The Anchor Is the Trade

A concentrated market does not diversify by growing. It concentrates further, because the capital that arrives buys what is already liquid, and what is already liquid is what everyone else already owns.

In May this newsletter called the vacancy: 75% of carry-bearing SPVs sat in 5 names, the top 20 took 81% of secondary trading value on Hiive, and the tallest name was about to leave. In July I reported that it had arrived, and that instead of the expected volume cliff the next tier was scrambling to re-anchor.

PitchBook has now measured both halves of that, and the picture is sharper than I put it. In the trailing 12 months through June 30, the US venture secondary market reached $121.7B, split between $107.1B of direct secondaries and $14.6B of GP-led. That is 2.14x the $50.0B annualized run rate of Q4 2024. The market more than doubled in 7 quarters, and it got narrower while doing it. On Hiive, the top 20 names took 86% of Q2 trading value, up 5 percentage points from the level I cited in May, and the top 5 alone took 50.3%, up more than 5 points. Caplight's top 5 accounted for 42% of trading value over the same 12 months.

The most important number in the report sits in the appendix. PitchBook sizes the pool of private stakes realistically available for sale at $94.2B. Excluding OpenAI and Anthropic, it falls to $60.6B. Two companies are 36% of the entire addressable opportunity, and both have filed to leave.

PitchBook expects volume to crater over coming quarters as those lockups clear and cash flows back into the ecosystem, and frames it as the natural effect of a top-heavy market rather than a crisis. That matches the staged release schedule I walked through in July, which pushes most of the recycled capital into Q4 and the first half of 2027. The supply gap and the returning demand still do not arrive together.

What is new since July is the other end of the barbell. On June 16 SpaceX agreed to acquire Anysphere, the company behind Cursor, for $60B in all stock, removing a name that had been trading near $29B in the secondary. So the count is now down to one. Of the 5 names that defined this market, SpaceX is public, xAI is inside it, Cursor is absorbed, Anthropic and OpenAI are both sitting on filed S-1s, and only Anduril still trades as a pure private secondary.

Three things follow, and only the first is widely discussed.

1. The discount is not a valuation. It is an age.

Forge's completed-trade data, cut by the year of each company's last primary round, shows discount tracking vintage almost monotonically. Names whose last round priced in 2026 trade at roughly par. Names last priced in 2025 trade around 5% below. Go back to the 2021 through 2023 vintages and the discounts run 45% to 59%.

Treat that as a pricing rule rather than a chart. What the market marks down is staleness, not illiquidity and not information asymmetry. The discount you are quoted is, to a first approximation, the market's estimate of how wrong the last round has become.

Which is why the headline convention has stopped working. In Q2, 62% of AI transactions on EquityZen traded above the company's last primary round. Not at a discount. Above. When the anchor is 12 months old and the company has doubled, a premium to the last round is still a discount to the next one, and buyers know it.

Two live examples make the point better than the aggregate. PitchBook's own valuation estimate put Databricks at $146.8B as of June 30. 16 days later, on July 16, Databricks signed a term sheet at $188B led by existing investor Coatue, 28% above an estimate that was barely two weeks old. The estimate was sound. It was simply overtaken, which is the whole point. Anduril tells the same story with more force: a $5B round led by Thrive Capital and Andreessen Horowitz that doubled the company to $61B in mid-May, then reported talks at roughly $100B on July 24. A 64% markup in 10 weeks, against the $2.2B of 2025 revenue the company reported in January, with a two-stage structure under discussion in which investors commit to funding a second round within a year at a higher valuation contingent on Anduril hitting certain financial benchmarks.

That structure is the tell. When insiders build the next markup into the paperwork, they are pricing the anchor's age themselves. So should you. Ask which anchor you are discounting from, how old it is, and what the next one will say. The size of the discount is the least informative number in the quote.

2. Access is repricing faster than the assets.

This is the leg most commentary misses entirely.

Anthropic, OpenAI, and Anduril have all reasserted board-approval requirements and declared unauthorized transfers void. Anthropic published a list of 8 firms it had not authorized to trade its shares, with language stating that any unapproved sale or transfer is void and will not be recognized on its books. 7 of the 8 were removed by mid-July. OpenAI has posted comparable language. Anduril warns on its own investor relations page that any offer to invest not coming from or through the company is very likely a scam.

The panic this caused was overblown, and PitchBook says so: voiding shares already held by existing SPVs would be an arduous legal undertaking, and issuers are unlikely to attempt it. The forward effect is what matters. New vehicles will be audited and vetoed. Access concentrates among providers with actual relationships to founders and boards.

Meanwhile the structural workaround keeps growing. Only the first layer of an SPV sits on the cap table and requires board approval, which is why single-layer SPVs have now surpassed direct-to-cap-table trades in value on Caplight this year. And ROFR rates on traded issuers have fallen to roughly 7% in the most recent reported quarter, against a 17.7% long-run average, so companies are exercising their right of first refusal less often even as they tighten who may transact at all.

Put those together. Issuers want to choose their counterparties rather than eliminate the trade. That means every one of these transactions now carries two prices: the price of the shares, and the price of admission. Most analysis prices only the first.

3. The buyer is already in the trade, and is rotating out of the foundation layer.

UBS's 2026 Global Family Office Report surveys 307 family offices across more than 30 markets, with average net worth of $2.7B. It is the closest thing to a census of the natural end-buyer for this paper, and it says three uncomfortable things.

Participation is saturated. Of those surveyed, 65% already allocate to AI, the broadest participation of any theme in the report. Only 7% plan to add over the next 12 months. Infrastructure sits at 37% allocated with 16% planning to add. Power and resources sits at 37% with 15% adding. The marginal dollar is going to the build-out, not to more foundation-model exposure.

The SPV formation data is consistent with that reading. On Sydecar, AI took 40.8% of secondary SPV capital in Q2, but the second-largest sector was aerospace and defense at 17.7%, the picks-and-shovels adjacency rather than another model bet. Where the allocator survey shows intent, the vehicle formation shows the intent being executed.

And the currency underneath is being hedged. Among UBS respondents, 65% expect confidence in the dollar's reserve role to weaken over the coming year against 6% expecting improvement. The dollar is the only major currency where a large share, 47%, describe themselves as overexposed, and 29% have already reduced or are reducing dollar-denominated exposure, with the Swiss franc and euro preferred. Every large-cap AI secondary is a dollar asset sold to a mostly non-US buyer base that wants less dollar. That is a pricing headwind, and it is structural rather than tactical.

A record 60% of family offices plan to change strategic asset allocation this year, against 35% last year. Private equity holds flat at 17% of portfolios, so the multi-year slide from 22% in 2023 has stopped. Share of wallet stabilized. What changed is the posture inside it.

The distinction

Scarcity is a supply condition with a repricing engine attached, not a valuation signal.

Anduril in talks at $100B and Databricks at $188B are not evidence that those businesses became more valuable between May and July. They are evidence that a shrinking number of private names are absorbing a pool of capital that has not shrunk. Those are different facts and they justify different bids. Conflating them is how a buyer ends up paying for scarcity and booking it as quality.

The practical takeaway

Price each leg separately, because each has its own anchor and its own exit.

The public and near-public names trade against the tape or the S-1. Use the SpaceX debut as the base case: strong demand, a first-day move higher, then a correction of more than 20% as float and unlock concerns surfaced. Underwrite the lockup, not the pop.

The scarce remaining privates trade against a round that may be weeks from being superseded. Underwrite the next round rather than the last one, check the vintage of your anchor before accepting any quoted discount, and read two-stage contingent structures as insiders telling you the anchor is already stale.

The absorbed names are gone. The exposure there is the acquirer's stock at the acquirer's multiple.

And price access as its own line. In a market where issuers are choosing their counterparties, a good price from a provider without a relationship may not be a trade at all.

Where does the discount still live? Structurally, in the cohort below the mega-caps, where anchors are older and names are smaller. PitchBook published a list of the next 10 names in venture secondaries with valuation estimates as of June 30, and that list is worth reading as a demonstration of everything above rather than as a price sheet.

4 of the 10 were overtaken almost immediately. Databricks is listed at $146.8B and signed at $188B 16 days later. Anduril is listed at $64.5B and is in talks near $100B. Crusoe is listed at $12.7B and is raising at roughly $30B, about triple its October mark. Polymarket is listed at $10.7B and raised at about $15B with Intercontinental Exchange anchoring. None of that is a failure of the data, which is carefully sourced and clearly dated. It is the anchor-age problem rendered as a table. In this cohort a good valuation estimate has a shelf life measured in weeks, and anyone quoting a discount off one is quoting a discount off a moving object.

What the list is genuinely useful for is direction. Stripe carries a 92% 3-year IPO probability, the highest in the group, on an estimated $171B. Shield AI sits at $13.9B with 37% at 3 years and 80% at 5. Count the verticals rather than the values and the rotation is unmistakable: aerospace, defense, compute, and infrastructure, not another bet on the model layer.

One note from the operating side, anonymized. A roll-up sourcing engine we run against owner-operator businesses depends on third-party contact data. On targets under 50 employees, that data degrades to roughly 60% deliverable accuracy, against the 80% or better needed to protect sender reputation. The fix is verification before the send, plus a hard refusal to sequence any list that falls below the accuracy floor. The parallel to the secondary tape is exact. The expensive error in both cases is committing on a stale input, whether a dead address or a last round that no longer describes the company. Verify the anchor before you commit the bid.

I want to know where your numbers land. If you are marking Anduril, Databricks, Shield AI, or pre-IPO Anthropic or OpenAI paper right now, reply to this email with three things: where you are carrying it, what vintage the anchor is, and what you are being shown on access. I read every reply, and the most useful ones shape what gets covered next.

Prince Capital tracks these shifts across our secondaries practice. If you are pricing entries in the next tier, sizing exposure against a stale anchor, or working out whether you have the access to transact at all, we welcome the conversation.

Reach out at info@theprincecg.com.