A $5 million check into a pre-seed company at a $100 million valuation is a terrible investment.
Unless you manage $15 billion. Then it's necessary. Because you cannot afford to miss one of the ~20 decacorns a decade produces.
That logic explains more about where venture is headed than any market map you'll see this year.
The top of the market is playing a different sport
Start with the numbers, because they're wild.
In January, a16z closed a little more than $15 billion in new funds. Ben Horowitz said that was over 18% of all US venture dollars allocated in 2025. In Q1 2026, per PitchBook-NVCA data, 73.1% of all LP capital raised went to just five firms. Roughly 75% of deal value went to five companies.
That's not a market. That's a velvet rope.
Now look at what those funds have to hunt. Rory O'Driscoll pointed out on 20VC that the US has produced roughly 130 tech companies ever valued at $10B or more. That's about 20 a decade, for five decades straight. (2026 has minted 42 privately with a quarter to go. That says more about private marks than about real exits.)
So run the math top-down. A $10B fund that wants to return 3x needs $30B back. Own 10% at exit and you need $300B of enterprise value. Across one portfolio. In a world that produces a couple dozen real decacorns a decade.
There's no strategy there except one: be in the handful of companies that matter. All of them. At almost any price.
That's why the $100M pre-seed doesn't need to work. Its job is to buy the right to write the $500M check later.
It's not an investment. It's an option premium.
Lemkin summed up the a16z logic on 20VC: "The price of overpaying for a winner is being in the super winner." O'Driscoll went further a year earlier: when outcomes are $100B+, "seed is for suckers."
The middle of the curve just got emptied out
Here's what nobody at the top has an incentive to say:
If every mega-fund is hunting the same 15-or-so trillion-dollar candidates, everything else gets left on the table.
And there's a lot of opportunity sitting where nobody's looking.
Building is cheap now. Five engineers with agents ship what fifty used to. Base44, one founder about six months in, sold to Wix for ~$80M. Midjourney reportedly hit ~$200M in revenue with a few dozen people and zero venture money. If you're not a neolab training frontier models, you probably don't need a war chest.
You need one round. Maybe.
Does that create a second kind of founder, and a second fund model to match?
Two founders, two math problems
Founder A: Trillion or bust. Raise early, raise huge, raise constantly. Compute, talent, distribution: capital is the weapon. This is the Anthropic path. Every round is a bigger bet on a bigger outcome, and the only exit that justifies the stack is a generational one.
Founder B: Seed-strap. Raise once. $3M. Get to profitability on it. No A, no B, no pro rata games. Build to real revenue and sell for a billion to a Fortune 500 buyer who can swallow a $1B deal without a board fight or a regulator in the room. This is the Zapier path: a ~$1.3M seed in 2012, no primary capital since, profitable since 2014, and a 2021 secondary at roughly $5B.
Venture has spent 15 years calling Founder B a lifestyle business.
That's lazy.
Ownership is the hidden variable
Traditional seed math: a seed check used to buy 15–20%. Today it's closer to 10%, if you're lucky. Then the A, B and C rounds each dilute you 15–25%. By exit you own maybe 4–6%, unless you keep paying up for pro rata, which a $100M fund can't afford anyway.
Seed-strap math: you buy ~10%. And you keep it. No later round comes along to dilute you.
10% of a $1B exit is $100M.
1% of a $10B exit is $100M.
Same check back. One of those outcomes happens a lot more often than the other.
Now run it through a $100M fund: 30 companies at ~$3M a pop, ~10% each. One seed-strapped company that sells for $1B returns the whole fund on its own. Hit three and you're at 3x. Make one of them a $2B exit, which a Fortune 500 buyer can still digest, and you're at 4x. Run the same playbook out of a $30M fund and a single $1B exit returns it more than 3x.
That's top-quartile performance. No decacorn. No follow-on.
The small fund doesn't need the trillion-dollar company. It needs ownership that survives and exits that clear.
"But cost of capital…"
Let me steelman it, because I can hear Harry Stebbings in my ear yelling it: cost of capital. Cost of capital. Cost of capital.
Here's the argument. When capital is abundant, the best companies get it cheap. A seed-strapped founder in a contested market gets steamrolled by a rival who raised $400M, bought distribution and priced at zero. Refusing to raise isn't discipline. It's unilateral disarmament.
In some markets, that's exactly what happens. Winner-take-all, capital-intensive, network-effect-driven: you can't seed-strap a foundation model, and you can't seed-strap a defense prime.
But.
Venture has always been about investing at the top of the curve. But with this much attention and capital piling in up there, pricing at the top only goes one way. The opportunity might be somewhere else.
Most AI tools aren't winner-take-all. The agent that runs dental billing, the AI underwriter for specialty insurance, the copilot built into one workflow nobody else understands: they're winner-take-a-lot.
The model underneath is rented, and it gets cheaper every quarter. The edge is the workflow, the data and the customer.
The real competitor isn't another VC-funded AI startup. It's the incumbent that would rather buy the AI layer than build it. In those markets a $400M war chest buys a bigger burn, not a bigger moat.
And "cost of capital" cuts both ways for the investor. 3x on a $100M fund is a great outcome for every LP in it. Nobody's apologizing for that.
Where the GP sits in all this
Here's the uncomfortable version for people like me.
Say you've built the sourcing, the tooling and the diligence engine, and you're writing $3M checks that hold their ownership all the way to a $1B exit. The question changes.
Why carry LPs at all?
Some have already walked that road. Chamath turned Social Capital into a holding company in 2018 and stopped taking outside money. We've watched a couple of firms quietly hand back the fund and move to the balance sheet. We've had that conversation internally too. When your edge comes from a machine you built, handing 80% of the upside to someone else starts to look like a strange trade.
The 2% game rewards raising. The 20% game rewards returns. A balance-sheet investor only plays the second one.
Most GPs can't make that move. They don't have the capital or the edge. But the fact that the conversation is happening at all tells you the mid-size fund model is being re-priced from the inside.
What I don't know
I'm not picking a side here. There isn't a right one.
Run the math and both can win. Easily.
The split might be clean: two lanes, two founder types, two fund sizes, each with its own math. It might be messy instead. The mega-funds could hunt down the seed-strapped companies that start working, offer them $200M they never asked for, and pull them back into the trillion-or-bust machine. Or the $1B M&A market could tighten if acquirers stay cautious.
Here's what I'm confident about.
Founders: decide which game you're playing before the first check. Your cap table will record that choice whether you meant to make it or not. If you're Founder B, pick investors built for one round, not ones who need you to raise four more.
LPs: stop grading every fund against a16z-sized outcomes. A $50M fund doesn't need a trillion-dollar company. Ask whether the GP's ownership survives to exit, and whether the portfolio's exit sizes match the fund's size.
Emerging GPs: the mega-funds aren't your competition anymore. They left the room. The only question is whether you'll build for the room they left.
The bottom line: the top of venture is buying options on the future of civilization.
The middle is buying companies.
Both can make money. Only one of them is crowded.
So which side of the curve are you building for?
Sources: TechCrunch, "The venture firm that ate Silicon Valley just raised another $15B" (Jan 9, 2026); SaaStr, "VC in 2026: 75% of All the Money Is Going to Just 5 VC Funds" (Apr 16, 2026; PitchBook-NVCA Q1 2026 data); BigGo Finance, summary of 20VC with Jason Lemkin and Rory O'Driscoll on the decacorn boom (Oct 15, 2026); SaaStr, recap of 20VC, "Why seed is for suckers" (Apr 17, 2025); Revenue Memo, "Who owns Zapier" (Aug 27, 2026).