of companies that no longer need capital.

Reserves stop making sense. Holding back around half of a fund for follow-on rounds is wasted when most companies either never need more money, or need far more than a regular seed fund can provide.

Where the old ladder survives

Three things cut against my argument, and they deserve a fair hearing. First, part of the squeeze is cyclical. Many of the weakest cohorts raised at 2021 valuations, and early data shows some recovery: 10 to 11% of 2025 seed companies reached Series A within a year, against 4 to 5% in 2022 and 2023 ^[2].

Second, the very best AI companies will keep raising, because when the prize is winner-take-most, speed beats efficiency. On 20VC, SaaStr founder Jason Lemkin and Scale Venture Partners' Rory O'Driscoll describe the winning playbook in AI as raising more than you need and outspending everyone else ^[17]. Top-tier firms now go further, picking a category winner early and funding it so heavily that rivals cannot keep up. The industry has a name for it: kingmaking ^[18].

Third, and most awkward for my own thesis: many of the specialist sectors winning today, robotics, defence and AI infrastructure, are capital-hungry. Their bottleneck is hardware, compute and certification, not headcount, and AI does little to shrink it. In these sectors the middle round survives, but it changes shape. It is sized to fund factories and compute rather than hires, and it often blends venture with grants, strategic capital and project finance. The ladder survives where atoms, not people, are the bottleneck.

At a glance

Rounds of $100M or more took 87.5% of US venture dollars in H1 2026.
Only 15 to 17% of 2022 seed companies raised a Series A within two years, about half the 2018 rate.
Venture rounds were a headcount ladder, a machine for financing hiring. AI decouples revenue from headcount: at the fastest-growing AI-native companies, $10M in revenue takes about 9 people, against 77 at a median SaaS company.
Without markups, the best raise-once portfolios will look worst on paper, so LPs judging on interim returns risk backing the wrong managers.
Generalist strategies among new funds fell from 22% in 2020 to around 5% by early 2026.
The result is a barbell: specialist funds win at the start, the largest platforms win at the end, and the mid-sized Series B fund loses its market from both sides.

A bet I would be happy to lose

Most people in venture have an opinion on this. Few are willing to put a number on it. So here is mine.

In 2018, roughly 1 in 3 startups that raised a seed round went on to raise a Series A within two years. I don't think we will see that again. My bet: neither the 2025 nor the 2026 seed cohort, the first to raise entirely in the AI era, will get back to 1 in 3, even as venture keeps breaking records.

The 2023 and 2024 cohorts are already tracking far below that mark, so the real test is what comes next. And the next cohort is recovering: of the startups that raised seed in 2025, about 1 in 10 raised a Series A within a year, more than double the rate of 2022 and 2023. If the second year adds about as much as the first, that points to roughly 1 in 5 over two years. If it follows the 2022 pattern, where most graduations came in year two, it gets uncomfortably close. That is why this is a bet, not a forecast.

If I am wrong, the old ladder still works, and the last few years were just a hangover from 2021. If I am right, the middle round is not coming back.

We will know by the end of 2028.

The round that splits in two

The headcount ladder is breaking in the middle, and the Series B is splitting in two: a round many of the best and most profitable companies will never need, and a round only a handful can raise.

For founders, the question shifts from "how do I get to the next round?" to "do I need one at all?" For investors, the question is simpler, and harder.

When investors ask me what to back today, I start with a different question than I did two years ago: not which fund is best, but which end of the barbell it is built for.

Standing in the middle is no longer a strategy. Pick a side.