Lost in the Fog of Warsh: was Citadel Securities’ rate hike prediction really the end of Situational Awareness?

People think that Citadel Securities’ public prediction that Fed Chair Kevin Warsh would hike base rates caused the collapse of Situational Awareness, and therefore Citadel (the fund) snapping up a rumoured $16bn of Situational Awareness’ public equities book was pre-meditated. This simply isn’t true.

Situational Awareness spiralled because it matched a book of short-term instruments against a very long-term investment thesis. Once the markets caught wind that the fund was struggling, it was over. This is because market perception can be so damaging when leverage is at play. Leverage removes your ability to back yourself and wait out a storm. As an investment strategy leverage can amplify your returns, for sure. But also your problems. It removes one of the most powerful tools in your arsenal as an investor: time.

Markets were also conscious that the struggling Situational Awarenss held very valuable positions in single companies. When there is a panicked seller with large positions selling off its shares, stock prices can go down even after good news. In public markets, stocks are a matching game. For every seller, there has to be a willing buyer and at huge volumes selling can carry a discount to market value.

This actually turns the conspiracy on its head. Citadel has in fact softened the landing because it absorbed the risk to the market of forced selling by stepping in as a holder rather than leaving the book to be liquidated piecemeal.

How did Situational Awareness’ descent start?

The AI complex was taken apart between 10 and 20 July, with most names down 30% or more. The Nasdaq 100 was off around 10% from its late-June peak. Situational Awareness’ holding Micron sat 39% below its record close of 25 June.

Leverage is a cruel mistress and it did the rest. Let’s quickly explain how leverage works (or doesn’t!). Say you hold £100 of your own money and borrow £300 against it, giving you £400 of assets, which is the sort of gearing reported here. A 25% fall in the value of those assets takes them to £300, which is exactly what you owe. So, your equity is gone. But you never get that far, because prime brokers mark positions daily and call for more cash long before it. Worse, they raise the margin requirement itself when volatility spikes or when a book looks concentrated and illiquid, moving the goalposts when you can least afford it. A broker preparing to reprice your risk is called a concentration watchlist. If you cannot post, they sell you out, and they choose what goes and when.

By July 23, Situational Awareness was rumoured to be trying to recover losses. On 23 July, Intel reported better-than-expected earnings and the shares fell anyway, which prompted speculation that somebody large was dumping stock into the good news. The FT identified the seller as Situational Awareness offloading a position to cover mounting losses elsewhere in the book. We know that at least one prime broker already had the fund on a concentration watchlist. Aschenbrenner (the founder) was ringing round investors towards the end of July after his July 24 positive investor update on an ad hoc basis with different terms to each; this behaviour suggests Aschenbrenner had lost control of his liquidity. All of this predates the Citadel Securities note of 27 July. Could a note like that push a fund already in this position over the edge? Theoretically, yes. In reality it was a light touch at best.

It is really difficult to argue that one research note manufactured a panic in a market, and Citadel Securities was hardly out on a limb: three Fed Committee members actually dissented in favour of a hike, the first hawkish triple dissent since 2016. There is also the reality that the Fed did not hike rates in the end. On the same day, Microsoft reported well and its share price rose by around 15%. These events should have relieved pressures, but Situational Awareness was crunched anyway.

What survives all this is a narrower and more interesting observation. Citadel Securities sees a quarter of US equity volume, it makes markets in the options through which most of this book was actually expressed, and it may well have worked out who was bleeding before the rest of the market did. Knowing where the body is buried is not the same as having put it there, but it’s knowledge all the same.

So what actually sent Situational Awareness spiralling?

There is a punchline available about a fund called Situational Awareness losing situational awareness. But I like my jokes to be original and weird, hence the title of this post. Anyway. Roughly 4x leverage, on a single theme, in names correlated to each other and to very little else, is risk on risk on risk. Aschenbrenner also ran a short against software stocks as a separate position. The long leg expressed the view that AI is constrained by physical inputs: chips, memory, data centres, electricity. The short leg expressed the view that AI would replace or commoditise much of SaaS. Two long-dated, short term financed investment thesis moving against Situational Awareness together.

The markets got nervous. Prime brokers got itchy. Aschenbrenner reportedly described the unwind to investors as a bank run. His concentrated, correlated claimant base lost confidence at the same moment and the asset side couldn't be turned into cash fast enough to meet the call. Situational Awareness was undone by the shape of its balance sheet. It wasn’t undone by Aschenbrenner’s fundamental investment thesis, which is right.

Infrastructure investment is one of the greatest challenges faced by AI.

Oh? You thought I'd write a blog and not mention energy transition? Wrong. Transformers are my muse. The binding constraint on AI is not chips, it is power: generation, transmission, transformers, cooling, and above all the queue to accessing power. I spend my working life in this space and the constraint is very real. Transformer lead times have gone from months to years. Connection queues run to years, for reasons you will have read in my many other articles.

Situational Awareness was operating on a thesis that resolves over ten years, funded by margin that reprices overnight. This fundamental structural mismatch was the undoing of the fund's public equities book.

Compounding this: you can't buy the constraint. There is no ticker for a transformer order book, no listed instrument for interconnection capacity, no way to take a direct position in a substation through your broker. So Situational Awareness' book migrated into bitcoin miners and data centre operators, Core Scientific, IREN, Applied Digital and their peers, which went from nothing to roughly a quarter of the disclosed portfolio in a year.

The difficulty is that those companies do not trade as infrastructure. They trade as high-beta crypto-tech sentiment, and their correlation to the physical power bottleneck is a great deal looser than their correlation to the Nasdaq. Buy a proxy for a bottleneck and you own the sentiment attached to the proxy - and on leverage no less - which behaves very differently in a liquidity crisis.

There is a whole cohort of people who believe that any thesis, however physical and however slow, can be expressed in liquid instruments with a bit of leverage on top, and that owning the exposure is the same thing as owning the asset. It’s just not true. The energy transition is a supply chain problem: order books, planning consents, skilled labour, grid connections, and supply chains are owned rather than traded. You get at them by writing a cheque to a business, sitting on its board and waiting out a lead time, which is unglamorous, illiquid and disconnected from what the two-year does in a given fortnight.

It is perhaps not shocking that the private equity investor looking thematically at companies working in energy transition and grid-enabling infrastructure lands on the argument that private equity is a risk appropriate asset class for this sector. I promise, I did get to this conclusion with good intentions.

So, what’s left of Situational Awareness?

The private book. It includes a large stake in Anthropic, which filed confidentially for an IPO on 1 June valued at c.$965bn.

Prime brokers do not extend margin on illiquid private positions so the one asset Aschenbrenner could not post as collateral still stands. ‘Illiquidity as accidental risk management’ was my second runner-up title for this piece.

The energy transition bottleneck will still be there in 2032. The pity is that Aschenbrenner was right about all of it, and will get to watch from the sidelines.