Mining for Capital Outside AI – $412 Billion Raised, Only $57 Billion Left in the Seams

US venture just printed a record half-year. Outside AI the pool of capital did not grow, the funds that deploy it got fewer and larger, and every raise now takes longer to land. The margin for a soft argument is gone.

86% of the Money Went to One Sector

PitchBook and NVCA published the Q2 2026 Venture Monitor on 9 July. The first half of 2026 put $412.7 billion into US companies, more than the whole of 2025. Of that, $355.9 billion went to AI – 86 percent of the dollars and about a third of the deals.

One sector understates it. Crunchbase counted $510 billion globally in the first half, and two companies, OpenAI and Anthropic, took $217 billion of it, 43 percent. In North America, Anthropic's $65 billion round was about half of everything invested in the second quarter. The record belongs to a sector, and inside the sector to a handful of names.

The record is real. It was also somebody else's.

Everyone Else Got the Same $57 Billion as Last Year

Subtract AI from the total and the rest of the market shared roughly $57 billion in the first half of 2026. Run the same subtraction on the first half of 2025 and the figure comes out near $59 billion. For a founder outside AI, a sponsor, a real estate manager or a first-time fund, the money available did not move while the headline doubled.

The Seams, by Name

Crunchbase's second-quarter sector dollars, global rather than US, give the market outside AI a shape. Cleantech and energy took about $8 billion in the quarter, the strongest since 2024, and $15 billion for the half. Cybersecurity took $4.4 billion, down about 30 percent on both the prior quarter and the year before, for $10.6 billion in the half. Defense reached $14.6 billion by May against $9.6 billion for all of 2025, on 107 rounds where 2025 had 206, which is the same concentration inside one sector. Biotech is running $36 to $40 billion a year and has not moved.

The stage numbers cut against the earliest-stage founder. North American seed funding came in around $4.9 billion in the second quarter, down 27 percent from a year earlier, and the early-stage deal count fell to its lowest point in five quarters. Two of these seams grew, two shrank and one stood still. None of them is where the headline came from.

Preqin's second-quarter real estate summary gives direction without dollar figures – fundraising up on the quarter, value-added strategies at their largest quarterly take since mid-2023, direct property deal count at its lowest since the pandemic. A real estate manager is raising into more money chasing fewer deals, and every one of those deals gets read harder.

Fewer Funds, Bigger Checks, Fewer Slots

Across all private markets, managers raised $658.1 billion in the first half of 2026, and funds larger than $1 billion took 78.2 percent of it, up from 59.1 percent in 2021. In venture, three firms – Andreessen Horowitz, Thrive Capital and Founders Fund – accounted for 48.1 percent of all capital raised. Experienced venture managers raised $64.5 billion; emerging managers raised $7.9 billion.

Private equity ran the same direction a year earlier. Buyout funds closed in 2025 fell 23 percent by count and 16 percent by dollars, to $395 billion. In real estate, 87.4 percent of the funds that closed in 2024 came from managers on their fourth vehicle or later, the highest share in more than a decade, into a pool of $85.8 billion against $241.7 billion in 2022.

Fewer, larger funds mean fewer slots and a sharper filter at each one. The committee that writes the check has a shorter list and more time to spend on each name.

The First-Time Fund Is Raising Into the Narrowest Seam

First-time venture funds raised $3.4 billion across 53 vehicles in the first half of 2026. Annualized, that is well short of the $11.4 billion first-time funds raised in all of 2025, and PitchBook has first-time fund formation on pace for its lowest year since 2016. A year earlier the same population had raised $1.8 billion across 44 funds. The first half is up on the first half of 2025. The year is not.

Set emerging managers against experienced ones and the share is what matters. In the second quarter of 2025 emerging managers took 23.1 percent of venture fundraising. In the first half of 2026 they raised $7.9 billion against $64.5 billion, which works out to about 11 percent. A quarter against a half is not a clean comparison, but the direction is not in doubt – the slot open to an emerging manager has roughly halved.

The 2025 figure is worth reading twice. First-time funds raised $1.8 billion in the first half of that year and $11.4 billion for the whole of it, so more than $9 billion landed in the second half. The year was made in its second half.

A Raise Now Takes Sixteen to Twenty-Two Months

The median US buyout fund took 16.2 months to close in 2024, up from 11 months in 2022. Real estate funds averaged 21.8 months, a record; the asset class had never averaged more than 16. Preqin's mid-2026 read on private equity has more than half of funds needing 19 to 30 months. Venture funds took a median 15 months to close in 2023.

The fourth quarter opens on 1 October. Allocators who close their books at year end are already writing next year's list. A raise that has not landed by then is competing against a budget that does not exist yet.

Two Hundred Screened, Four Funded

The average venture firm looks at 200 companies a year and invests in four. One in ten of those 200 arrives cold from the company itself; the rest come through the firm's network or its own sourcing. When 885 VCs were asked what they weigh, 95 percent named the team and 47 percent named it first, ahead of the product, the market and the model.

The same survey found that 20 percent of VCs do not forecast cash flows at all when they invest, and 9 percent use no quantitative metric. The model is not the first thing examined; the founder's account of it is, in person, with follow-up questions. The people asking are the ones who told the same survey the team was a factor in their successes, 96 percent of them, and in their failures, 92 percent.

The team is the argument. The materials are how the team is examined before anyone meets it.

Same Word, Two Examinations

At the top of the 86 percent the materials are close to optional. Sequoia's guidance to founders on writing a business plan, on its own site, covers financials in one sentence – if you have any, please include. Y Combinator's standard deal is fixed before the founder says a word, $125,000 for 7 percent plus a $375,000 uncapped note, the same terms for every company in the batch. That is what it looks like when capital is not the constraint. The deck is optional at the top because the examination happens elsewhere – in the reference calls, in the batch, in the network and the firm's own sourcing, which between them supply nine of every ten deals the average firm sees.

In the seams the examination runs the other way. A sponsor's pro forma goes to a quality-of-earnings desk that rebuilds it from the general ledger. A first-time manager's track record goes to an allocator who asks which deals were his, deal by deal. A founder outside AI walks into a partner meeting at a firm that will fund four companies this year and has time to pull on every number. The word is the same. The test is not.

Loose Money Forgave a Bad Foundation

When capital was loose, a decent deck and a warm introduction could carry a flawed thesis to a term sheet. The market absorbed the error; the next round or the next fund covered for it. That capacity is what the arithmetic above removed.

What is left is the argument underneath the materials, examined by people whose job is to break it. On the sponsor side the committee reads the thesis for the assumption doing the most work and moves it. The lender reprices the debt the model assumed and reads the coverage ratio again. The quality-of-earnings team rebuilds EBITDA from the ledger and finds the add-back that was a hope. The operational due diligence desk asks whether the process the memo describes exists, and calls the people who would know. Each has a form, a sequence and a person whose year depends on finding the flaw.

On the venture side there is no form, and the test is harder to see. It is a partner meeting with a shorter list and more time for each name, where the associate opens the model and moves the assumption, and the partner calls the customer named on the traction and roadmap. The partner meeting is where the 47 percent who rank the team first do the ranking, and it runs on the founder's answers, not the founder's deck.

Rejection Is Not Feedback

Most founders and most first-time managers run the test in the wrong order. They build the materials, send them, and wait for the market to tell them the truth by passing. Rejection is not the truth. It is a polite substitute for it. “The timing is not right” carries no information about the flaw that ended the meeting.

The last dispatch, Burn After Pitching – Lose the Raise or Anchor the Argument, argued that the raise is lost in the twenty minutes after the pitch.

The Machine Agrees With the Founder

The 2026 version of the wrong order is faster. A founder asks a frontier model to build the deck, the memo and the model, and it does. It does not tell him the foundation is wrong. Stanford researchers tested eleven leading models, in a study posted in October 2025, and found they affirm a user's actions 50 percent more often than a human would; the people who used them came away more convinced they were right, and trusted the model more for it. OpenAI rolled back a GPT-4o release in April 2025 by its own account for being “overly supportive but disingenuous.”

The result is not a deck that looks machine-made. Every deck now looks machine-made. The result is a market size nobody built, a model that agrees with its own inputs and a use of funds that answers nothing a committee would ask, all of it fluent, all of it complete, none of it able to survive the first question. A sponsor who runs his assumptions past a friendly LP has done the same thing in a better suit.

The Stanford participants also rated the agreeable answers as higher quality. The founder is not being fooled by a bad tool. He is choosing, from among good tools, the answer that agrees with him, and the vendor's incentives point the same way.

The Only Route Left

Attacking one's own argument before the market attacks it used to be good practice. In a seam that did not grow, with fewer allocators and longer closes, it is the only route. Either the founder finds the flaw first, or the allocator finds it and says the timing is not right.