57% of Fundrise's Innovation Fund Is Not in Companies. It Is in Ten LPs and LLCs Whose Contents the Filing Does Not Disclose

What the fund actually holds

Four times a year, every registered fund in America has to file a list of everything it owns. Almost nobody reads them. For a venture-style fund sold to retail investors, the NPORT is the only document that tells you what you actually bought, as opposed to what the landing page says you bought.

Here is Fundrise Innovation Fund on June 30, 2026, sorted into what the positions actually are:

Fundrise Innovation Fund, portfolio composition at June 30, 2026

The single most important row is the first one. More than half of a fund marketed as ownership in great private companies is not, on this filing, in any company you can name. It is in ten pooled vehicles.

One number from the same filing sharpens this considerably. SaxeCap Advisors VIII, LP was acquired on December 6, 2023 for $8,534,000 and is carried at $128,169,000 — roughly 15x cost. We wrote that up separately when we went through what the June 30 schedule of investments does to VCX's NAV and its premium to the market price; that piece is about the price of the fund, this one is about the shape of it. Read together they say the same uncomfortable thing from two directions: the single position doing the most work in this portfolio is an LP whose underlying investment the filing never names.

What an "access vehicle" is, and why it is normal

Before the criticism, the defence, because it is a real one.

When a company like OpenAI, SpaceX or Anthropic raises a round, it does not want three hundred names on its cap table. It allocates to a handful of lead investors, and everyone else gets in — if at all — through a special purpose vehicle: a single-purpose LP or LLC set up to hold one stake, with many investors pooled inside it. The cap table sees one entity. The investors see their share.

This is completely standard. Most crossover funds, family offices and, yes, retail-facing vehicles reach the best private rounds this way, because the alternative is not reaching them at all. Nothing about SaxeCap Advisors VIII, LP or AI Access 12, LLC appearing in this filing implies anything improper, and we want to be unambiguous about that before going further.

There are, however, three consequences an investor should be able to state out loud.

One: you cannot see what you own. The NPORT lists the vehicle, not its contents. When 56.93% of the fund sits behind ten such entities, the majority of the portfolio is opaque at the level of the disclosure that exists precisely to make portfolios transparent.

Two: there is usually a second fee layer. An SPV normally charges management fees and often carried interest to its own investors, on top of whatever the fund charges. The NPORT does not disclose the terms of these vehicles, so we cannot say what Fundrise pays inside them, and we are not going to guess. What we can say is that the question is worth asking, and that it does not arise for the 24 positions the fund holds directly.

Three: the valuation chain gets longer. A direct stake in Databricks is marked at Fundrise's estimate of Databricks' value. A stake in an LP that holds a stake in something is marked at Fundrise's estimate of the LP's value, which depends on the LP's own estimate of its holdings. Each link is defensible; the chain is longer than most investors picture.

The names tell you something, and also nothing

Read the ten vehicles as a list and a pattern is obvious:

Four of them have "AI" or an AI-suggestive string in the name: AI Growth GRT, AI Access 12, AI-LLM, RWL AI. One is called Quiet OA Access LP. We are not going to tell you what "OA" stands for or what "AI-LLM" holds, because the filing does not say and inference is not reporting. What the filing does establish is that the fund's exposure is increasingly routed through entities whose names gesture at a theme while disclosing no issuer.

That is the crux. An investor who wants to know whether this fund gives them exposure to a particular private company — the single most common reason someone buys a vehicle like this — cannot answer that question from the fund's own quarterly holdings report, for the majority of the fund.

The rename that is fine, and the rename that is just unexplained

Two positions changed name between the two filings while keeping an identical mark. They are not the same kind of event, and separating them matters.

dbt Labs became Fivetran, and that is explained. At March 31 the fund held "dbt Labs, Inc." at $15,000,018. At June 30 it holds "Fivetran, Inc." at $15,000,018. The two companies completed their merger on June 1, 2026, with Fivetran as the surviving name. A holder's stake converting and the mark carrying across unchanged is exactly what you would expect. Nothing to see, and we checked precisely so we would not report it as something.

DBH1 LP became AI Growth GRT, LLC, and that is not explained. At March 31 the fund's largest single position was "DBH1 LP" at $72,479,780. At June 30, "DBH1 LP" is gone and "AI Growth GRT, LLC" appears at $72,479,780 — the same figure to the dollar. It is now the second-largest holding in the fund at 9.33% of net assets.

There are ordinary explanations for this. A vehicle can be renamed, restructured or rolled into a successor entity without anything changing economically. What we can only report is what the filing shows, plus one detail that is genuinely unusual: AI Growth GRT, LLC is reported with no fair value level at all. Every other position in the filing carries a Level 1, 2 or 3 designation. The fund's second-largest holding, $72.5 million of a $777 million fund, carries none. The prior filing left DBH1 LP's level blank in the same way, so this is consistent rather than new — but "consistently absent" is not the same as "disclosed".

Twelve marks that did not move at all

Across the two quarters, twelve positions worth more than $1 million are carried at exactly the same value, to the cent:

Quiet OA Access LP ($64,613,489), Databricks ($23,256,000), HOF Capital AP Growth ($19,548,977), Epic Games ($19,179,610), Vanta ($10,115,608), Canva ($9,598,642), Inspectify ($5,999,965), Erebor Bank ($4,999,920), Handshake ($3,415,350), Anyscale ($2,493,723), Rhino Labs ($1,414,783) and Immuta ($1,021,600).

This is normal and expected for private holdings. In the absence of a new financing round or a comparable transaction, the standard practice is to carry a position at the last round price. Marking it up on vibes would be worse.

It is worth saying plainly anyway, because of what it implies about NAV. When a majority of a fund is marked at last-round prices that do not move for quarters at a time, the fund's reported volatility is an artefact of the valuation method, not a property of the assets. A private-markets fund that looks less volatile than public tech is not less volatile; it is less frequently re-priced. Anyone comparing this fund's NAV chart to the Nasdaq should understand they are comparing a photograph to a video.

The 8.5% that is not venture capital at all

One row in the portfolio deserves separate attention: $65,925,325, or 8.48% of net assets, in asset-backed securities referencing data centres. The issuers are QTS (five separate tranches), SWITCH, Vantage and EdgeCore, and they are reported at Level 2 — meaning observable market inputs, unlike the private equity.

These are bonds. Specifically, they are securitisations backed by data-centre properties and their lease revenue, and the tranche labels — B, C, D, E — indicate positions in the capital stack, with the letters further down the alphabet subordinated and higher-yielding. The fund also holds $30.5 million of reverse repurchase agreements with Barclays, which is a borrowing against securities.

There is a coherent logic to this: data centres are the physical substrate of the AI theme, and credit against them pays a running yield that private equity does not. But an investor who bought an "Innovation Fund" expecting venture-style equity exposure should know that roughly one dollar in twelve is instead structured real estate credit, held with a degree of leverage.

It is also, of all things, the part of the portfolio that most resembles the rest of Fundrise's business.

What the 24 direct holdings look like

For completeness, here is everything the fund holds under a company's own name, at June 30, 2026:

Note Stripe at $617,652, which is 0.08% of the fund — about eight hundredths of one percent. Stripe is one of the most recognisable names on this list and appears in Fundrise's own communications about the fund.

We want to be careful here, because the obvious inference is the wrong one. This does not mean the fund has minimal Stripe exposure. Any of the ten access vehicles could hold Stripe, and from the filing there is no way to know. The direct line is the visible Stripe exposure, not necessarily the total.

Which is, once again, exactly the finding. In a portfolio that is 57% pooled vehicles, every statement about what the fund owns is a statement about the 43% you can see.

What happened next: a $1 billion offering at 1.76 times this NAV

Ten days after we published this page, the fund registered to sell up to $1,000,000,000 of new shares against the net asset value described above.

On September 9, 2026 Fundrise Innovation Fund filed an automatic shelf registration statement (Form N-2ASR, File No. 333-298833), and on September 10 it filed the prospectus supplement that opens the offering. It is an at-the-market programme: the fund has an Open Market Sale Agreement with Jefferies LLC, which will sell shares into the NYSE from time to time and be paid up to 3.0% of gross proceeds.

The two numbers that matter are on the cover of that supplement, both struck at the close on September 8, 2026:

The prospectus sets a floor rather than a ceiling: "the minimum price on any day at which Common Shares may be sold will not be less than the current NAV per share plus the per share amount of the commission to be paid to the Sales Agent." So the offering is designed to issue at or above NAV, and at the price on its own cover that means issuing at about 1.76 times the accounting value of what the buyer is getting.

In fairness to existing holders, that is accretive to them, and that is not a criticism. Selling shares above NAV adds net asset value per share for everyone already in the fund; a closed-end fund trading at a premium is one of the few places where raising money makes existing holders richer on paper. The question this page is about is the other side of the trade.

Because the NAV being sold against is the one described above. At June 30, 2026 — the most recent composition the fund has disclosed — 72.89% of net assets were Level 3, valued with unobservable inputs; 56.93% sat in ten LPs and LLCs whose underlying holdings the filing does not name; and the second-largest position, $72.5 million, carried no fair value level at all. A buyer at $38.12 is paying a 75.7% premium to a figure that is, by the fund's own disclosure, mostly an estimate of things it does not itemise.

The filing says as much about the premium, in its own words:

"Common Shares of the Fund recently commenced trading on the NYSE and have traded at prices substantially above their NAV. ... Investors purchasing Common Shares while they are trading at a premium may incur significant losses even if NAV is stable or increases. The Fund cannot predict whether the Common Shares will trade at, below or above NAV."

That paragraph is the fund's, not ours. What this page adds is what the NAV on the other side of that premium is made of.

The direction of travel

The single most useful thing in these two filings is not the June snapshot. It is the change between March and June, because three things moved together in one quarter:

The fund grew, and the new money went disproportionately into pooled vehicles marked with unobservable inputs. SaxeCap alone absorbed $71.8 million of the quarter's growth, and AI Access 12 another $25.5 million.

That is a reasonable thing for a fund to do when it is scaling and the direct allocations are hard to get. It is also a shift in the character of the product, from one where a retail investor can read a holdings list and understand it, to one where they mostly cannot. Nothing about this is hidden — it is in a public filing, which is how we found it — but it is not what the marketing describes, and no investor was told it had changed.

What we are not saying

Three claims we are deliberately not making, because the filing does not support them:

• We are not saying the fund is mismarked. Level 3 does not mean wrong; it means valued with unobservable inputs, which is unavoidable for private companies.
• We are not saying the access vehicles are a problem in themselves. They are the industry-standard route into oversubscribed rounds, and without them the fund would hold less interesting assets.
• We are not saying what any vehicle holds. Not Quiet OA Access, not AI-LLM, not AI Growth GRT. The names invite inference. Inference is not evidence, and we have been wrong before by treating a suggestive name as a fact.

What we are saying is narrow and checkable: the majority of this fund is now one disclosure layer removed from the investor, that share rose by more than eight points in a single quarter, and the fund's second-largest position carries no fair-value level. An investor is entitled to know all three before deciding whether the product is what they thought it was.

How to check this yourself, for any fund

This took about twenty minutes, and the method works on every registered fund:

1. Find the fund's CIK on EDGAR. For this one it is 1867090.
2. Filter its filings to NPORT-P. These are quarterly and they list every position.
3. Open the raw filing and look at the invstOrSec blocks: each gives a name, a value, a percentage of net assets and a fair value level.
4. Sort by value and read the top ten names. If they are companies, the fund holds companies. If they are LPs and LLCs, the fund holds vehicles, and the marketing is describing the second derivative of what you own.
5. Compare against the previous quarter's NPORT. The composition drift is usually more informative than either snapshot.

Step four is the one that takes ten seconds and that almost nobody does. It would have answered the central question about this fund at any point in the last year.

Sources

All portfolio figures are taken from SEC EDGAR primary filings, retrieved August 30, 2026:

• Form NPORT-P, Fundrise Innovation Fund, LLC, filed August 28, 2026, accession 0001867090-26-000109, period ended June 30, 2026 — net assets, all 62 positions, values, percentages of net assets and fair value levels.
• Form NPORT-P, Fundrise Innovation Fund, LLC, filed June 1, 2026, accession 0001867090-26-000054, period ended March 31, 2026 — the comparison quarter, including the DBH1 LP and dbt Labs positions.
• Our companion piece on the same June 30, 2026 portfolio, read for NAV and premium rather than composition: VCX's NAV Is $21.70, Not $18.97. The cost basis of SaxeCap Advisors VIII, LP quoted above ($8,534,000, acquired December 6, 2023) comes from the schedule of investments analysed there.
• The Fivetran and dbt Labs merger completion on June 1, 2026 is from Fivetran's own press release.

The bucketing of positions into access vehicles, direct holdings, asset-backed securities and cash is ours, as are the aggregate percentages, the quarter-on-quarter comparison table and the identification of unchanged marks. Every underlying value is the fund's own reported figure.