This is Wazz and Rosie, welcome to Issue 018 of Uninvited.

In this issue

1. Raises. 269 rounds, $5.9B: down 13% on count, 37% on capital. Median cheque up 43%.

2. AI. Round share steady at one in six; dollar share is mega-round timing (58% to 17% inside three weeks).

3. SPV Confessions. We put the public SPV failure stories in one place and opened an anonymous form for the rest.

4. Prediction markets. Polymarket’s venues reject its reported $21B in unison. Kalshi’s venues disagree with each other by 2.7x.

5. Fresh rounds. The secondary market re-prices new rounds within weeks: 25 of 43 already trade below their round, with tails from -60% to +51%.

6. Briefly. SB Energy files, Anthropic’s window opens, OpenAI cuts Cursor off, zero exits.

   Subscribe now

Fewer, bigger deals

269 rounds, $5.948B in the week to Aug 30, against 310 and $9.466B the week before: count down 13%, capital down 37%, median cheque up 43% to $8.00M.

AI: watch the round share, not the dollar share

AI’s share of rounds has held at one in six for six straight weeks.

Its dollar share is mega-round timing: 55-58% in the two early-August weeks carrying Anthropic’s and Databricks’ $5.0B rounds.

Largest AI round of the week: Instinct, $250M. 11 rounds of $100M+

The largest rounds, Aug 26 to Sep 1

• Polymarket and Sunrise are reported, not confirmed

• Chinese deep tech took two of the seven slots. Sunrise builds GPUs in Hangzhou for the Nvidia-shaped hole in China’s supply; AusperBio ($120M Series C, just below the table) is developing hepatitis-B therapeutics.

• Owner is the sleeper. $240M at $2.3B, led by Goldman Sachs Alternatives, for marketing software that fills restaurant seats

• Energy hardware keeps raising. Quaise closed its $180M Series B for millimeter-wave drilling, aiming to open geothermal anywhere on earth, weeks after Base Power’s $1B round for home batteries.

• Ajaib’s $270M Series C is the week’s biggest emerging-market round: Indonesia’s retail-investing app, now past $500M raised.

• Instinct is $250M at $2.5B.

SPV Confessions

Okay, this one is from us. This week we launched an archive of SPV deals gone wrong: confessions.uselegion.com

SPVs made private markets easier to get into. They also made the bad plumbing easier to hide: layered fees, allocations you can’t verify, managers who stop answering. When something goes wrong, people usually don’t say it out loud. Industry is so small that talking can cost you the next allocation. So the same structure gets offered to the next person.

We put the public stories in one place and added an anonymous form for the ones that can’t have a name on them. If you’ve been burned by an SPV, tell us what happened. Read more about this here.

Prediction markets price everything except themselves

Polymarket, reported round: $21B. But every venue that quotes it sits between $11.4B and $14.7B, 30 to 46 percent below. Our proprietary blended valuation puts it at $14.9B.

Kalshi is the opposite failure. Last priced round $22.0B, the $40B is a single report of talks, not announced or confirmed.

Thirty-day ranges: Tessera $14.6-18.4B, PreStocks $20.3-32.9B, the secondary marketplaces $22-40B. Our proprietary blended valuation: $26.5B.

There is also a court risk under looming over both. On Aug 28 the Ninth Circuit ruled that federal law does not shield sports event contracts, where most of both companies' volume sits, from state gambling regulators, directly contradicting an earlier Third Circuit ruling. Splits like that tend to end at the Supreme Court.

The market re-prices new rounds within weeks

43 companies priced a round in the last 90 days and carry a live marketplace mark.

25 trade below their own round, median -9.2%.

The 30-day tails run both ways: HappyRobot -60%, Polymarket -43%, Databricks -36%; Anthropic +51%, Physical Intelligence +27%. Eight below, eight at or above.

Seven things in one minute

• SB Energy filed for Nasdaq (SBE): $138.7M H1 revenue (+66%), ~$439B backlog, three AI data centers under construction, none operating. The reported $50B+ valuation is not in the filing.

• Anthropic’s IPO window opens after Sep 7; prospectus expected for a late-September listing.

• DeepSeek: reported ~$7.4B at ~$74B pre-money. The biggest AI round in the window is not American.

• OpenAI moved to cut Cursor off two weeks after SpaceX’s $60B purchase closed, invoking change of control

• Bybit lists options on its SpaceX and Nvidia perpetuals on Sep 17: a derivative of a derivative.

• The robotics IPO window is repricing the private cohort’s exit math: Unitree, the class of 2026’s benchmark listing, trades about 46% below its day-one open — the multiple every late-stage robotics round (Figure AI $39B, Physical Intelligence $11.2B) is implicitly underwritten against.

• Zero exits this week. The latest remains Cursor, $60.0B to SpaceX, Aug 14.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

This is Wazz and Rosie, welcome to Issue 017 of Uninvited.

In this issue:

1. Raises. Venture dollars fell 48% in the last 60 days and the whole fall is AI (-71%, everything else -6%). Power, chips and fiber tripled their share to 16%.

2. Hugging Face and Nvidia’s private book

3. Anthropic. A $2T IPO valuation, four weeks after a block of its stock was auctioned at a 20% discount.

4. Pre-IPO perpetuals are having a moment

5. Robotics. Unitree listed at about 30x its last private round, then fell 29% in five sessions. 1X faces a Down Round

6. Down rounds. SoftBank’s $6B control talks for 1X, Groq at half its mark

   Thanks for reading Uninvited! Subscribe for free to receive new posts and support our work.

The last 60 days of venture rounds

Disclosed venture rounds over the last 60 days total $103.1B, down 48% from $197.3B in the 60 days before.

Almost all of the fall is AI: AI rounds went from $126.8B to $36.9B (-71%), everything else from $70.5B to $66.2B (-6%).

The same number of large financings got done: 808 rounds of $10M or more in the prior 60 days compared to 822 in the last 60. The $1B+ rounds went from $127B to about $42B, and from 87% AI to about 60% AI.

Power, chips and fiber

Money is moving into what AI runs on. Clean energy, nuclear and fusion, semiconductors, and telecom and connectivity together raised $16.3B, up 57% from $10.4B.

The prior 60 days had one $1B+ round in these sectors, the last 60 had five:

• Joulent $1.75B (baseload power for data centres)

• Eaton Fiber $1.5B,

• Base Power $1B at $13B

• Commonwealth Fusion Systems $1B

• Valar Atomics $1B at $6B.

Two more infrastructure rounds sit under AI:

• Firmus $2B at $10.5B (AI factories)

• Etched $700M at $21B (inference chips)

  

Repricing accelerating

Etched raised a $300M Series C at $10.3B on July 23 and a $700M Series D at $21B on August 18: a doubling in 26 days, and 4.2x since its $5B mark in January.

Other companies that raised in the window at large step-ups from their previous priced round:

• Emergent $300M to $1.5B (5.0x in 176 days),

• Hadrian $1.6B to $7.9B (4.9x in 209 days, raising $1.37B),

• Castelion $2.8B to $13B (4.6x in 254 days),

• Higgsfield $1.3B to $5.4B (4.2x in 214 days),

• Valar Atomics $2B to $6B in 123 days,

• Moonshot AI $20B to $35B in 83 days,

• Firmus $5.5B to $10.5B in 122 days and

• SambaNova $4.0B to $11B in 149 days.

The market is repricing the winners at a record speed, and three of the fastest are chips, power and nuclear rather than model companies.

AI round sizes

AI is 13.9% of rounds and 35.8% of dollars and places 9th on our ranking of median round sizes. The mean is $133.7M, because AI has both a large tail of sub-$5M rounds and about half of the market’s rounds above $1B.

Use the median, not the mean, when comparing AI to other sectors.

Nvidia is reported to be buying Hugging Face for $12.9B, 3.5x its secondary price

The Information reported yesterday (Aug 26) that Nvidia has agreed to buy Hugging Face, the open-source model repository, for about $12.9B.

Business Insider, which first reported on Aug 23 that Hugging Face had hired a bank to gauge interest at $13B or more, reported the same day that the talks are serious but there is no signed agreement and the deal could still fall apart.

Hugging Face last raised in August 2023: $235M at $4.5B from Salesforce, Google, Amazon, Nvidia and Intel. It turned down a $500M Nvidia investment at $7B late last year.

Revenue is about $150M annualized and the company is near profitability, so the reported price is roughly 86x run-rate.

Our blended secondary price, from three venues in the last 30 days, is $3.7B: the secondary market has priced Hugging Face below its 2023 round, and the reported price is 3.5x that.

Nvidia’s private book: Perplexity above $30B, Poolside, and 101 companies worth $3.23T

Three days before the Hugging Face report, The Information reported Nvidia in talks to invest in Perplexity above $30B, up more than 50% on its January round.

On Aug 20 Nvidia agreed to pay Poolside a $6B non-exclusive licence for its model-training system, invest $1B at a $12B pre-money, and hire 109 of its engineers.

In our data Nvidia has backed 108 companies, 101 still private, worth $3.23T combined (whole-company values, not Nvidia’s stakes)

Largest rounds this week

Anthropic: a $2T IPO valuation, four weeks after a block was offered at 20% off

Bloomberg reported on Aug 20 that Anthropic plans to file for its IPO as soon as the end of August and expects to raise more than SpaceX did, the reported values are $100B or more at a valuation of up to $2 trillion.

The Wall Street Journal reported on Aug 25 that the pitch puts the addressable market above $30 trillion

But just a few weeks earlier the same stock was offered at a discount.

In late July the Situational Awareness fund, a forced seller, ran two sales of its Anthropic stake at the same time: 20% below the market with a 12-hour deadline to Sequoia, Greenoaks, DFO and XN, and about 10% below to Citadel and Millennium (TNW, Aug 18). Neither closed.

On Aug 25 the New York Times reported that the SEC subpoenaed Goldman Sachs, JPMorgan, Citi and Bank of America for information on the timing of the fund’s trades and its communications with lenders about leverage, the fund’s July unwind was mostly of public positions, and no wrongdoing has been alleged.

On the two sale days our cross-venue median was $1,425B, so the 20% discount implied about $1,140B, below every daily median of the last two months, the median has since risen 37% to $1,958B.

Pre-IPO perpetuals are having its moment

Entropy raised a $14M seed led by Ribbit Capital plus a $40M HYPE stake and relaunched Anthropic pre-IPO perpetuals on Hyperliquid.

The market opened near $1.96T with about $3M of first-day volume.

Trade.xyz and the Hyperliquid Policy Center filed an SEC comment letter arguing pre-IPO perpetuals should be brought onshore as an IPO-pricing tool

Multicoin’s co-founder petitioned the CFTC to open them to US retail. Stock-related perpetual volume on the top crypto venues went from about $831M a month in July 2025 to about $34B in May 2026.

Unitree listed at about 30x its last round

Unitree, the humanoid-robot maker, listed on Shanghai’s STAR board on Aug 19 at RMB 150.80, closed its first day at RMB 845 (+460%) and closed Aug 25 at RMB 602.80, 29% below the first close.

Reuters put the day-one market value near $50B, roughly 30x the $1.70B Series C of June 2025 in our records. 9.78 million retail investors subscribed. It is the first public price for a humanoid-robot company.

The same week XPeng’s robotics unit Dogotix raised $900M+ at $6.3B in its first round, General Intuition raised at $6B, and Gravis raised a $200M Series A.

Down rounds: SoftBank offers 1X control at $6B and Groq halves

Last fall 1X sought $1B at $10B and raised less than half of it. OpenAI’s acquisition talks fell apart, and OpenAI now has its own humanoid program. And now SoftBank is in talks to buy a majority stake in the humanoid maker at about $6B

The troubles with 1X don’t end here, there were 10,000 NEO humanoid robot pre-orders in five days last October, and no confirmed customer delivery yet.

In September 2025, Groq was an independent chip company valued at $6.9 billion, built to challenge Nvidia. As of August 2026, it is worth half that amount, and one of the investing companies is Nvidia itself. Following a $350 million Series A at a $3.5 billion valuation, led by Disruptive

Of 108 companies that raised a priced round in the last 90 days with a prior mark to compare, 87 re-priced up, 16 flat, 5 down.

Every down round is late-stage, re-prices a two-to-five-year-old mark.

One large accredited-investor marketplace reports its median July trade 7% below the last round, from par in June, with 2024-vintage rounds 17% off.

Also this week

• Secondaries. Record first half, $118B to $124B; single-asset continuation vehicles $34B (+88%); Carta counted 71 tenders and about $3B in H1, a four-year high; Linear’s second employee tender at $2.5B.

• Vanguard is buying Altruist for about $4B (WSJ)

• Instinct, a one-year-old AI assistant, raised $250M led by Index and Benchmark at $2.5B, about 5x its mark of a few weeks earlier. Generalist raised about $200M led by 8VC at $3B.

• Exits. SpaceX closed its $60B all-stock purchase of Cursor. Stripe bought OpenRouter at about $7.5B, 5x its valuation three months earlier.

• Retail access. Robinhood’s RVII opened 10% below issue.

• OpenAI. CFO Sarah Friar told staff the IPO is “2027, or sooner”; the $7B employee tender priced it at $852B

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

This is Wazz and Rosie, welcome to Issue 016 of Uninvited.

Today, we’re diving into what might become the biggest IPO ever, and the very weird problem with trying to put a price on it.

The situation: TL;DR

Anthropic wants to list in October at around $2 trillion. The last time it sold shares, in May, the price was $965 billion.

Normally you would test a price like that against the secondary market, where existing shareholders sell to each other and a real price gets set. In May, Anthropic closed that market to itself. OpenAI polices its own the same way.

So the two most valuable private companies in the world currently have no price anyone outside them can verify, eight weeks before one of them goes public.

In 1889, a cylinder made of platinum and iridium was locked in a vault outside Paris. It didn’t represent the kilogram. It literally was the kilogram.

Copies were sent to national labs, and every few decades they were brought back and compared with the original. Over time, the copies started to drift. On average, they ended up about fifty micrograms heavier.

The weird part was that no one could say for sure what had changed. By definition, the original still weighed exactly one kilogram, no matter what happened to the metal. The mismatch was real, but its source was impossible to pin down. Maybe the copies had gained mass. Maybe the original had lost some. There was no way to tell. All anyone knew was that the system had stopped agreeing with itself.

That’s the real lesson. The issue wasn’t just that drift happened. It was that the system gave no way to say where the drift came from.

Anthropic’s $965 billion valuation is the cylinder in the vault.

Where the $2 trillion comes from

Anthropic hasn’t actually said it wants to go public at $2 trillion.

It filed confidentially on June 1, and its early investor meetings reportedly haven’t included a target valuation or even specific financials. The October IPO timeline comes from backers speaking to the Financial Times. So does much of the excitement around the number. One investor went as far as calling $3 trillion “the incredibly low end.”

But $2 trillion is already being treated like the starting point.

Reuters traced the math. Investors are taking Anthropic’s own forecast of $190 to $200 billion in 2028 revenue and applying a multiple of roughly 42 to 53 times.

There is nothing inherently wrong with valuing a company on future revenue. The strange part is how little there is to check the result against.

Anthropic supplies the forecast. Investors supply the multiple. Its last private round gives you the $965 billion reference point. And since Anthropic shut down secondary trading in May, there is no active market outside that circle saying: actually, we think it is worth something else.

Maybe $2 trillion is right. Maybe it is cheap. But right now, the people producing the number are also the only people with a number.

Anthropic left the market

In May, Anthropic told shareholders that any sale not approved by its board was void. It also blocked the pooled vehicles investors had been using to buy exposure and named eight secondary platforms.

Now only one of those platforms still shows Anthropic.

Investors, naturally, found workarounds. Before the crackdown, an EDGAR search finds 35 registered vehicles with “Anthropic” in the name. None of them were Anthropic. Together they sold $34.7 million of stakes in ten weeks.

Since then, brokers have become the next best source of a “price.” Which is worth remembering when you see Anthropic quoted at $1.5 trillion. That number can simply mean a broker has sellers on his books asking for prices that imply a $1.5 trillion valuation. Nobody necessarily paid it.

The Narrowing

Anthropic does still have a price. It is produced by four venues, all of them crypto exchanges selling exposure to anyone who wants it, and since 8 August they have been the only source of one. Until that date a second class of venue was quoting too, the platforms restricted to wealthy investors, and the two classes were as much as a third apart on the same company on the same day.

Once one market disappeared, so did most of the disagreement.

What is left behind agrees with itself almost perfectly. Those four venues have never been more than 2.3% apart on any day, which reads as precision and is closer to the opposite; they draw on the same thin inputs, so what you are looking at is one measurement repeated four times rather than four measurements taken. Dispersion is information, and losing it is not the same as resolving it.

It is worth being precise about why this was a problem at Sèvres, because the failure was not incompetence. Nobody at the Bureau was lying. The instruments were excellent and the verifications were carefully done. What was missing was a second thing to weigh against, and no amount of care inside the vault substitutes for that.

The cluster is also small enough to see the bottom of. Six private companies, and it does not track who has left; xAI was absorbed into SpaceX in February and went on printing a price for four more months.

There is one recent case of a private number being tested against something that had to settle. OpenAI bought shares back from its own staff at a fixed price, and that price landed within two-tenths of a percent of what the venues were saying, which is a real point in the venues’ favour. Over the twelve days that followed, the venues moved about ten percent and the buyback stayed exactly where it was.

Only one of those two numbers was under any obligation to be a price.

The SpaceX Case

There is exactly one completed experiment in pricing a company this size and it was SpaceX: listed in June at a single fixed number, with no marketed range for investors to push against.

There is one complication. The SpaceX that listed had already absorbed xAI, so the company had changed between its private and public lives. That makes headline valuations messy.

Then the market opened.

SpaceX nearly halved from its IPO price before climbing most of the way back.

So the private market was not wildly wrong. But it was not the final answer either. It took nine weeks of public trading to find out how much disagreement was hiding underneath that one private price.

The arc says two things at once. The private venues were not deluded; the stock has come back within sight of where they had it. And they were not sufficient, because a near-halving and recovery inside nine weeks is what price discovery looks like when it is finally allowed to happen.

Anthropic proposes to skip that and settle 2028’s number today.

One more beat, because it shows what the newly public do next. SpaceX spent its fresh paper on Cursor at a price the private market had not reached.

From $965B to $2T

Most companies behind Anthropic in the IPO queue have decided to wait. They raised more private money instead of testing the public market. The one that pushed ahead without giving buyers what they wanted never priced.

Anthropic is doing the opposite. It is heading toward the biggest valuation jump on the chart, with one of the thinnest markets to check it against.

There is a good argument for that: its $965B valuation is fresh. The worst IPO cuts in the chart came from companies trying to defend private valuations set years earlier. Anthropic’s was set in May.

Fresh helps. It just does not make $965B an independent check on $2T.

If the October IPO is happening, we should see the public S-1 around mid-September. So far, EDGAR has nothing.

A price should survive an argument

The kilogram problem was eventually fixed by getting rid of the one thing nobody could check: the kilogram itself.

The cylinder stopped being the standard. A kilogram became something any lab could measure for itself and compare.

That is basically the problem with Anthropic’s valuation now.

A broker saying sellers want $1.5T is not a trade. A modelled price is not a trade. And $2T based on Anthropic’s own forecast is still a number built on Anthropic’s own numbers. Maybe $2T is right. Maybe it is cheap. But a price gets more useful when someone is allowed to disagree with it.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Wazz and Rosie, welcome to Issue 015 of Uninvited.

Five months ago, Robinhood introduced a new way for people to invest in some of the world’s most valuable private companies. And for a lot of people, it was the first time they could buy into companies they normally had no way of accessing.

The launch was successful. At one point, RVI was trading at almost 3x the value of what it actually owned.

Robinhood did pretty well out of that too. The premium helped it book a $106 million non-cash gain on the RVI shares it was still holding.

Robinhood is listing Fund II tomorrow and that is the reason we’re digging into it today.

Except this one charges 20% on realized gains, takes a 4.5% sales load, and starts with roughly 90% of the money in cash.

Subscribe now

The wall between you and private companies is coming down

The reality for most people over the past few years was this: top companies were staying private longer, and unless you had a lot of money, it was almost impossible to own a piece of them.

Private companies now hold an estimated $10T+ in value. And under current accredited-investor rules, only about 13% of Americans qualify to invest in them directly.

Then, over the past year or so, the wall started coming down:

• The no-accreditation listed vehicle became a category: DXYZ, Robinhood’s RVI, AngelList’s VCX, and Powerlaw’s PWRL trade today, and RVII listing tomorrow.

• The biggest players bought the secondaries layer. Schwab closed Forge in March, Morgan Stanley closed EquityZen in January.

• Wall Street and Washington are moving together. Wellington, Blackstone, Vanguard, and ARK are building private-market products for individual investors. At the same time, the SEC has made the “responsible retailization” of private markets an explicit policy goal, while retirement-plan access and accreditation reform move forward.

For the first time, we don’t have to argue about what happens when retail finally gets into private markets. We can just look.

What Robinhood promised in February

On February 20, Robinhood held an SEC-filed roadshow livestream where Vlad Tenev mentioned the main barriers keeping regular people out of venture capital: accreditation, high minimums, illiquidity, and fees, including 20–30% carry.

Robinhood’s promise was to get rid of all of them.

But we have a completely different picture with Fund II after just five months. Now, it takes 20% of the profits made, and the sales load has gone up to 4.5%. The leverage cap has also been doubled. If you invest $25, $1.125 of that goes straight to the underwriting banks before any actual investing happens. And yes, "no carry" claim is nowhere to be found.

What lists tomorrow

Fund II keeps the Robinhood name but changes almost everything else:

• The new structure brings back carry. RVII is a business development company, a listed vehicle for investing in private businesses. That structure allows Robinhood to charge 20% of realized gains. Fund I does not.

• $20.6M across 80 companies, 79 SAFEs plus one July conversion (the book’s only mark above cost, +$48,377), under ~$175M of incoming cash: the fund opens ~90% cash, with up to 36 months to deploy, charging 2%/yr on all of it from day one, no waiver.

• Buyers start about 7% underwater, by the prospectus’s own math. Of each $25 share, $1.125 goes to the underwriters, leaving $23.875 of NAV. Shareholders also absorb $4.2 million in offering costs, or $5.9 million including organizational costs. That puts estimated day-one NAV at $23.10 to $23.35. Robinhood reimbursed $2.98 million of Fund I’s expenses. Fund II receives no such support.

• Yearly expenses ≈ a third of the portfolio’s entire cost basis: 4.18% all-in before carry ≈ $8.4M/yr against a $20.6M book. The prospectus’s own illustration: $461 of costs per $1,000 over ten years.

• The lock-up self-releases: 30 days, but a close ≥20% above $25 after day 15 frees the sponsor two weeks in. (Fairness: Robinhood cashes out only $9.55M here and ends under 8%)

What’s inside the two funds

We indexed both portfolios from the filings into our database, all 93 companies, so we can put the funds’ own marks next to everything else we track.

RVI holds famous names and marks nearly all of them at exactly what it paid. Ten of eleven audited positions sit at fair value equal to cost, to the dollar. One Databricks markup accounts for 100% of the fund’s $6.67M unrealized gain. NAV moved from $24.05 on March 31 to $25.02 on June 30, the latest number struck (both funds only mark NAV quarterly). Realized gains so far: zero.

RVII looks less like a portfolio and more like a subscription. Exactly two YC batches. 78 identical $250K checks, every single one marked at cost. Nine of the 80 companies have raised priced rounds since the fund’s SAFE, $76.2M combined. Tasklet’s earlier $20M round at a $175M valuation is the biggest thing in the book, and only two companies have any post-money valuation on record at all: Tasklet at $175M and Complir at an earlier $8M.

The entire paper gain across the whole book is $48,377. That’s 0.23%.

What did retail pay for access?

RVI gives us a clean look. Its portfolio is valued once a quarter, so the chart below has exactly two NAV points. And because there are no redemptions and no arbitrage mechanism, nothing pulls the share price back toward what the assets are worth.

DXYZ shows how strange this can get. In fiscal 2025 its NAV rose 209.6% while its share price fell 48.0%. Both numbers are correct. Selling new shares at premiums above 100% added $9.95 per share to NAV.

How the premium turned into $106 million

The central finding comes from Note 4 of Robinhood’s Q2 2026 10-Q, titled “Deconsolidation of RVI.”

After the IPO reduced Robinhood’s ownership, accounting rules required the company to revalue its remaining RVI shares at the market price. Because RVI traded far above NAV, Robinhood recorded a $106 million non-cash gain.

That gain did not come from the underlying portfolio rising by $106 million. It came from the premium public investors paid for the fund’s shares. The same change also allowed Robinhood to begin reporting advisory fees from RVI as revenue.

But financially: the premium is not a side effect of the product. The premium is the product.

From one fund to a franchise

Around March 2027, RVI could become eligible for a short-form shelf registration, which would let it sell new shares quickly whenever it wants. If the shares trade above NAV, it can sell them at that premium, which raises NAV for existing holders and grows the pool of assets Robinhood earns fees on. DXYZ ran this exact play in 2025.

And Robinhood is preparing to scale it. Funds III and IV are already formed, and it’s asking the SEC to let its venture funds invest alongside its European tokenization entities. Tenev has said he wants to build “the world’s biggest venture capital firm.”

So what did five months of RVI teach us?

RVI answered one question pretty quickly: people want this.

Give retail investors a ticker that holds private companies and they’ll buy it. At one point, they were willing to pay almost three times what the underlying portfolio was worth.

That’s where it gets more interesting. RVI made private markets easier to access, but it didn’t make them easier to price.

And Robinhood clearly sees enough demand to keep going. Fund II lists tomorrow, this time with higher fees and 20% carry. Funds III and IV have already been set up.

So we don’t think the question anymore is whether private markets are coming to retail. They are. The question is what retail ends up paying for the privilege.

What to watch next

1. August 13: RVII starts trading. Compare its price to the estimated NAV of $23.10 to $23.35, not the $25 offer price.

2. Around August 29: RVI files its second-quarter portfolio. Watch whether anything besides Databricks gets marked above cost.

3. September 9: RVI’s fee waiver expires and its management fee doubles.

4. The SEC co-investment decision: approval would connect Robinhood’s venture funds to its European tokenization entities through a shared deal pipeline.

5. Spring 2027: an RVI shelf filing would let the fund sell new shares into a premium, the same move DXYZ made in 2025.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Wazz and Rosie, welcome to Issue 014 of Uninvited.

In this issue, we’re going to do a deep dive on Airtable’s sale and the class of 2021: why a company can triple its revenue and still sell for 89% less than its last round, why software is getting hit the hardest, and what the rest of the 2021 vintage is really worth now that the marks are catching up.

Subscribe now

Companies whose last priced round landed in 2021 now trade at a median 48.7% below it.

How far any one of them fell comes down mostly to two things: how aggressively it was priced, and whether it sold software.

Airtable checked both boxes.

It raised at roughly 75 times revenue in 2021, then tripled that revenue, started generating more than $100 million in free cash flow a year, and signed up 80% of the Fortune 100. This week it sold for 89% less than the value of its operating business at that last round.

A company now can turn into a better business and still lose most of its paper value along the way, and that is what happened here, to Airtable and to most of the class of 2021.

Airtable’s acquisition Puts a Price on the Software Reset

Bending Spoons is buying Airtable in an all-cash deal at a $1.285B enterprise value. Counting Airtable’s net cash, the equity value is about $2.25B. Against the $11.735B post-money of its December 2021 Series F, that is 89% off the business and 81% off the equity.

A number like that usually signals a company that fell apart, and Airtable didn't. ARR is roughly $480M growing over 20%, gross margins near 90%, free cash flow above $100M, 500,000 customer including 80% of the Fortune 100. It also never spent about $965M of what it raised, and that unspent cash is a large part of the purchase price.

That is what makes Airtable’s acquisition the right anchor for the wider story.

Airtable was repriced twice. First, the market unwound a 2021 round struck at 75 times revenue. Then AI changed what investors were willing to pay for software itself, weakening the assumption that today’s applications will retain the same moat, pricing power and strategic value tomorrow. AI has not yet eaten Airtable’s revenue, but it has eaten away at the durability investors once assigned to that revenue.

Grown too fast?

Airtable multiplied its valuation 10.7 times in 37 months: $1.1B in November 2018, $2.585B in September 2020, $5.77B in March 2021, then $11.735B that December at $187.28 a share. The last two doublings happened inside one year.

Then the secondary marks went to work. First print below $8B in October 2022, ten months after the round. Below $5B by May 2023. Then eleven quarters pinned between $4.16B and $4.73B, a stable-looking plateau that lasted almost three years. It broke in April 2026 when real order flow arrived and marked the company at $2.4B to $2.6B.

Our blended mark came the closest to the announced price at the time of the acquisition.

Subscribe now

Who actually loses

The preference stack is about $1.5B, every series 1x non-participating, the clean terms 98% of deals now carry. At $2.25B of equity the stack is covered, so the pain lands unevenly.

• Series F bought in at $187.28 a share and takes its 1x back. Money returned, zero return, 4.7 years gone.

• Thrive (Series D at $2.585B) and Greenoaks (Series E at $5.77B) do modestly at best. The deal’s equity value sits below the Series D post-money.

• Common holders, meaning employees and anyone who bought on a secondary desk, absorb the rest.

Under the modeled terms, preferred investors can recover their investment or convert into common, but not both. More aggressive terms would leave less for employees and other common holders.

Who Gets Paid—and Who Takes the Loss

At $2.25 billion of equity value, Airtable can repay its preferred investors, but little more.

Series F is modeled to get its money back after 4.7 years: no loss, no return. Series D and E also recover only modestly at best.

What remains goes to common shareholders: founders, employees and secondary buyers of common stock, who absorb most of the valuation collapse.

Because the preferred shares are modeled as 1x non-participating, investors cannot take their preference and then share again in the remainder. More aggressive terms would have left common holders with far less.

The class of 2021, priced today

Airtable is not an outlier. It is the median outcome. Here is every company in our data with a live secondary mark, grouped by the year of its last priced round.

Nearly half the 2021 cohort (48.5%) has lost at least half its value. One in seven is down more than 75%. Only one in six trades above its last round.

The bigger the 2021 round, the bigger the haircut. PitchBook-derived reporting puts the same cohort at 68% below on an average basis, our dollar-weighted list lands at 68.6%.

Widen it from the ten shown to every 2020 to 2022 company with a billion-dollar round now marked below it, and the scale of the write-down is 245 companies, $750B of peak paper value against $340B today. Fifty-five cents of every peak dollar, gone.

Priced too high, and in the wrong category

Two things predict how far a 2021 mark fell, and the first is simply what the round was priced at. Sorted by the size of the last round, the gradient runs one way and keeps running.

Measured against current revenue, the pattern is the same: valuation haircuts deepen from 26% to about 61%.

But this reflects two things at once: the price paid in 2021 and how much each company grew afterwards. Airtable raised at roughly 75x estimated ARR and is now being acquired at about 2.7x current ARR.

And software paid twice. It raised the bigger rounds, which is the first penalty, but it also fell further than everything else that raised the same amount. Hold the round size fixed and the gap is there in every band.

Among companies that raised under $1B, where both groups raised a median $0.5B, software is down 30.3% against 16.3% for everything else. Between $1B and $3B, both at a median $1.4B to $1.5B, it is 52.1% against 38.8%. Fourteen points and thirteen, on the same money.

This is the AI trade showing up in private prices. Software was hit harder than almost every other category. Fintech companies that lend or move money are down 26.2%, compared with 38.3% for the broader non-software group.

Anything priced on a 2021 growth multiple got repriced. Software got hit twice.

That is why Airtable is a good case study. It raised at the top of the market, in the category investors now trust less, then did almost everything right. Revenue tripled, but the price still collapsed.

The same repricing, in public

Median SaaS EV to forward revenue went from 20.8x in February 2021 to 3.8x in July 2026. SaaS Capital’s index tells the same story, and its June reading of 3.21x was the lowest since 2011.

The cleanest fact in the file: between December 2025 and July 2026 the median multiple fell 32% while median revenue growth held at 12.6%. The public market re-rated the moat, not the numbers, exactly as the private one did. Airtable going from 75x to 2.7x is perhaps just the canary in the coal mine

On 3 February 2026, Anthropic shipped Claude Cowork plugins and Goldman’s software basket fell 6% in a day, with roughly $285B erased inside 48 hours. A second leg followed in July on OpenAI’s Presence launch. Gartner now puts $234B of enterprise application spend at risk from agentic AI through 2030.

“A big chunk of their revenue was enterprise, where it was probably used as a CRUD app with proper controls. That is low hanging fruit for Claude and friends.” Hacker News, on the Airtable deal

AI has hit software valuations before it has hit software revenues.

The clearest examples of SaaS replacement have weakened. Enterprise app counts remain stable, Copilot seats are growing, and major software vendors are still beating expectations. The market is not pricing the death of software today. It is pricing the risk that AI makes its moats and pricing power less durable tomorrow.

537 minted, 35 exits a year

The bigger problem is supply.

The 2021 boom created 537 new unicorns. In 2025, only 30 to 40 exited. In the US, 859 unicorns worth $4.34 trillion are still waiting, and roughly half have not raised in at least three years.

These companies are no longer being repriced through new rounds, but rather being resolved through acquisitions, secondaries and shutdowns.

The next wave will look like Airtable: healthy software companies sold at broken prices because AI is eroding the moats that once made their revenue look untouchable, but these businesses do not have to die for the 2021 valuations to be wiped out.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie and Wazz, welcome to Issue 013 of Uninvited.

In this issue, we’re going to do a deep dive on SPVs: what are they, why people hate them so much, where the bad reputation comes from, and what a better version could look like.

Subscribe now

Pre-IPO used to be a room that regular people couldn’t access. It was reserved for the wealthy.

The current narrative is about democratizing access, letting everyone in. Now, anyone can buy a share of SpaceX or Anthropic. While some problems from the past have disappeared, we now have brand-new ones to deal with, namely more bad actors in the space.

For this issue, we pulled the SEC filings behind thousands of these vehicles, reviewed fee data and surveys, public industry reports, and, of course, browsed Twitter to capture much of the current narrative around SPVs.

SPVs are not niche anymore

SPVs are now the way to get into private markets. This stopped being a niche a while ago.

We pulled every Form D we could find from four of the biggest SPV administrators who have filed with the SEC since 2013: 35,327 vehicles, $27.1 billion of capital, 1.79 million investor positions. The median vehicle is small, about $203,000, held by roughly twenty people. Tens of thousands of tiny pools, each assembled for one deal, each carrying its own terms that nobody is required to publish.

Why some people might hate SPVs?

The reasons are:

• layered fees you never see

• personal access resold for a fee

• bad terms sold to buyers who cannot read them

• no exit until the GP says so, and total losses when it goes wrong

• vehicles claiming shares they do not own

• founders losing track of their own cap table

The fees you see, and the ones you don’t

AngelList, Sydecar and Assure report zero sales commissions on essentially 100% of their Form D filings. They are not free obviously, but a Form D asks about exactly one form of compensation: sales commission.

These deals have setup fees. A percentage of the investment or carry plus a most deal leads also charge a management fee. Neither of these is a sales commission so they never get reported publicly.

Then there is the fee that nobody files anywhere because it is built into the price itself. The person in charge sets the price you buy at. Nothing stops them from building a markup into it.

So where do those numbers come from? Not from a disclosure regime. They come from the platforms themselves.

Carta's data shows the median SPV charged a 1.9% management fee in 2023, with the middle half landing between 1.5% and 2%, and that is before carry.

Sydecar, which has pushed much of the market past the old "2 and 20," reports average carry of about 12% across its SPVs; on the layered secondary-share vehicles that sell you SpaceX or OpenAI, the average runs 2.19% management plus 10.24% carry.

And as the SPV market scaled, it got more expensive, not less.

On Carta, the share of larger SPVs charging a management fee jumped from 41% in 2021 to 67% in 2023

On Sydecar, the share charging both a management fee and carry rose from 37% to 56%.

Layering breaks the math

One SPV buys into another SPV, which buys into a third and somewhere at the top sits the stock. Every layer charges a fee. Every layers fee comes out of the returns. Industry estimates put the combined drag at between 15 and 30 percent of net returns, spread across separate vehicles.

You do not control the exit

The other half of the problem is a question people don’t know they can ask before investing: who decides when the shares get sold? Usually the person in charge decides. We sorted 78 accounts of SPVs going wrong and manager discretion came out as the most common complaint.

Founders should care about this more than they do. Once vehicles start nesting that single line on the cap table can be hiding anyone. You approved one investor. Who is behind them two layers up? You have no way of knowing.

Fees are the quiet failure. Here are the loud ones.

Sometimes the whole structure gives way and investors get to find out what they owned, which is less than they thought.

Each time it was the structure itself letting someone in the middle do something the investor at the end could not see until the money was gone.

The seven ways an SPV fails you

Sort the first-person accounts and they fall into a clean map. Buy an SPV and you are exposed to all seven, usually without a document that names any of them

Programmable SPVs move the rules out of a manager’s hands

The way things are set up now is because of one decision: one person is in charge of everything, including the shares the records, the decisions and the value and the investor just gets a document that says what the person in charge claims is true.

Better disclosure improves the document, it does not change who holds the power dynamics. But a different construction can take those four things out of any single party's hands. Call it the programmable SPV: a vehicle whose terms and transfers live in code on a shared ledger instead of in a manager’s discretion and a private agreement.

Yet nothing shipping today fixes the issue entirely, the underlying shares still sit in a legal entity that somebody has to administer, in a jurisdiction, with a bank account and a signature, and any ledger storing the SPV records are a claim on that entity, not the stock certificate itself.

A rule that executes a distribution faithfully still executes it over whatever the sponsor put in. Giovanni Pennetta sold $10.5M of Anduril shares he never owned. A programmable vehicle would have settled that fraud cleanly, on time, and in public.

The question of who attests that the entity actually holds what it claims is the hard one.

Boring, obvious, and already on the table

None of this requires invention. The SEC’s own Investor Advisory Committee, in September 2025, recommended almost exactly the guardrails this market needs:

• Standardized, layered disclosure of fees and risks, in plain language

• Explicit disclosure of valuation methodology when a sponsor deviates from a third-party appraisal

• Plain disclosure of conflicts and sales compensation at every layer

• Prudential investment limits tied to income or net worth for less-sophisticated buyers

• An accredited-investor test that measures sophistication, not just wealth

• Strict enforcement of the Form D and Regulation Best Interest rules already on the books

The SPV democratized access. That was the promise, and it was real. But access without disclosure is just a faster way to lose money to people you can't see, and almost none of what you would need to see is written down anywhere public.

We’d love to hear your story too. If an SPV ever hurt you, you are in one now and something does not feel right write back and tell your story. You can stay anonymous if you want. Send us a message.

This is Rosie and Wazz, welcome to Issue 012 of Uninvited.

If you know a bit about private markets, you probably understand that a company’s value usually only changes when it gets new investment.

There isn’t much buying or selling between funding rounds.

But this system is starting to change now.

Subscribe now

A market that never used to exist

Private markets have always had secondary markets, but they were mostly invisible. A broker might match a buyer and seller, a fund might sell a position without announcing it, or employees might access liquidity through a tender.

Today those transactions are happening across dozens of venues, each publishing its own view of what a company is worth.

This creates something private markets never really had before: prices that update all the time.

The problem is that these markets don’t agree. So which price can you trust?

Is there a gap in the venues?

Over the past month, we followed 119 companies listed on more than one market during the same week. What we found:

• the middle difference between the highest and lowest price was 1.37 times.

Largest gaps:

• Lovable and Deel traded at valuations almost 2x apart across venues.

• Revolut and Anduril both sat at nearly 1.4x.

The IPO is still the closest thing to the truth

An IPO is one of the few moments where years of private pricing collide with a liquid public market. That makes IPOs one of the only ways to evaluate which pricing sources were closest.

We witnessed the biggest one ever last month, so let’s check its numbers.

In its final week as a private company, SpaceX had fourteen different prices, from $1.37 trillion to $3.3 trillion. On June 12 the IPO priced at $135 a share, a $1.60 trillion valuation. Here’s everyone who tried to call it and who got close:

Not all markets behave the same

Take Anthropic, for example. On four crypto contract exchanges, the price barely moved, stuck within 0.7% of each other and settling at $1.08 trillion. Over-the-counter markets for accredited investors showed a bit more stretch, with prices bouncing between $1 trillion and $1.2 trillion, while the last funding round, was just $965 billion.

If you’re wondering why, the reason is these markets work differently.

Crypto markets move at breakneck speed. Traders jump in and out, closing price gaps almost instantly. Spot a price that’s way off? Someone swoops in, fixes it, and everything snaps back to normal.

Private secondary markets are a different animal. Deals can take weeks to settle. You can’t bet against prices on the spot, either. That kind of friction lets prices drift apart and sometimes they stay apart, at least for a while.

Every source tells a different story

Depending on where you look, you get a different piece of the truth. Trading venues show live bids, asks, and deals happening in real time. Deal records show the final price people paid: extra fees, fund overhead and all. Modeled data does its best to estimate a fair value, usually by mashing together various sources and math.

Stack them up, and trends pop out. Deal records land about 40% above price feeds, mainly because they bake in extra costs. Trading venues, where people want to cash out fast, generally run 13% lower.

For Anthropic right now: crypto-anchored perps point to $1.08T, modeled feeds range from $1.01T to $1.15T, retail asks leap up to $1.64T, and the Polymarket high watermark is $1.93T.

The last round is losing its monopoly

It used to be simple: the price set in the last funding round was the gospel. Now, that number turns stale in a hurry.

Sometimes the market moves sharply lower. Other times, buyers are happy to pay way above the old round.

So no single marketplace gets the final say. If you average prices across venues, you iron out weird spikes and get a truer sense of value. Ultimately, you want to see how closely those blended averages match up with the IPO price.

The gap is narrowing, but it’s still there

Looking at a fixed group of 24 companies tracked by the same two venues since the start of 2024, we see the average price gap drop from 37.5% down to 16.1%.

Now everyone’s watching China’s CXMT - the country’s leading DRAM maker. It just set a record with its STAR-market IPO, which priced at $81B and starts trading July 27.

Five days before the IPO, the perps on three venues all clustered tightly between $428B and $430B (within 0.5% of each other), but that’s still more than five times higher than the IPO price. Maybe the IPO price is just low, but clearly there’s a gulf. Once the shares start trading, we’ll get the best look yet at how useful these secondary market signals really are.

If there’s a sector you want us to run through the database next, private credit, defense, robotics, whatever you’re weighing, reply and name it. We read every one, and the good ones become issues.

-Uninvited

If you’re new here, you can subscribe for free below.

This is Rosie and Wazz, welcome to Issue 011 of Uninvited.

This issue began as a reply. A founder answered a post on Rosie’s personal account:

We treated this as a research prompt.

Yes, Europe is losing the money race

Start with the bear case, because it’s real. A decade ago, North America took 74.4% of the world’s venture capital and Europe 8.8%. Since then the American share has only climbed, to 86.2% by 2025, while Europe, after a brief 14.3% peak in 2022, has fallen back to 7.7%, with the partial 2026 year running lower still.

This is the chart that launches a thousand “Europe is falling behind” op-eds, and on its own terms it’s undeniable. If the game is total dollars raised, Europe is not winning it, and the trend line points the wrong way. Any honest case for European private markets has to start by conceding the scoreboard.

Share

But total dollars is the wrong thing to count, and Europe recently recorded its strongest venture quarter in four years. Here's what that number is made of.

The “gap” is really a concentration story.

Now ask a different question, not how much money, but what it’s buying. Of every venture dollar deployed in North America this year, 85% went into AI. In China, 79%. Two of the world’s three largest venture markets have become, functionally, AI monocultures: one technology, absorbing four dollars in five.

Europe’s AI share? 18%. It is the only major bloc where AI is not a majority of venture capital. The US and China went all-in on a single hand. Europe didn’t.

Where Europe’s money goes instead: everywhere.

So where does the other 82% go? AI is Europe’s biggest sector at 17.7% and then the field fans out: fintech 10.8%, industrial and manufacturing 10.3%, clean energy 7.9% and defense 7.4%, biotech 6.3%, proptech, automotive. AI wins the ranking, but it takes less than a fifth of the share

In a year when the American venture market is, in effect, a leveraged position on one technology, Europe is behaving like a diversified index that happens to be overweight AI.

Subscribe now

There’s exactly one place Europe is concentrating: its own defense.

A diversified bloc that refuses to go all-in on anything, except that isn’t quite true. There is one corner of the map where European capital has stopped hedging and started concentrating, and it is the one the moment demands. European defense-tech funding went from $70M in 2018 to $1.9B in 2025, a 35× rise, and already $5.5B in the first half of 2026 alone.

Widen the lens to strategic technology, defense, space, quantum and cybersecurity together, and the line runs from $350M to $7.1B in seven years.

Just a year and a half ago, hardly anyone had heard of Helsing, a Munich-based defense startup. Now, it’s one of the most valuable private companies in Europe, worth an impressive $14.6 billion. Not far behind is ICEYE from Finland, a company specializing in satellite intelligence, which has also soared to a valuation of $11.1 billion.

Europe is betting big on sovereignty. It shows on the sector map, too: defense has climbed from a rounding error to 7.4% of Europe’s 2026 venture, small beside America’s 85% AI bet, but the fastest-rising slice on the board.

European venture is back. American money owns it.

Look at who backs Europe's best companies and the same American names keep appearing. General Catalyst sits on ten, Andreessen Horowitz and Index on nine, and it doesn't thin out much below them.

The quarter

European startups raised $24 billion in Q2, per Crunchbase. That is the best quarter in four years, up two thirds on the $14.4 billion of Q2 last year and up a third on Q1. Late stage did $12.1 billion, up 90 percent. M&A held too: 154 European venture-backed companies were acquired in the quarter for a combined $11.5 billion and change.

The distribution of venture is not spread evenly. The money is not spread evenly. Britain raised $18.5 billion this year, 36% of all European venture, more than Germany, France, the Netherlands and Spain combined. When we say Europe is back, we mostly mean Britain.

Diversified doesn’t mean dormant.

A portfolio strategy is only vindicated if it returns something. Europe’s does. The bloc counts 112 private companies worth $1B+, worth $518B together, 9 of them worth $10B or more, Revolut alone at $70.7B. And it exits: 219 European companies have reached a liquidity event since 2015, 88 via IPO ($252.5B in value), 71 via acquisition ($38.2B), and 60 that shut down, erasing $2.4B of last-reported value, according to our data.

A choice, not a failure.

US and China have gone all-in. Europe has not. The all-in crowd may well be right: if AI is the race of the century, it goes to whoever commits the most.

But Europe has a war on its border, and the security guarantee it leaned on for eighty years now comes with conditions. You buy the drones before you buy the data centers.

So Europe runs a spread book. It backs AI harder than anything else on the continent, and keeps its deepest reserve for the one race it cannot outsource.

The founder who started this issue said venture doesn’t exist in Europe. Our database counts 112 private companies worth $518 billion, a defense sector that raised $5.5 billion in six months, and 219 exits since 2015.

It exists. But the most active buyers are in San Francisco.

This whole issue exists because one person hit reply. So: if there’s a sector you want us to run through the database next, private credit, defense, robotics, whatever you’re weighing, reply and name it. We read every one, and the good ones become issues.

-Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie and Wazz, welcome to Issue 010 of Uninvited.

Roughly 12% of American adults are now on a GLP-1. The Ozempification of the economy has arrived, and the world is eating, drinking and snacking less because of it.

Public markets already had their turn with the story. This week we looked at the private side, to see where the money went.

Before that, the week’s raises.

The week in private markets

The rounds worth knowing from the past week.

Larger rounds:

• Kling AI, $3.0B at an $18.0B valuation. AI video, backed by Baidu, CITIC Securities and ICBC. The week’s biggest by a distance.

• Quantum Systems, $1.2B at $8.0B, its valuation up 2.3x. Defense drones, with Advent, Airbus and Blackstone.

• Keyfactor, $1.0B at $1.3B. Cybersecurity, led by Insight Partners, Sixth Street and Summit.

• Together AI, $800M at $8.3B, up 2.5x. AI cloud, with General Catalyst and March Capital.

• EnergyX, $225M for lithium extraction, backed by Eni.

• Celea Therapeutics, $180M. Biotech, with Leaps by Bayer, PureTech and RA Capital.

• Yotta Data Services, $150M at $3.9B. Indian data centers.

• Even Realities, $150M at $1.0B. AI glasses, backed by Tencent and Meituan.

• Twelve Labs, $100M. Video AI, with Amazon and Index Ventures.

Notable and smaller:

• Oratomic, a $300M Series A, backed by ARCH and Robinhood co-founder Baiju Bhatt.

• Turing, $78M Series A at $600M. AI.

• Venice, $65M Series A at $1.0B. Private AI, with Coinbase Ventures and Dragonfly.

• 10Beauty, $24M for a robot that does your manicure.

AI, defense and data-center power took most of the week’s biggest cheques, and China printed several of the largest rounds, from Kling to Chang Guang Satellite’s $736M.

Subscribe now

How Private Markets Are Betting on GLP-1s

The public market version of the GLP-1 trade is pretty well understood by now. Alcohol stocks sold off, food companies got repriced, and Eli Lilly became one of the most valuable companies in the world. JPMorgan estimates these drugs could pull $30 to $55 billion a year out of US food and drink spending by 2030.

The private market story is much newer.

Before Wegovy, obesity showed up in venture portfolios only occasionally, with a few companies raising capital each year. Then the category went vertical. Since 2021, obesity has turned into one of the busiest areas in private healthcare, going from a niche corner of biotech to a steady stream of new companies, products, and business models competing for capital.

Delivery out-raised the drugs, $2.6B to $1.7B

Of the 52 companies in our dataset, the 22 operating in telehealth and delivery have raised roughly $2.6 billion combined.

The 19 companies developing obesity drugs have raised closer to $1.7 billion.

Private investors are putting more money behind the businesses that get these drugs into patients’ hands than the businesses trying to invent the next generation of them.

Access, prescribing, insurance coverage, side effects, nutrition support, adherence and affordability matter just as much as the molecule itself, and the companies that manage that relationship between the patient and the healthcare system, tend to capture a meaningful share of the economics.

The most valuable weight-loss companies

The valuations point in the same direction.

The most valuable private companies in the cohort are all delivery businesses: eMed and Virta at $2 billion, Nourish at $1.7 billion, SheMed at $1 billion and Noom at $941 million.

The market is paying up for the front door to obesity care.

Noom remains the best-funded company in the entire group, with $664 million raised to date, yet its last valuation was set back in 2021 at $941 million and hasn't been revisited since.

Who’s developing the next obesity drug

The companies building the drugs themselves have raised less capital and attracted less attention, but they are working on a different problem.

Nobody is trying to prove that GLP-1s work anymore. The race now is around better formats: pills instead of injections, patches instead of pills, and fewer trade-offs for patients.

Verdiva leads the category with $411 million raised around an oral GLP-1 program. Behind it are oral and non-incretin approaches from companies including Marea, Rivus and Syntis, alongside Anodyne’s transdermal patch.

If one of those approaches works, the upside looks very different from telehealth or distribution. Better formats come with patents, exclusivity and potentially an entirely new standard of care.

The latest raises: The easy money phase is ending

The latest rounds suggest capital is still flowing into the category, but investors are becoming more selective about what exactly they are buying.

Some companies are focused on access and care delivery, including eMed, Nourish, Ilant, Coral and Motivia.

Others are betting on the next product cycle entirely: Ambrosia with oral small molecules, Anodyne with transdermal delivery and RenaissThera with oral therapies.

The broad GLP-1 halo that lifted almost everything a few years ago appears to be fading. Companies increasingly need a specific problem to solve and a clear place in the value chain.

The safest bet may not be the drug

For now, private markets seem fairly convinced about where value sits.

In chronic care, the company that owns the patient relationship often ends up owning the economics as well.

That is what makes the delivery layer so powerful. The platforms sitting between patients, prescribers, insurers and pharmacies do not need to perfectly predict which obesity drug wins. If the next standard is an oral GLP-1, a patch, a dual agonist, or something entirely new, they can still be the place patients go to access it.

Drug makers have to be right on the science, the format, the pricing, the coverage and the adoption curve. Delivery companies have a different bet: own demand, then route it to whichever therapy the market chooses.

In a category moving this quickly, that flexibility is starting to look like the asset private investors are really paying for.

Charts by Legion.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie and Wazz, welcome to Issue 009 of Uninvited.

Robotics is having the biggest funding year in its history, and the story underneath it is AI leaving the screen and picking up a body. pulled our data and mapped where the money is moving.

Before that, the week’s raises, and a chat with a French VC in San Francisco who runs twenty agents and lets seventy CTOs vote his deals in.

Subscribe now

The week in private markets

The rounds worth knowing from the past week.

Larger rounds:

• DeepSeek, $7.3B at a $50B valuation. AI, and the week’s biggest by a distance.

• Stegra, $1.6B for green steel.

• SambaNova, $1B at $10B, its valuation doubling in the round. AI chips.

• Dreamplug (CRED), $900M at $4.5B, a 30% down round. Fintech, led by Meta.

• Etched, $500M at $5B. Transformer chips, with Andrej Karpathy, Arthur Mensch and Fei-Fei Li on the cap table.

• CuspAI, $400M at $2.6B. Materials discovery, backed by Bezos Expeditions and Kleiner Perkins.

• Airwallex, $320M at $11B, up nearly 40% from its last mark. Fintech.

• General Intuition, $320M at $2.3B. AI, led by General Catalyst and Khosla.

Outside the AI column:

• Mirendil, a $200M seed at a $1B valuation, led by a16z, to build systems for AI research. A billion-dollar seed, in the same week we spent arguing the seed stage is shrinking.

• A24, $75M at $3.5B, with Google investing. The indie film studio now has a hyperscaler for a shareholder.

• Premier Lacrosse League, $100M Series E, led by Ares.

• Dominion Dynamics, $98M at $400M. Defense.

A $50B AI raise, a billion-dollar seed, and a fintech marked down 30%, all in seven days. The marks are moving both ways this week.

The Uninvited Interview

This week we caught up with Nathan Denier, who landed in San Francisco in January, less than a year out of school, and helps run Batch Ventures. In a city where the information turns over this fast, he thinks that humility is the whole job: “if you’re not ready to learn a new subject every time, you will never be a good VC.”

A few of his sharper takes:

• AI now does 80% of the work, the grind he used to do by hand in Europe. The last 20% is taste, the read on a founder you can’t prompt your way into. Get that wrong and none of the 80% mattered. It also happens to be the only part he ever wanted to do.

• Chasing founder-celebrities is the most overrated move in venture. Everyone checks the YouTube channel and the follower count first, but a big audience has never once made someone a good CEO. Community wins you the tie, he says, not the round.

• His second fund is backed by 60 to 70 CTOs who vote the deals in. If the room likes a dev tool, the fund invests, and some of those same CTOs then go and buy the product. Seventy CTOs shrugging at your tool says more than any memo he could write.

Read the full interview on X.

Follow Nathan at @Natedenier.

AI has found a body

For most of robotics history, progress was real but painfully incremental. Robots did remarkable things inside tightly controlled environments: a factory arm repeating one motion, a warehouse robot following a mapped path, a Boston Dynamics machine with world-class locomotion, but the stack underneath was hand-engineered. Motion planning, perception, grasping, navigation, safety: all built on rules, control systems and deterministic assumptions about the world.

The real world doesn’t cooperate. Boxes sit crooked, floors change, humans get in the way, tools move. Traditional robotics was brilliant inside bounded environments and struggled the moment it left them.

AI changed that. Vision-language and vision-language-action models let a robot interpret its surroundings instead of executing a script: read a shelf, a label, a door handle, a spoken instruction. Google DeepMind’s RT-2 showed vision-language models could be adapted into robot-control systems; OpenVLA showed the same in open source, training generalist policies on large sets of real-world robot demonstrations.

That’s why investors now reach for a new phrase: Physical AI. The hype isn’t only about the bodies. It’s about the model layer that lets machines perceive, reason and act in physical space.

Our data shows the shift plainly: capital is moving into humanoids, robot foundation models, autonomy software, simulation, sensors and full-stack platforms. Outside data points the same way: Crunchbase puts 2026 robotics funding at $18.8B globally, already past the $15B raised in all of 2025, while F-Prime’s State of Robotics calls 2025 a record year, with the caveat that mega-rounds, not broad strength across stages, are driving the gains.

AI is finally good enough to give robotics a software-like learning curve.

1. The U.S. is setting the price; China is setting the pace

The chart isn’t subtle. Over the last two years, U.S. robotics companies have pulled in far more disclosed capital than Europe or China according to our data, the U.S. is where investors are most willing to price robotics like a platform market.

That changes what a robotics company can be. It no longer has to look like an industrial equipment supplier. If investors believe it can own the model, the data loop, the fleet or the operating layer for physical labor, it raises like AI infrastructure, which is the logic behind the mega-rounds for humanoids, general-purpose robot brains and autonomy platforms.

China is a different story. It shows less funding here, but it’s already the world’s most important deployment market. The IFR puts China at 54% of global industrial-robot installations in 2024, with more than 2M robots operational, the largest installed base of any country.

So this isn’t a simple “U.S. wins, China loses” chart. The U.S. holds the capital-markets advantage: bigger checks, bigger platform narratives, bigger valuation step-ups. China holds the deployment advantage: dense manufacturing clusters, cheaper hardware supply chains, faster iteration and a vast domestic automation market. Europe sits smaller but increasingly relevant, especially in industrial robotics: see NEURA Robotics.

That split is the real geopolitical story: the U.S. is trying to own the intelligence layer, China the physical scale. The chart shows where the money is going. It doesn’t show where the robots are already working.

2. The cap tables have changed

The most-active investor list is the clearest sign yet that robotics has crossed into the AI trade. The top names aren’t industrial-automation specialists, they’re AI-era venture firms and strategics: Lightspeed, General Catalyst, a16z, NVIDIA, Lux, Elad Gil, 8VC, Gaingels, Sequoia and Uber. This crowd treats robotics as a platform play, not a hardware bet.

Subscribe now

NVIDIA’s presence is the tell. Robotics runs downstream of accelerated compute, training models, running simulations, processing sensor data, deploying inference at the edge.

Uber’s is another: years around autonomy, mapping, logistics and marketplace labor taught it the value of turning physical-world operations into software-coordinated networks.

The mix says what the market now believes: the robotics companies that win may not just sell machines. They’ll own the data, software and operating systems behind automated labor. That’s the real prize.

3. Themes of Physical AI

The raises chart of the past 2 years isn’t a “robotics is hot” slide, it maps where private capital thinks physical-world AI turns strategic. Roughly $11B in disclosed rounds over six months, clustering into three themes.

Defense autonomy. Shield AI raised $2B at $12.7B, its Hivemind software built to fly drones and aircraft in GPS-denied conditions; Saronic $1.75B at ~$9.3B for autonomous surface vessels; Helsing at the intersection of AI software, drones and European defense urgency. Defense is the shortest path from demo to paid deployment: funded customer, urgent need.

The intelligence layer. NEURA’s up-to-$1.4B Series C: Europe’s largest robotics round — funds humanoids, cognitive robots and its Neuraverse learning platform, with an orderbook north of $1B. China’s entrants sit here too — X Square Robot’s $293M Series B and Galbot’s $364M round (now the country’s most valuable unlisted humanoid maker), both building embodied-AI foundation models. Skild AI’s $1.4B Series C is a pure model-layer bet. The common thread: the market is paying for the general-purpose software layer that makes many kinds of machine useful.

Autonomy in deployment. Wayve raised $1.2B at $8.6B to commercialize its end-to-end driving platform; Zipline $600M at $7.6B on the back of live drone-delivery operations; Apptronik added a $520M Series A extension to scale Apollo.

The through-line: once AI can perceive, decide and act in the real world, the market stops looking like “robotics hardware” and starts looking like labor, logistics and defense. Half these names aren’t robotics companies in the old sense, they’re autonomy companies.

4. The markup chart is the market’s loudest signal

These are AI-platform markups landing on companies that work in the physical world: Apptronik’s 17× seed-to-Series-A at the top, Figure’s step to a $39B post-money the headline mark. The bet is that a few become the control layer for automated labor.

The catch: these marks price in years of execution that still has to happen, certainly not without challenges

The private-market read

Robotics is now one of the few private-market categories with a clean narrative, strategic urgency and a credible AI tailwind.

The labor argument writes itself: developed economies are aging, manufacturing is reshoring, warehouses need automation, defense demand is climbing, and China is automating fast while the U.S. would rather not depend on its hardware. That’s a rare mix of venture interest, corporate demand and government relevance.

But it still isn’t SaaS. Hardware costs, field support, manufacturing complexity, safety, integration, long deployment cycles: the best companies may earn software-like learning curves, but none of them get software-like friction.

So the next phase is the real test. The funding market has voted, now customers have to. The winners will turn AI progress into physical productivity: robots that work longer, fail less, learn faster and cost less per task than the human or legacy alternative.

For now, private capital is making one of its clearest bets of the cycle: AI is leaving the screen, and the next platform may be labor itself.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

A note on the charts: the robotics data comes from Legion (@uselegion on X), a new-media account for private markets.

This is Rosie and Wazz, welcome to Issue 008 of Uninvited.

Different format this week: We're skipping the usual roundup for a single deep dive, because someone published research we couldn't ignore, and we had the data to take it further.

Before that, last week’s raises and a few things from our chat with the guy who sees 3,000 startups a month.

The week in private markets

Larger rounds:

• Odyssey, $310M Series B at a $1.45B valuation, led by Natural Capital. AI world models and real-world simulation.

• Chronograph, $140M growth round, led by Sixth Street Growth. Portfolio monitoring and diligence software for private capital investors.

• Hydra Host, $100M Series A, led by Kindred Ventures, with Founders Fund and Nvidia. Bare-metal GPU platform.

• Ent.AI, $100M seed out of stealth, led by Decibel, with Sequoia and Felicis. AI workspace security.

• Twenty Technologies, $100M Series B at a $1B valuation, led by Accel. Offensive cyber for the US military.

• Atom Computing, $100M Series C, led by Third Point Ventures. Neutral-atom quantum computing.

Seed and smaller:

• Radical Numerics, $50M seed, led by Emergence Capital. Biological AI for drug discovery.

• Traysar, $25M seed. Subterranean defense tech.

• Architect Labs, $24M seed. AI chip design.

• Critical Energy, $19M seed. Geothermal turbines.

Defense, AI infrastructure and quantum took most of the week’s larger cheques.

The Uninvited Interview

11 truths from the man who sees 3,000 startups a month

This week we interviewed Stephane Nasser, who co-founded OpenVC, where roughly 3,000 startups show up every month to raise. He has the dataset, no fund to sell, and no reason to flatter anyone.

A few that stuck with us:

• LPs cannot be talked into a belief they don’t already hold. You build the fund they will buy, or you don’t have a fund.

• The least important part of running a fund is the actual investing. The job is LP relations and admin first.

• If you need a $1M seed to build SaaS in 2026, the problem isn’t the money, it’s the team. Building is commoditized, and capital is retreating to the hard things: deeptech, biotech, robotics, nuclear.

Read all 11 on X.

Now the part we held the issue for.

The deep dive: the $8.4 trillion that’s really $4.6 trillion

Earlier this month, Nnamdi Iregbulem published one of the more interesting pieces of venture analysis in a while, arguing the seed stage is quietly shrinking, and the number that matters is the stock of live startups, how many are actually alive at any moment, and at the seed stage, that stock is shrinking. Post-2021 cohorts are dying off faster than new deals replace them, and the AI funding torrent isn’t enough to offset it.

His central problem was missing data. Nobody publishes the live count of startups, because companies announce their seed rounds and celebrate their Series A but die in silence. So he inferred the exits, fitting curves to graduation benchmarks and assuming that whatever doesn’t graduate eventually leaves. His stock curve is honestly labeled what it is: a model-based estimate, peaking around Q3 2022.

We read it and did the obvious thing: we ran the same question through our own data. Two things came back. The first confirms him, at larger scope. The second is the part his essay explicitly leaves for “part two, coming soon”, the valuation reckoning. We didn’t wait.

Fewer in, more out: the funnel is narrowing

Start with the two flows, companies coming in, companies going out, because everything else is downstream of that.

New companies entering our tracked market peaked in 2021 and have been falling since. Recorded exits, IPOs, acquisitions, and visible shutdowns, have climbed the other way. The two lines are converging.

The weakness in every “startups are dying” take is that death is unobserved, you’re estimating a number nobody reports. When actual new company entries and exits converge, market contraction becomes real

IPOs are out, getting bought or buried is in

Aggregating exits into one line hides the more interesting story, which is how companies are getting out.

In 2021, the exit was an IPO. The window was wide open and companies poured through it. Then it slammed shut. Since 2022, the exit door has been acquisitions, and increasingly, shutdowns. The total exit count is now higher than it was at the 2021 IPO peak, but the composition is upside down: fewer companies graduating to the public markets in triumph, more getting absorbed or quietly wound down.

Subscribe now

The class of 2022 didn’t graduate

The engine driving all of this is the graduation rate, the share of each cohort that makes it to the next stage. Nnamdi cites the well-documented collapse, our own cohort data shows the same fracture.

Seed-to-Series-A conversion held in a comfortable band for a decade, then fell off a cliff for the 2022 cohort, down to roughly 31%, with Series-A-to-B dropping under 20%. The plumbing that used to move companies up the stack seized. Recent cohorts are recovering, though the most recent ones are still partial and the rebound should be read with that in mind.

When fewer companies graduate, more sit in place until they exit. That’s the mechanism. The graduation collapse and the rising exit flow are the same event seen from two angles.

Past peak: the startup pool is draining

Put the flows together and you get the stock. Nnamdi found the seed population peaking in Q3 2022. Our pool of active venture-backed companies, those that raised in the last three years and haven’t exited, held up longer, then turned

Peak 2023, declining since. And the mechanism is the opposite of what you’d guess: the stock isn’t falling because exits overtook entrants, they never do, entrants win every year.

It’s falling because the huge 2021 boom cohort is aging out of the rolling three-year window faster than the post-2022 slowdown can refill it, and because fewer companies are graduating to later rounds, so more sit in place until they quietly time out.

So seeds cracked first, in Q3 2022, and the broader, later-stage population coasted on momentum before rolling over in 2023.

The $8.4 trillion that’s really $4.6 trillion

If the population of companies is shrinking, what is the population that remains actually worth? Because the aggregate private-market cap everyone quotes is built on a convenient fiction: every company is carried at the valuation of its last priced round, however long ago that was.

We stripped that out. Drop any company whose last priced round is more than three years stale, the “zombie unicorns” coasting on a 2021 mark that no one will re-test, and the picture changes hard.

The raw, carry-forward number is about $8.4 trillion. De-horned, counting only companies with a live, recent mark, it’s about $4.6 trillion. The gap, roughly $3.8 trillion, is filled with valuations frozen at the top of the last cycle, attached to companies that in many cases belong to the very cohort now exiting through the side door.

The companies aging out of the active population are disproportionately the ones holding 2021 marks no one is willing to refresh. As the stock shrinks, the share of headline value that’s fictional grows. The market is both shrinking and overstated, with the distortion driven by the companies most likely to exit.

What this means for your book

Nnamdi’s rule was to watch the stock, not the inflow. Ours goes one further: watch the marked stock, not the headline cap.

• The contraction is real, and we’re being conservative about it. The exits we can actually count are already closing on a falling inflow, and the quiet deaths we can’t count only widen the gap. The funnel is narrowing at least as fast as the chart shows, across the whole venture-backed stock, not just seed.

• The exit changed character, not just volume. 2021 was the IPO. The 2020s exit is an acquisition or a shutdown, liquidity by absorption and attrition, not by parade. Underwrite for it.

• The headline cap is built on stale marks. ~$3.8T of the quoted $8.4T is marks no one has re-tested in three years. Some of those companies are fine, some are zombies coasting on a 2021 number. Either way the mark is unverified, so treat aggregate private market cap as a ceiling, not a clearing price.

So, are private markets cooked?

What’s really happening is a repricing: fewer companies, harder exits, and a lot of valuations that haven’t been tested in years. That’s not a market to look away from; it’s a market where knowing the real number is worth more than it’s been in a decade.

Framework and the open thread come from whoisnnamdi’s “Taking Stock of the Seed Stage.” The data, the broader scope, and the valuation cut are ours; charts by Legion (@uselegion on X).

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Wazz and Rosie, welcome to Issue 007 of Uninvited.

This week:

• Washington switched off Anthropic’s two best models on Friday, which is awkward timing for their IPO.

• Binance has won the pre-IPO perp market.

• The secondary desks are quoting half the AI and defense complex, which is what the start of a hot IPO summer would look like.

• And this week’s funding rounds, from a $175M DeFi raise down to a fistful of seed cheques, all landed on the same two things: AI and fintech.

Subscribe now

The race for pre-IPO markets

Binance is currently the main venue for pre-IPO perps.

With SpaceX no longer driving the category and Ventuals, the Hyperliquid-based venue, winding down, the market has narrowed quickly. The recent oracle issue leading to a $1.5M loss on Ventuals’ SpaceX market also showed how hard this product is to get right: these are liquid markets built around assets that were never designed to trade that way.

Hyperliquid, once one of the more visible venues in the category, is now absent from the current leaderboard. That leaves Binance with the deepest OpenAI and Anthropic markets today.

Coinbase is now moving into the same lane. It started with SpaceX and says OpenAI and Anthropic are next.

The question is whether Coinbase can turn brand and distribution into real liquidity. Binance has the lead today, but Coinbase is one of the few venues large enough to make the race interesting.

Hot IPO Summer

If “hot IPO summer” becomes real, the first signs are already showing up in secondary quote flow.

Polymarket, Stripe, Kraken and Kalshi are leading activity over the last 30 days, while OpenAI, Anduril, Shield AI and Crusoe are also seeing consistent interest, according to our data from leading venues.

The million-percent gain

The Cursor deal makes the hindsight hard to ignore.

SpaceX agreed to buy Anysphere, the company behind Cursor, for $60B in stock, turning one of FTX/Alameda’s old private-market positions into one of the clearest missed upside stories of the cycle.

FTX’s estate sold what it had to sell, but the timing looks brutal in hindsight.

Stakes tied to Anthropic, SpaceX, Robinhood, Cursor and Genesis Digital have re-rated meaningfully since FTX first invested, with some of the biggest gains coming after the bankruptcy process began.

That is the strange part of the FTX estate story: underneath the collapse was a book of private-market exposure that, in another context, would now be one of the most sought-after portfolios in venture.

A model ban, right before the IPO

On Friday the US Commerce Department banned non-Americans, including Anthropic’s own staff, from using Fable 5 and Mythos 5, the company’s two best models. It cited export controls and national security. Anthropic shut both models off for everyone to comply.

The timing matters because Anthropic is mid-IPO. Its S-1 leans on safety as the main advantage: the most careful lab earns the most trust and the most enterprise revenue. This is the first time a regulator has used that same positioning against the company.

The question for the IPO is whether investors read this as a one-off or a pattern. If the government keeps restricting Anthropic’s models, the safety moat starts to read as regulatory risk.

The week in private markets

The rounds worth noting this week.

Larger rounds:

• Morpho, $175M at a ~$2B valuation, co-led by Paradigm, a16z crypto and Ribbit. The largest raise a DeFi protocol has done.

• Current, $80M Series E at a $1.5B valuation, led by Springcoast, with a16z and Tiger Global. Consumer fintech.

• Poetic, $50M Series A at a $500M valuation, backed by OpenAI, Kleiner Perkins and Founders Fund. AI agents.

• Minerva, $20M Series A, led by 8VC and Lingotto. Market analysis AI.

Seed and pre-seed:

• Enlaye, construction risk AI, $5M

• Soffi, collaborative dev environment, $4.6M

• Recess, kids activities marketplace, $4M

• Intellectible, AI proposal writing, $3M

• Sika, cross-border settlement, $2M

• BeatpulseLabs, multimodal AI training data, $1.8M

• Onedosh, stablecoin payments, $1M

Most of the week’s capital went to AI and fintech infrastructure.

The Uninvited Interview

How the kids who grew up trading JPEGs will rewrite venture

This week we sat down with Oliver Morris. He DM’d a Bitcoin developer at 12, made Rare Pepes at 13, still brokers six-figure NFT deals at 23, and is now eyeing a move into venture.

To us he reads as a preview of the first generation that learned markets before jobs, and liquidity before cap tables. Hand them a private share they can’t price, sell, or transfer without clearing three gates, and the honest reaction is just: why. They won’t ask permission to enter markets, because they never learned that markets required it.

Read the full interview here.

Follow Oliver at @official_coder.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie and Wazz, welcome to Issue 006 of Uninvited.

Quite a week. SpaceX got advertised next to the sea moss at Erewhon. OpenAI announced its S-1 with a shrug: “we expect it to leak so we’re just announcing it.” Robotics is no longer a niche hardware category. And the GPs spent the week telling LP horror stories on the timeline.

So let’s get into it.

The SpaceX smoothie arc

SpaceX is days from the biggest IPO in history, and the marketing has reached the smoothie aisle.

Erewhon membership costs $200 a year and exists so that a specific kind of person can buy a specific kind of smoothie as a personality. It is, to be fair, exactly the demographic you’d model for this: high disposable income, brand-led, allergic to being left out of anything.

We don’t have a take on whether the SpaceX-as-wellness-good moment is the top of the cycle. We do have a take on what it means when private-market hype reaches the part of LA where people pay $20 to drink charcoal.

Front-running the biggest IPO ever

The IPO is a formality at this point. SpaceX already trades onchain across five venues: Coinbase, Lighter, PreStocks, Ventuals, trade.xyz, with implied valuations between $1.71T and $2.07T. Pre-IPO access has hit its peak: continuous, liquid, public price discovery for a private company. The biggest IPO ever will be the first one where the market already knows the price.

Wait, what is OpenAI doing here

OpenAI submitted a confidential S-1 and announced it on X, on the logic that it would leak anyway. No timing decided yet, though Altman expects it “within the next year.”

“The faster the potential RSI takeoff looks like it could be, the more it could advantageous to delay an IPO,” he reportedly told his staff, because the “technology and the world may change in surprising ways, and there might be good reasons to be a private company during that time.”

Robotics is now a generalist trade

Look at who’s most active in private robotics right now and you won’t find robotics specialists, you’ll find Coatue, Lightspeed, a16z, SoftBank’s Vision Fund, NVIDIA, and even Fidelity. The same capital that powered the AI trade has moved decisively into robotics, and the sector has effectively been absorbed into it. Per PitchBook data, robotics startups have raised over $23 billion in 2026, nearly matching 2025’s full-year total of $26 billion, and on pace to surpass it.

The takeaway for private market investors: robotics is no longer a niche hardware category with its own capital base. And when crossover money, public-markets giants like Fidelity, and the sector’s biggest strategic in NVIDIA are all setting the pace, robotics gets priced like software: faster markups, bigger rounds. Whether software-style capital has the patience for hardware-style development cycles is the open question.

The IPO is lava

Databricks is in talks to raise at up to $175B. The round is a Series M. You heard that right. A company that large, with the listing window this open, choosing a thirteenth private round, is not an accident.

Are companies afraid of the Big 3 drinking the pool? Nobody else is getting in the water this season?

Meanwhile: $500M, no fund

TechCrunch profiled Sabertooth Capital’s Justin Ernest. Nearly $500M into ten companies in twelve months, with no fund. SPVs, single-asset vehicles, a good network, and company-approved allocations in names like Anthropic, Anduril and SpaceX. His read on the moment: “I think this will end up being one of the best vintages of our lifetime.”

A traditional fund takes a year and a half to raise. The trade, evidently, is to skip the back office, keep all the gains.

“Dear LPs, let’s do better,” beg the GPs

The GP group chat moved to main this week. All over the timeline, fund managers are publicly trading stories about the worst LP behaviour they’ve survived.

It started with Rick Zullo of Equal Ventures. An LP asked to pitch at 7am on a Monday, two days after his daughter was born, showed up 45 minutes late, ate a breakfast burrito through the meeting, and passed.

Turns out everyone had one of these saved up:

• The LP who summoned a solo GP to a tennis club 90 minutes outside London, dialed in late from his car, then asked why her father or husband wasn’t funding her. (@sarahdrinkwater)

• The LP who no-showed his own meeting in the Presidio, then emailed “come back” at the end of the slot. (@oanaolt)

• The two LPs who defaulted, sued, lost, and sued again. (@DesireeCachette)

• And the LP who verbally committed, received subscription docs the next day, then pulled out because his astrologer said the stars weren’t aligned. (@pavelprata)

The stars, as of press time, remain unaligned.

The Uninvited Interview

About Zach Holman and Signed.com

After 230 Startup Investments, One Lesson Stands Above the Rest

After writing 230 startup checks, the founder of Signed.com, Zach Holman, believes many founders are optimizing the wrong variable.

A $250k investor who disappears may be less valuable than a $1k investor who brings your first enterprise customer.

In Zach’s words:

“The minimum check size is zero.”

It’s a simple sentence.

But it may explain one of the biggest shifts happening in venture capital. More in this week’s Uninvited Interview.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie, and welcome to Issue 005 of Uninvited. I’m back with and this week.

We had a long week. Another big name joined the IPO race, and yes, I’m talking about Anthropic.

Two of the biggest private companies in the world have now agreed to be publicly priced. That makes OpenAI the only holdout.

But this hasn’t stopped the speculation, as retail has been betting on the price of these companies on Prediction markets and Onchain venues for months.

Plus, we present an interview with Jonas Lamis of Remount, on why the next layer of private markets may be about owning autonomous fleets rather than equity in the AI labs that drive them.

If something interesting happened in your corner of the market this week, hit reply. The best stuff in here usually comes from friends and supporters.

The starting gun

Anthropic’s S-1 landed Monday at a number a few percent off where Forge, Caplight and the rest of the broker network had been quoting it for the last year.

SpaceX is already public-bound. OpenAI is the last holdout. The three are worth roughly $4 trillion at their last marks, and they’ve all been priced by the same handful of brokers working off slightly different assumptions. Two of those assumptions now have to live in a filing.

Where’s the real price?

Depends on which crowd you ask. While the bankers were polishing roadshow decks, the rest of the market had been pricing all three live, and the prices don’t always agree. Three different crowds, three different numbers for the same company: accredited investors buying secondaries; retail buying synthetic exposure Onchain; event-driven traders pricing the IPO itself, in probabilities, on Polymarket and Kalshi.

Legion (@uselegion on X) has been pulling the three together.

SpaceX trades at $1.66T on OTC, $2.26T onchain and $2.28T on Polymarket. Onchain and Polymarket agree almost exactly. Spread: 37%.

OpenAI is the most contested name on the page. $834B OTC, $1.39T onchain, $924B Polymarket. Leveraged retail has it at full moonshot. Polymarket is calling it back down. The perp and the prediction market disagree by 33% on the same company. Spread: 66%.

Anthropic is the most aligned of the three. OTC and Polymarket land in the same place at $1.18T. Only the onchain is high, at $1.60T. The cheque-writers and the event-traders agree. The leveraged crowd is alone in the room. Spread: 36%.

If you want consensus, it’s in Anthropic. If you want disagreement, OpenAI. The name where OTC is most disconnected from where the market is trading is SpaceX.

As the IPO market heats back up, it’s worth remembering what happened last time.

Many of the biggest VC-backed listings have underperformed after going public, with several still trading well below their IPO-day valuations. The window may be reopening, but public markets have been consistently unforgiving to growth stories that can’t defend their price.

Is passive investing going to die?

Probably not but this has been the discussion of the week after Index providers change the rules for the SpaceX IPO

Subscribe now

The Uninvited Interview

Jonas Lamis, Remount

Jonas Lamis runs Remount out of Palo Alto. He’s spent twenty years building software companies and is now building a different kind of vehicle: one that wraps ownership of autonomous infrastructure the way old SPVs used to wrap startup equity. His usual conversation opener with Uber drivers, on the way to a meeting, is asking when they think a car is going to replace them. He likes the gap between their answer and his own.

What’s left to own when intelligence is cheap.

Five to ten months on the clock

Jonas asks every Uber driver the same question. The driver’s answer is usually five to ten years before autonomous vehicles do them out of a job. His own answer is five to ten months. The gap between what he and his peers in Palo Alto see coming and what the people whose work is about to be replaced believe is wider than at any point he’s tracked. Even the Palo Alto tech workers, he says, haven’t fully adopted Claude, Codex or Cursor yet. Let alone the rest of the country.

For an LP, the interesting question is whether that kind of perception gap creates the asymmetric entry point Jonas thinks it does. He thinks it does. Historically, the gap between expert view and broad-market belief is the one that pays.

Physical SPVs

The thesis at the heart of Remount: SPVs don’t need to wrap startup equity. They can wrap ownership of the physical autonomous systems themselves. He’s starting with Tesla’s Cyber Cabs.

Remount’s modeler gives users the ability to project potential returns on their Cyber Cab SPVs.

The next phase could be autonomous trucking fleets and, eventually, humanoid robots.

The structural shift for GPs is real. Instead of betting on equity in a robotics software company, a GP can syndicate a Fleet SPV (say, financing 100 autonomous vehicles) and offer LPs a yield-generating asset class with depreciating physical collateral underneath it.

For LPs, it’s a way out of the standard ten-year venture horizon. Asset-backed, amortised against loans, generating near-immediate yield from real-world operations: cleaning, charging, driving.

The unspoken question Jonas’s structure is answering is the one most LPs are sitting on this year. Where does private capital go if the AI lab equity trade is about to be publicly priced?

The size of the firm is collapsing toward zero

Ronald Coase, the economist who explained in 1937 why companies exist (the short answer: transaction costs), is the reference point. Jonas’s reading is that AI tools are letting individuals coordinate machine labour at a scale that used to require an organisation. The next generation of founders won’t manage employees, departments or office leases. They’ll orchestrate AI agents, autonomous workflows and fleets of machines.

If that’s where this lands, tomorrow’s businesses will have tiny headcounts, massive leverage and near-zero marginal labour costs. Most of the assumptions about hiring scale and organisational complexity that anchor modern VC underwriting will need to be reconsidered.

One layer down from the model layer

Most AI investors are still focused on the model layer: OpenAI, Anthropic, inference, agents, benchmarks. Jonas is looking one layer down. The interesting bet, in his view, isn’t owning equity in the company that builds the model. It’s owning the autonomous system the model controls.

The entry point that makes this more than a thesis is the access. Remount’s structure means ordinary people can buy pieces of autonomous systems for as little as $1,000. There are real caveats: Tesla’s network economics, autonomous utilisation rates, the regulatory rollout. Any of them can break the trade. But if autonomous systems replace meaningful categories of labour over the next decade, ownership of those systems will be one of the bigger economic questions of the decade.

Jonas closes on the line that more or less holds the whole conversation together. Intelligence is about to be abundant. Ownership of the systems intelligence runs through may not be. The editorial above is about two of the biggest private companies finally being priced in public. Jonas’s read is that the next layer of private markets may be about who owns what those companies build, rather than the companies themselves.

Learn more about Jonas and Remount at remount.co

Posts we loved

@saxena_puru (Puru Saxena) ·

“Moral of the story: do NOT chase hot IPOs. Year-1 average drawdown = 55%. Year-1 median drawdown = 54%. (Table: Truist)”

A useful base rate to keep in the back pocket while the IPO trio gets ready to list. Median year-one drawdown on hot IPOs is 54%. The S-1 doesn’t print that number anywhere.

@credistick (Dan Gray) ·

“A portfolio of the top 10 [most-VC-funded companies] would be at –120.5% today, relative to the S&P 500.”

Pairs with the index-fund-that-isn’t section from a different angle. Dan’s argument: the most-VC-funded companies underperform the S&P by 120 percentage points, and the mega-VC 2% fee layer has become mostly a tax on the LP rather than payment for picking outliers.

(@NWischoff on X) ·

“Even A16Z at 75 deals per year only invested in 1.6% of all seed rounds. The top 10 funds do roughly 8% of total seed deals.”

A counter-take for any LP who’s been hearing the “mega-funds have eaten seed” line this fundraising season. A16Z does 1.6%; the top 10 combined do 8%. The data point worth saving is buried lower in her thread: Series A graduation rates have dropped from 54% pre-2020 to 24% since 2023.

@johnfelix123 (John Felix) · 5d

“A lot of the best GPs I’ve backed gave me a strong gut feeling in the very first meeting. Most of the work afterward was really about pressure testing that initial reaction.”

A grace note for the picking conversation that ran through Issue 004. Most LPs and GPs spend their diligence schedules pretending the first-meeting reaction wasn’t already the call. John’s just being honest about it.

A note on the charts and data: most of the pricing numbers came from Legion (@uselegion on X), a new-media account for private markets. If you want more cool charts, make sure to follow Legion on X.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

This is Rosie, welcome to Issue 004 of Uninvited.

You’re getting a longer one this week, because a lot happened. SpaceX filed an S-1 at a $1.75 trillion valuation, which would make it the biggest IPO in history. Polymarket launched prediction markets for private companies the same week, and the traders there are already pricing SpaceX half a trillion dollars higher than the bankers.

While that was happening, I sat in on a private LP/GP dinner about how GPs evaluate opportunities and , who’s joining the byline this issue, sat down with a seed investor who thinks the scarcity equation in venture has flipped.

If something interesting happened in your corner of the market this week, hit reply. The best stuff in here usually starts as a whisper.

Polymarket enters the arena, pricing the biggest IPO ever

Polymarket is pricing the biggest IPO ever. The number is higher than the bankers’.

SpaceX filed its S-1 last week. The company is targeting a Nasdaq listing on June 12 at a valuation north of $1.75 trillion. But traders on Polymarket are already pricing it higher: the median bet has SpaceX touching $2.3 trillion before July 1, and Lighter, the onchain perp venue, has it at $2.43 trillion. That is a half-trillion-dollar gap between what the bookrunners are aiming for and what the prediction markets are willing to pay, before a single share has changed hands.

If Polymarket is right, SpaceX closes day one as the sixth-largest company in the world.

Is the price discovery for private companies about to move out of bank spreadsheets and onto prediction markets? Will LPs start watching Polymarket the way they watch their quarterly statements? SpaceX is the first test case.

How GPs do due diligence

“Where’s the soul in what you’re building?”

Last week I was invited to a private dinner co-hosted by Roundtable Ventures, Talok Capital, and GLO. There were a couple of interesting discussions during the evening, mostly about how GPs do due diligence on founders.

Here are my favorite due-diligence tactics from the night, no names attached:

1. The customer call. Send the founder to a real buyer with no warning. If the buyer wants in personally, the customer did the diligence for you.

2. The small check. Write a small initial check ($25K to $50K) and watch what the founder does with it.

3. The soul question. Ask the founder how their company contributes to a more flourishing human future. The answer (or lack of one) is the filter.

4. The network IC. For deep tech: keep an expert network on call, and run an investment committee of partners from deliberately different backgrounds.

5. The hometown visit. Visit the founder’s childhood hometown. Talk to school friends and parents.

6. The hunger test. Hunger is the easiest variable to fake. Look for evidence the founder has done something in that space before.

The stories behind them:

The story behind the customer call came from a GP who has spent years building a network of Fortune 500 executives that they can call up to take cold pitches from their founders. They told us about sending a calling-infrastructure startup to the CIO of T-Mobile with no warning. Halfway through the pitch, the CIO said he wanted to invest personally. The contract that followed was seven figures. “The customer did the work,” the GP said.

Evan Buhler of Talok Capital had a different approach. He writes initial checks of $25K to $50K and watches what happens: who the founder hires, how fast they come back for more, what they spend the money on. “That’s how you find out who you’re dealing with,” he said, riffing on Kendrick Lamar. The first check, in his version, is the experiment. The investment, if there ever is one, comes later.

On the AI side, another GP had a strict filter for direct investments. They wouldn’t make one, they said, if the founder couldn’t answer one question on the spot: how does what you’re building contribute to a more flourishing human future? Half the pitches they see identify a need and call it a thesis. “Where’s the soul in what you’re building?” they asked. The founder either has an answer or they don’t.

For deep tech, the problem is different. The investor at the table runs pre-revenue companies with no commercial signal to test against, and the foundation models change too fast to bet on long-term. They keep a network of professors and post-docs on call for technical assessments, and an investment committee of four partners with deliberately different backgrounds. They decide by consensus. “You can understand the primitives,” they said. “After that you need the network.”

For evaluating the founder personally, one GP had a method that surprised me. They visit the founder’s childhood hometown when they can. Talk to school friends, talk to parents. Watch how they treat their spouse. “You’ll see their taste,” they said. “You’ll see their values. You’ll see what kind of mission they’re really on.”

Bryce Allen Bagley, who works with Evan at Talok Capital and once ran a research group at Stanford, extended the idea. The most important variable in venture, he said, is also the easiest one to fake: hunger. “There are a lot of people who are very good at acting hungry for success. Until you’ve seen them actually accomplish something in that space, that hunger is borderline meaningless.”

A few of the best methods came out of deals that didn’t work. One GP once backed a founder in the ADHD treatment space who admittedly had very severe ADHD. The takeaway: being the world’s best community member in a category isn’t the same job as running the company. Another talked about a founder who’d been so charismatic he had Cisco signing a term sheet within weeks. The fund still passed. “He can convince us. He can convince Cisco. Can he convince the world? No.”

Evan made the night’s most candid confession too. He’d invested in a startup he would later describe as “an AI-native spreadsheet.” Every signal had told him not to. He did it anyway. “I really liked the founder. I wanted them to win. I went in despite my better judgment because I wanted to make something happen.” The product flopped. The company is pivoting. Evan still thinks the founder will win.

The bet was always the founder.

Getting into rooms like this and writing back what gets said is half the point of Uninvited. Thanks to Roundtable Ventures, Talok Capital, and GLO for letting me in on this one. Hope I keep getting invited :)

The Uninvited Interview

caught up with Andy (@andyhyfi on X), a communications professor turned seed investor. The two of them got into AI, founder risk, distribution, and what’s still worth underwriting in 2026. Here’s what came out of the conversation.

A Seed Investor’s Take on How Scarcity Has Moved

Why the most valuable assets in venture are no longer what they used to be

If everyone can build, what becomes scarce? In a recent conversation with communications professor-turned-seed investor @andyhyfi, one theme kept surfacing: AI may be making product creation abundant while making conviction, distribution, and focus more valuable than ever.

From the conversation:

1. AI isn’t reducing founder risk. It may be increasing it.

AI makes it trivially easy to build something. That makes it harder to know whether a founder has conviction.

“Is this one of twelve half-finished experiments, or the founder’s actual business?” Andy asks.

Historically, building a product required enough effort that effort itself filtered unserious founders. AI destroys that filter. Now anyone can ship prototypes, landing pages, demo products, agent wrappers. Diligence frameworks likely need updating as signal shifts from product velocity to founder focus.

2. “AI psychosis” is real, but not in the way most people think

Andy defines AI psychosis less as delusion and more as builder compulsiveness. Symptoms include endless agent tuning, tab switching between models, constant iteration without shipping, and optimizing systems instead of building businesses.

It’s a fascinating LP lens. A way to make sure your seed portfolio isn’t full of founders who appear hyperproductive but are trapped in AI-enabled pseudo-productivity.

Velocity is not necessarily progress.

3. Distribution has become the moat again

While product creation is getting commoditized, distribution isn’t. If everyone can build quickly, code stops being scarce. Interfaces stop being scarce. MVPs stop being scarce. Only one thing remains: attention.

That means every venture you back must have a viable route to distribution. Otherwise, even a strong product may never find escape velocity.

4. The future may belong to hyper-niche businesses

Andy’s thesis: the internet first enabled niche communities. AI now enables niche monetization.

Instead of broad TAM storytelling like “we sell gardening products,” the better opportunity may be: “we serve Midwest local gardening obsessives.”

This has implications for venture. Traditional VC requires massive TAMs. AI may make smaller, highly profitable niche businesses viable at scale. That suggests an emerging barbell: mega-scale AI platforms on one end, highly efficient niche cashflow businesses on the other, with less room in the middle.

5. Algorithmic fragmentation is becoming a market reality

As a communications PhD, Andy is especially tuned into social algos and their ability to create personalized informational universes.

“Everyone lives in their own tunnels,” he says. That impacts investors. Consensus itself becomes harder to measure. When market participants occupy fragmented information environments, narratives spread unevenly, adoption curves distort, cultural products fragment, and distribution strategies require precision.

As these universes deepen, they may create even more overlooked investment niches.

The bottom line

Scarcity is shifting.

Old scarcity: engineering talent, product building, capital access.

New scarcity: conviction, distribution, focused execution, authentic engagement, economic utility.

The timeline this week

Curated posts we kept open in a tab.

The Biggest PE Firms Are Now Insurance Companies

@fabrizio_builds (Fabrizio):

“The biggest private equity firms in the world are insurance companies now, and that’s by design. The top PE firms stopped competing for pension money and started buying the companies that already hold it.”

Biotech VC Has Quietly Outperformed for Years

@credistick (Dan Gray):

“6% of all VC deals generate 60% of the proceeds. 13% of biotech VC deals generate 62% of the proceeds. Leading biotech firms are sitting on an exit rate of ~40%.”

Why Emerging Managers Need “Gravity Makers”

@pavelprata (Pavel Prata):

“Solo GPs probably won’t need a traditional CoS the same way going forward. I think EMs will increasingly need what I’d call ‘gravity makers.’ In a world where competition for access among funds is insane, the most important thing from day one is building enough ‘gravity’ that the best founders naturally come to you first.”

SpaceX’s IPO Is a Trainwreck

@edels0n (Ed Elson):

“SpaceX-stasy. This IPO is a trainwreck. At a roughly $2 trillion valuation, it will be the largest IPO in history. The go-public may also make Elon Musk the first trillionaire.”

How to Raise (A VC’s Perspective)

@neuralunlock (Arjun):

“Fundraising is a necessary grind. This is usually the part that founders hate the most, and every moment you spend fundraising is less time that goes to building your company. The pitch deck should NOT be an exhaustive list of information about your company.”

Emerging Managers Outperform

@nic_detommaso (Nicole DeTommaso):

“More than half (55%) of top decile VCs in the last ten years are Fund I’s. 73% are Fund I’s or Fund II’s. 73% of the top performing VC funds over the last 10 years have been under $100M in size.”

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

If a friend would get something out of this, send it their way. That’s the easiest thing you can do for us, and honestly, the one that helps the most.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

In this week’s edition:

• Three new retail vehicles for private markets trading 3,000% above their net asset value

• Four onchain venues pricing a pre-IPO SpaceX at $2 trillion

• First interview on Uninvited: Evan Buhler of Talok Capital

If something quietly happened in your corner of the market this week, hit reply. The best stuff in here usually starts as a whisper.

Wall Street keeps opening doors, just not the right ones.

$DXYZ, $RVI, $VCX what do they all have in common? Some of a new trend of public traded funds which offer exposure to private markets.

On paper, that sounds like progress. More access. More liquidity. More ways for everyday investors to participate in an asset class that has long been reserved for institutions and the ultra-wealthy.

But the reality has been messier.

In just three months, these vehicles have traded across an extreme range: from a 3,000% premium to NAV to an 16% discount.

Wall Street keeps building new doors. The question is whether retail is being invited in, or simply being left holding the handle.

The Pre-IPO Onchain craze

SpaceX still has no public ticker, but it already has a live market. Across its onchain multiple pre-IPO venues, traders are pricing the company at roughly $2T+, with premiums heating up fast compared to their offchain counterparts.

The market everyone trades, no one measures

Private secondaries have become too large to ignore and too opaque to trust at face value. Depending on whose lens you use, the market is a $100B venture liquidity channel, a $226B global secondaries market, or part of a much larger private-market backlog waiting to clear.

The Uninvited Interview

Evan Buhler, Talok Capital: creativity, consciousness, and culture.

Evan Buhler runs Talok Capital out of the wreckage of a startup he built after quitting an elite law firm and posted about when it failed.

The startup was a legal-tech company. Big law culture, he says, is don’t post online. A respected partner at Fenwick told him as much on his way out the door: don’t do TikTok, you’ll embarrass the firm and the profession. He didn’t listen. When the company failed in the middle of 2023, he posted about that too. A former client at a family office read the post, reached out, and asked for help on the investing side. That email became the front door of Talok Capital.

Buhler is 32, runs his fund alone, and has built a thesis around “creativity, consciousness, and culture.” He says it on the record, in print, on his timeline, and to LPs in pitch meetings, because the entire premise of his fund is that an emerging GP’s edge is the person running it.

From the conversation:

“DPI is the most overrated thing in venture right now”

In early-stage venture, specifically. The growth-stage version is a different argument and he wants the qualifier on the record. “If you started a pre-seed fund in 2023 and you already have DPI today, that’s a potential red flag, not a credential. It would be outrageous. It would mean you’re maybe good at picking short-term winners. It doesn’t suggest you’re going to be good at running an enduring firm.”

The growth-stage version of the take comes a minute later. “I’m confused what we’re celebrating: someone who’s picking up companies at Series C who has the right network. There’s not much risk at that stage.”

“Institutions are not creative”

The issue is incentive design. A twenty-partner shop has to speak as one voice. “Like a country,” he says. “The president represents us even though we all have dissenting opinions.” A solo GP skips the drag. AI widens the gap. “It’s a net-new equalizer for me. Institutions have the capital to pay someone to do all of this. I get capacity I never had. AI favors creatives, and institutions are not creative.”

LPs should ask GPs: “If you fail in the next five years, what happened?”

The question is in past tense on purpose. The failure has already happened; the GP has to construct a plausible cause of death. The thing to listen for is humility. “Arrogance is really dangerous in investing in super speculative things.”

Reference checks get a sharper variant. The standard ask produces three referrals who’ll say good things, which is to say three referrals who say good things. Evan’s version: ask the GP for someone they’ve worked with who didn’t have a great experience, and why. “It tests humility, self-awareness, learning from mistakes. A lot of this job is social awareness.”

“Don’t wait to deploy”

The operational version is one word: deploy. Don’t wait for permission. If an LP will give you a million, use a million. A hundred thousand, use a hundred thousand. Use your own capital. Get a competitive founder on the cap table for $25,000 if that’s what it takes.

The signal isn’t that the founder accepted a small check. It’s that the GP negotiated below the minimum. “Negotiation is one of the core skills in business. If you’re good at negotiating with less, I want to see what happens when you have more.”

The personal version is the line you remember. “I’m willing to go broke. With my last startup I pulled out before zero. With the fund, I won’t. If you really love something, you’ll go to zero for it.” Then the half-joke that lands: “Which is insane if it doesn’t love you back. But venture capital loves me back.”

The throughline of the interview is the throughline of the fund. He wouldn’t be running Talok if he hadn’t posted about failing in 2023. He keeps posting because that’s the strategy. The partner who told him it would embarrass the profession is welcome to judge.

Find Evan and Talok Capital here:

Evan Buhler

Talok Capital

The timeline this week

Curated posts we kept open in a tab.

Are we in a Zombie VC era?

@pavelprata (Pavel Prata):

“85% of 2017–2018 vintage funds have still not returned capital. The median 2017 fund has returned only $0.34 on $1. The median IRR for that cohort is 11.6%. Only the top 10% of funds from that vintage have crossed 1x DPI.”

Private Markets Are Filtered for the Wrong Thing

@fabrizio_builds (Fabrizio):

“You can’t invest in a US private equity fund unless you prove one thing: a net worth of $1 million or an income of $200,000. That filter controls access to a ~$15 trillion asset class. It isn’t the filter that decides who wins inside it.” “Officially the rule is net worth. Unofficially it’s reputation. Only the unofficial one predicts returns.”

The Outlier Quotient

@adamshuaib (Adam Shuaib): How we spot generational talent

“Truly exceptional founders make you uncomfortable in the first few minutes of meeting them. They are not rude, but there is an intensity to how they think and talk that sits outside the range of a normal professional conversation.”

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

If a friend would get something out of this, send it their way. That’s the easiest thing you can do for us, and honestly, the one that helps the most.

Uninvited

If you’re new here, you can subscribe for free below.

Subscribe now

Welcome to Issue 002 of Uninvited.

In this week’s edition:

• What Anthropic’s lawyers and Charli XCX have in common this week

• Why media people are landing partner seats at top venture funds

• Five things we heard off the record, plus the posts we kept open in a tab

If something quietly happened in your corner of the market and we should know about it, send us a message.

The Anthropic recall

For three years, if you had enough money and patience, you could buy a slice of Anthropic. Not from Anthropic, which doesn’t sell to outsiders. You bought it through what people in this market politely call structures: a fund inside another fund inside a third fund, with each layer taking a tidy cut.

But something has changed.

About $1.5 trillion of those stitched-together positions evaporated in a single memo. The middlemen who built the layers were not invited to the call.

The other AI labs are reading the memo. “I have OpenAI exposure” is going to mean something very different in three months than it did last week.

Five things we heard this week

01 / SECONDARIES

Wall Street is staffing secondaries like a primary market.

In six weeks: Lazard bought Campbell Lutyens for $575M, JPMorgan poached Will Boyle, Goldman and Ardian split $1B of CIC’s US PE stakes. Blackstone’s secondaries arm crossed $100B AUM. CIC, for its part, sold $1B while simultaneously renewing its US allocations, which is the kind of trade you make when diversification is a process and not a thesis.

02 / FUND MATH

TVPI keeps going up. DPI doesn’t move.

Bain, AngelList, TrueBridge and thefundcfo are all printing the same shape this quarter. Pending PE exits sit at $1T. April’s recovery came through secondaries and continuation vehicles. The continuation fund stopped being a workaround and started being the product.

03 / PRIVATE CREDIT

The asset class is having its first real liquidity test.

HNWI inflows collapsed 45% YoY in Q1. Partners Group is publicly weighing gating thresholds. The SEC flagged “emerging pressures.” Dawson Partners closed a $7.7B credit fund and, in totally unrelated news, opened a Hong Kong office. Blackstone’s Joe Baratta talked up Middle East de-escalation as a deal catalyst. Three separate defenses, all executed by managers who were never supposed to need one.

04 / FUND FORMATION

The barbell is the whole shape now.

73% of new VC money in Q1 went to five firms. This week alone, Emerald Lake closed Fund III at $825M, 137 Ventures pulled in $700M+, BMW i Ventures launched a $300M agentic-AI fund. On the other end, three Fund Is closed on pure specificity: Brett Calhoun’s Redbud II at $25M, Adam Koopersmith’s $50M SNAK I, Jordan Wan’s $30M CoFound II. Above $500M, capital is one decision. Below $100M, it’s a thousand specificity bets. The middle is empty.

05 / IPO

Cerebras priced at $185, opened at $400, closed at $86B.

Underwriters do not normally call that an error, but it is one. The biggest US tech IPO of 2026 traded at more than double the offer on the first print. PitchBook had flagged it weeks earlier as “potentially the year’s hottest IPO,” which now reads less like a forecast and more like a memo to the syndicate desks.

Subscribe now

The timeline this week

Curated posts we kept open in a tab.

@uselegion · Embed:

The biggest US IPOs of the past 6 years

“10 of the biggest US IPOs of the past 6 years: Airbnb, Coinbase, Coupang, Snowflake, Figma, DoorDash, Toast, Affirm, Palantir, CoreWeave. 7 trade below their IPO price today. Three trade above: DoorDash, Palantir, CoreWeave.”

Seven of ten. The IPO did not transfer the upside. The pre-IPO holder kept it.

On the Anthropic situation

@lex_node (Gabriel Shapiro) · May 11 Embed:

“There is an active secondary market purportedly in Anthropic stock or derivatives, including on fairly reputable platforms like Forge. Anthropic is calling them out specifically, by name, and essentially saying 100% of these are illegal.

@ankurnagpal (Ankur Nagpal) · earlier today Embed:

“While we do invest in some funds and SPVs, we’ll aim to also invest directly on the cap table of promising startups. We avoid triple layer garbage and SPVs of SPVs and any investments where we can’t verify the underlying.”

Venture and media are merging

@clairejyz (Claire Zau) · 18h Embed:

“Venture and media are converging, and the next generation of great firms will need to show up differently.”

A 350K-follower creator hired as a partner at a multi-stage platform fund, with media buildout in the title. The Item 04 thesis with a name attached.

@fabrizio_builds (Fabrizio) · May 12 Embed:

“Three years ago this would have been a $2M endorsement deal. Today she’s a shareholder. Cap-table seats are the new media buy. Distribution is being priced as equity.”

Fabrizio on Nothing bringing on Charli XCX as shareholder and first global brand ambassador. The deal that used to clear through a marketing budget now clears through a cap table. The marketing line item it replaced was probably bigger.

How GPs and LPs are behaving

@endowment_eddie · 3h Embed:

“Met a privates-focused LP who manages $15B. She didn’t know what gross/net retention was. Deploys $1B annually in tech PE and venture. GPs, KISS method.”

@johnfelix123 (John Felix) · 2h Embed:

“Met a Fund I GP this week with a pretty unusual carry structure: 0% carry below 4x net, 25% above. One of the strongest alignment signals I’ve ever seen.”

Off the record

@endowment_eddie · May 8 Embed:

“The problem is that it’s bubble-ish. We’re still compute constrained.” (Growth GP) “Inflection-stage AI applications feel like suicide.” (VC FoF) “Everyone needs to stop trying to ape Josh Kushner.” (Multistage GP)

Uninvited observation

The Anthropic story and the Charli XCX news both landed last week. The more I look at them, the more they feel like the same story.

A broker used to own the relationship between a private company and the people who wanted to invest in it. An agent used to own the relationship between a brand and the audience it was trying to reach. The company in the middle was paying both of them, and never quite questioning why.

Then Anthropic cut out the brokers. Nothing skipped the agency and gave Charli equity instead.

My guess is we’ll see more of this. The middlemen on both sides are having a worse year than they thought.

If you’re raising, allocating, or sitting on a position that suddenly feels less stable than it did last week, hit reply and tell us what you’re seeing. We read every one.

If a friend would get something out of this, send it their way. That’s the easiest thing you can do for us, and honestly, the one that helps the most.

Uninvited

If you’re new here, you can subscribe for free below.

Every Wednesday going forward, you’ll get one issue with a long read on the week in private markets, five things we heard off the record, and what we couldn’t post anywhere else.

If you have a tip, a story, or a deal that closed without anyone writing it up, send us a message.

Welcome aboard. Let’s dive in.

Amazon went public at $438M. Today companies hit $500B before retail can buy a share.

Amazon IPO’d in May 1997 at a $438M valuation. Eight years from founding to public, with an S-1 the whole market could read on the way in.

That used to be the deal.

In 2024, the median tech company took twelve years to reach an IPO. There were 196 IPOs on US exchanges that year, compared to 651 in 1999. Meanwhile, the largest private companies in the world cleared $200B and $300B without ever filing.

The audience for the story these companies tell has moved to a different room.

The S-1 has been replaced by a data room. The roadshow has been replaced by a tender offer with a $25M minimum. The earnings call has been replaced by a board update sent to a list. The diligence is real, the valuations are real, the cash flows are real. None of it is public.

Their LPs know the numbers. Their employees know whether the recent tender priced up or down. Their late-stage investors know whether the round just closed flat. Everyone else is reading press releases and guessing.

That asymmetry is what Uninvited is built to close.

We started writing because the coverage we wanted to read didn’t exist. Most private-markets coverage either sells subscriptions or republishes GP-approved press releases for free. There’s a third version, which is the one we are trying to write. The one an emerging manager, an allocator, or a curious operator would want to read.

Five things we heard this week

01 / FUND FORMATION

Solo capitalist spinouts are at their 2021 peak.

Most institutional LPs are at or over target on venture and trying to reduce line items, with appetite for new emerging-manager strategies described privately as “fatigued.” The supply curve and demand curve are visibly mismatched, and nobody on the LP side seems in a hurry to close the gap.

The cultural read: this is the first cohort of GPs spinning out who didn’t raise during the boom and don’t have boom-era LPs as anchors. They are pitching ZIRP-era allocators who are themselves trying to convince their boards that venture as an asset class still works. Two layers of hesitation, one round to close.

02 / LP BEHAVIOR

85% of LPs have killed an allocation over a GP’s social posts.

A survey by Equidam: 98% say firm-personnel social media profiles are important when investing. 85% want to see personal statements on geopolitical and societal issues. 85% have walked away over a post.

The cultural read: this is the most consequential change in fundraising etiquette since the LP day became a calendar fixture. GPs spent the last two years deciding whether to be publicly opinionated. The data says LPs already decided for them. Either you’re legible online with a worldview LPs can underwrite, or you’re filtering yourself out before the meeting.

03 / FUNDRAISE TACTICS

GPs who hedge on fund size are losing rounds.

Heard repeatedly from the LP side this week: “Your fund size is your strategy. Talking in percentages is ridiculous.” The hedge reads as a tell that the GP doesn’t know their strategy. The advice from one allocator: pick the number, defend it, move on.

The cultural read: hedging on fund size used to signal flexibility. In a market where LPs are over-allocated and over-pitched, the same hedge now reads as indecision. The cohort raising fastest is naming the number on the first page of the deck and treating it as load-bearing, with no negotiation range underneath.

04 / DEAL FLOW

OpenAI and Anthropic are both selling AI agents into PE portfolio companies.

OpenAI’s head of private equity left for Google last week. Both labs are now positioning AI tooling for PE operating partners as the next major enterprise channel. Semafor reported the partnerships 20 hours ago.

The buyer of the next decade is a PE operating partner with a fiduciary mandate to extract more EBITDA. Operating partners are also the easiest sell in enterprise software because their job is to find the next 200 basis points of margin, and AI agents are the cheapest path. This realigns where the AI labs go for revenue. It also means the PE roll-up playbook is about to absorb a lot of horizontal SaaS that VCs sold in 2021.

05 / MARKET STRUCTURE

The 10-year fund is structurally dying.

A new analysis from Odin argues the math on holding portfolio companies past year ten flattens or declines, suggesting the standard 10+2 structure is now actively destroying returns past its midpoint. Combined with the secondary boom, LP appetite is shifting toward shorter-duration, faster-DPI structures.

The fund structure itself is becoming a vintage signal. Funds raised 2014-2018 are now in the part of their lifecycle where holding longer hurts. LPs are doing the math and asking GPs about exit pressure earlier than they used to. Expect more 7+1 structures, more continuation funds, and a quiet end to the “we hold great companies forever” pitch.

Uninvited observation

The GP timeline is the new data room.

There’s a sentence we heard three different ways this week, in three different conversations: “We can’t underwrite a GP we can’t read.”

By “read” the LPs meant something specific. They meant the person, the worldview, and how the GP behaves when something inconvenient happens to a portfolio company. The deck barely registers.

The implication is that the most consequential fundraising work a GP does in 2026 is the ongoing public record they leave on the timeline, which any LP can pull up at any point and use to underwrite or kill the relationship.

This makes private markets the new public markets in one specific way: the social cost of what you publish is now priced into your raise.

For emerging managers, that’s a tax on inauthenticity. For established firms, it’s a tax on the marketing-comms layer that used to scrub partner posts before they went live. The new equilibrium is strange. The GPs being chosen are the ones willing to be specific in public. The GPs being passed on are the ones being too careful.

We’ll write more about this in Issue 002.

If you are raising right now, allocating right now, or watching from the sidelines wondering if this is the bottom or the middle, reply to this email. Tell us what you’re seeing.

Forward this to one person who needs it.

Issue 002 lands next Wednesday.

Uninvited