Situational Awareness: What Assets Are Left After the Citadel Sale

- The 13F headline is not a stock portfolio. Situational Awareness’ last filing, for 31 March 2026, reports $13.68 billion — but 72% of that is the underlying value of options: $8.46 billion of puts and $1.36 billion of calls. The actual common-stock book was $3.86 billion.
- Split properly, the filing shows a coherent structure: long the infrastructure, put against the large-cap names. The stock book is Bloom Energy, SanDisk, CoreWeave, IREN, Core Scientific, Applied Digital, Riot and CleanSpark — power, storage and compute hosting. All 11 puts are on large-cap technology — the VanEck Semiconductor ETF, Nvidia, Oracle, Broadcom, AMD, Micron, TSMC, ASML, Intel, Corning and Infosys, three of which are not chip companies. Nvidia common stock: $497,912.
- What is left is mostly private, and Anthropic is about half of the fund’s reported total assets — not half of the private book, which is a different denominator. The Wall Street Journal put the stake near 20% of a ~$20bn fund in June, which works out at about $4 billion; after the July sale Bloomberg’s figure is about $5 billion against ~$10 billion of assets. Eight weeks apart, from different unnamed sources, one reported and one derived — not directly comparable, and not demonstrably contradictory either.
- Neither figure makes it large inside Anthropic. Against the $965 billion post-money valuation of Anthropic’s Series H in May 2026, $5 billion is 0.52% of the company and $4 billion is 0.41%.
Leopold Aschenbrenner’s fund sold most or all of its public equity book to Citadel on 30 July 2026 — accounts differ on which — and kept the private one. Before asking what is left, it is worth being precise about what was there — because the number everyone is quoting for the public book is not what it appears to be.
Start with what the fund actually filed.
What was there to sell?
The fund’s most recent legally required disclosure is a 13F for 31 March 2026, filed on 18 May. It reports $13.68 billion across 42 line items — and that headline number is not a stock portfolio.
Only $3.86 billion of it is common stock. The rest is the value of the shares underlying options positions: $8.46 billion of puts and $1.36 billion of calls, together 72% of the reported total. A 13F lists options by the market value of the underlying, not by premium paid, so adding those lines to the stock lines produces a figure that describes nothing.
Split properly, the filing says something much more specific.
The stock book: $3.86 billion, and 98% of it is infrastructure
The ten largest of 26 common-stock positions, together 94% of the $3.86bn; the remaining 16 come to $234m.
| Common stock | Value | Share of the stock book |
|---|---|---|
| Bloom Energy | $878.7m | 22.8% |
| SanDisk | $724.4m | 18.8% |
| CoreWeave | $556.1m | 14.4% |
| IREN | $401.0m | 10.4% |
| Core Scientific | $389.1m | 10.1% |
| Applied Digital | $320.0m | 8.3% |
| Riot Platforms | $142.2m | 3.7% |
| CleanSpark | $104.5m | 2.7% |
| Solaris Energy Infrastructure | $62.5m | 1.6% |
| T1 Energy | $43.9m | 1.1% |
Power, storage, and the companies that host compute. Bloom Energy makes fuel cells; SanDisk makes memory; CoreWeave, IREN, Core Scientific, Applied Digital, Riot and CleanSpark run or convert data centres; Solaris Energy Infrastructure supplies power equipment and T1 Energy manufactures solar modules, cells and batteries.
Not quite everything. Alongside them sit token stock lines in exactly the names the fund holds puts on — AMD at $20.2m, the VanEck ETF at $10.3m, Intel at $8.9m, TSMC at $7.6m, ASML at $6.1m, Micron at $5.9m, Corning at $0.7m and Nvidia at $497,912. Together those are $60m, 1.6% of the stock book. It is tempting to read them as share legs attached to the options, and the size and the exact overlap of underlyings both point that way — but a 13F cannot show it. The filing proves only that the fund held small amounts of these shares and options on the same names; it does not link them into one trade, and this article does not claim elsewhere to know the fund’s intent from a 13F. One more holding, ProPetro at $13m, fits neither group.
The puts: $8.46 billion notional, mostly against large-cap chips
Every put in the filing — all 11 of them:
| Put, by underlying | Notional |
|---|---|
| VanEck Semiconductor ETF | $2,042.7m |
| Nvidia | $1,568.3m |
| Oracle | $1,072.9m |
| Broadcom | $1,006.2m |
| AMD | $969.2m |
| Micron | $583.7m |
| TSMC | $535.1m |
| ASML | $494.1m |
| Intel | $159.1m |
| Corning | $21.0m |
| Infosys | $6.8m |
| All 11 puts | $8.46bn |
Mostly semiconductors, but not only: Oracle is database and cloud software, Infosys is IT services, and Corning makes glass and optical components. Together those three are $1.10bn of the $8.46bn. The accurate description is large-cap technology weighted heavily toward chips, not “the semiconductor complex”.
That is the shape worth reading. On the evidence of this filing the fund was long the picks-and-shovels — power, storage, hosting — while holding large put positions against the big listed technology names. Whether those puts were a hedge on the long book or a directional view is not something a 13F can tell you: it reports positions, not intent, and it does not report the premium paid, the strikes, the expiries or any short positions at all.
One thing it settles, with a qualification. Our earlier article said the fund held no meaningful position in Nvidia and the other obvious large-caps. On common stock that is right: Nvidia’s line is $497,912. But it is not right to say there was no Nvidia position at all — a $1.57bn notional put is a substantial derivative exposure to Nvidia, even though it is neither $1.57bn of capital nor necessarily a bet against the company. A first pass at this filing that ignored the put/call field made this look like a simple contradiction, and led to a correction on that article which has itself now been reversed.
What a 13F does not show
Worth stating plainly, because it bounds everything above. A 13F is a regulatory filing, not an audited statement, and the SEC does not vouch for its completeness. It excludes short positions, cash and borrowings, most derivatives outside listed equity options, private holdings, and therefore total assets under management. It is a partial view of one side of one book on one date.
So the $3.86 billion stock line is not “what the fund was worth”, and the $13.68 billion headline is not “what there was to sell”.
So what is actually left?
Three things, on the reporting — and the reporting does not fully agree with itself.
The private positions. Anthropic, the chip designer MatX, and the data-centre firm Fluidstack. Multiple outlets agree none of this was part of the Citadel transaction.
Possibly a residual public book. Here accounts diverge: the Wall Street Journal and TechCrunch describe the majority of public equities being sold, Bloomberg and Business Insider report a small remaining public position with the manager continuing to trade, and Axios describes the entire public portfolio going. That conflict is unresolved.
Whatever the sale realised, net. The price has not been reported, and because the book was leveraged, proceeds may have gone to repaying financing or meeting redemptions rather than sitting as cash.
How much is the Anthropic stake worth?
Two figures circulate, at two different dates. One is a reported valuation and one is derived from two rounded numbers:
- In June 2026, the Wall Street Journal reported the stake at around 20% of the fund, when assets were about $20 billion — implying roughly $4 billion.
- After the July sale, Bloomberg’s figure, via TechCrunch, is about $5 billion, against total assets of about $10 billion.
Both come from unnamed sources and neither from the fund, and they are eight weeks apart. That makes them not directly comparable rather than contradictory — there is no evidence the stake actually rose from $4bn to $5bn in the interval, and the $4bn is itself a product of two rounded figures multiplied together, not a reported valuation. What they do share is that Anthropic is now roughly half of the reported total, and that it is the reason the fund is described as moving toward a private holding structure. Note the ceiling on that: half makes it the largest single asset, not a majority, and both the numerator and the denominator are estimates from unnamed sources.
What does that stake mean at Anthropic’s valuation?
Anthropic’s Series H in May 2026 valued the company at $965 billion post-money. Divide the reported stakes into it:
| Stake figure | Basis | Implied share of Anthropic |
|---|---|---|
| $5.0bn | reported valuation, after the sale | 0.52% |
| $4.0bn | derived: ~20% of a ~$20bn fund, in June | 0.41% |
Both land at roughly half of one percent of the company. That is the arithmetic worth holding onto: a position large enough to define what a multi-billion-dollar fund now is, is a fraction of a percent on the cap table of the company it is invested in.
This is an inversion, not a disclosure — nobody publishes what percentage of Anthropic the fund owns. And if the shares were bought at an earlier, lower valuation, the cost basis differs from the mark; no source gives the entry price.
How did the assets move?
The reported sequence, all of it from unnamed sources:
| Reported assets | |
|---|---|
| Peak | around $45bn |
| Recent months | around $20bn |
| After the Citadel sale | around $10bn |
Against that, the first-half return was reported at 439%, and the fund is reported to have begun in September 2024 with roughly $225 million of seed capital from Patrick and John Collison, Nat Friedman and Daniel Gross. That last figure is a press number and no filing confirms it — the fund’s first 13F, for December 2024, reports about $254.8 million of listed equity, which is a different thing measured at a different time.
Both things are true at once: reported assets more than halved in weeks, and the fund is still enormously larger than it started. Note the wording — a fall in reported assets is not the same as an investment loss. Selling a leveraged book shrinks it, repaying financing shrinks it, and redemptions shrink it, and none of those is the same as the portfolio going down. Nobody has published which of them this was.
What would actually settle this?
The Q3 13F, in November. The Q2 filing due mid-August covers 30 June, a month BEFORE the sale, so it cannot show what left — it only updates the pre-sale book. November’s is the first filing that can show the aftermath, and it will still only cover listed equities.
Any filing describing the sale itself. None has been identified. Everything about the Citadel transaction currently rests on press reporting from unnamed sources.
Whether the fund raises new capital. Reporting indicates it is seeking it. Investors deciding whether to fund a vehicle whose largest single asset is one private stake is a more informative signal than any valuation estimate.
Anthropic’s next round. The single asset that now defines this fund is marked to a private valuation, and private valuations move in steps, not ticks.
Sources
Each was checked on 1 August 2026.
| Source | What it supports here |
|---|---|
| SEC EDGAR filing index, Situational Awareness LP (CIK 0002045724) | The 13F-HR for 31 March 2026 filed 18 May, the absence of any Q2 2026 filing, and the split of its 42 line items into $3.86bn of common stock, $8.46bn of put notional and $1.36bn of call notional across 29 issuers |
| TechCrunch, 30 July 2026 | The retained private positions including Anthropic at a reported $5bn, MatX and Fluidstack, the assets falling from a ~$45bn peak to about $10bn, and the 439% first-half return |
| Wall Street Journal profile of the fund | The June 2026 reporting that the Anthropic stake was around 20% of the fund, which is the basis for the derived $4bn figure, and Jane Street as an investor |
| Business Insider via Yahoo Finance, 30 July 2026 | Anthropic’s $965bn Series H valuation of May 2026, that the private portfolio was not sold, and that the manager kept a small part of the public book and will continue trading |
| Proactive Investors, 30 July 2026 | The retention of the Anthropic stake and the transition toward a private holding structure. An earlier version of this table cited it for a “20% of assets” figure; that figure is the Wall Street Journal’s, from June, and this page does not carry it |
How we verified this
The holdings tables are parsed from the fund’s own filing on SEC EDGAR — form 13F-HR, CIK 0002045724, period 31 March 2026, filed 18 May — and specifically from its putCall field, which separates common stock from options. That field is the whole article. An earlier version of this piece ignored it, added put and call notional to shareholdings, and reported a $13.68 billion stock book with Nvidia as the second-largest holding. Both statements were wrong, and the error also produced a mistaken “correction” on our earlier article about this fund, which has since been reversed. The lesson is specific enough to be worth stating: a 13F total is not a portfolio value, and any parse that does not read putCall is not reading a 13F.
What a 13F does not contain bounds everything here. It is a regulatory filing, not an audited statement; the SEC does not confirm its accuracy or completeness. It omits short positions, cash, borrowings, most non-listed-equity derivatives, private holdings and total assets. It reports options by the market value of the underlying shares, not by premium, so an $8.46 billion put line is not $8.46 billion of capital.
One inference is flagged as an inference in the body and repeated here because it is the kind of thing that hardens into fact on rereading: eight small common-stock lines sit in exactly the names carrying large puts, which is suggestive of share legs attached to those options, and is not demonstrated by the filing. A 13F shows positions side by side; it does not say they are one trade.
Direction is not claimed for the options. A 13F reports positions, not intent, and gives no strikes, expiries or premiums — so this article says the fund held large puts on the chip complex, and does not say whether they hedged the long book or expressed a view.
Two aggregator errors were caught by going to EDGAR. Several sites list a “Q2 2026” filing for this fund; EDGAR has none, and Q2 is not due until mid-August. Others present the $13.68 billion as a stock portfolio, which is the same conflation described above.
Everything about the sale rests on unnamed sources and the accounts conflict: the Wall Street Journal and TechCrunch describe most public equities sold, Bloomberg and Business Insider a small residual book, Axios the entire portfolio. Aschenbrenner has not commented publicly and no filing describing the sale has been identified. No investment view is offered here.