Situationally Aware, Permanent Blind Spot

Many of you probably saw the news last week about the large AI investment firm, Situational Awareness (led by AI wunderkind Leopold Aschenbrenner), that collapsed as a result of the recent pullback in tech stocks.

If so, you’ve probably seen some of the analysis regarding why it fell – too much leverage, concentrated bets, etc. That’s all true, but there is a much simpler explanation that is important to understand, along with a simple solution that could have kept him in business.

He wasn’t running an investment portfolio. He was running the same trade over and over and over again.

To be clear, it was a great thesis. Leopold was absolutely prescient to anticipate the demand for memory, semiconductors, power, etc. and the risk to entrenched software business models.

If he just owned an unlevered portfolio of Nvidia and Micron and a few others, he’d still be successfully running money.

Where he went wrong is he kept doubling down on his bet. Even if you’re 99% likely to be correct on each hand, if you double down after every win, you eventually lose it all.

Double Trouble

What do I mean by doubling down? There are three different aspects to it.

First, all his long positions were highly correlated. Sandisk and Coreweave and Intel are all largely the same position. They trade in the same way and react to the same news and/or sentiment. Thus, having a bunch of similar longs really acts as having one big long called “AI on steroids”.

Second, his short positions were mostly inversely related to his long positions. When Intel goes up, Adobe goes down. However, as he learned, when Intel goes down, Adobe goes up and his losses amplify. In other words, he didn’t hedge, he…doubled down.

Third, he added leverage. It has been widely reported he operated with 4X leverage, so he didn’t double down, he quadrupled down!

Put all three together and it was a near certainty his fund would collapse at the first sign of a change in AI sentiment because his entire investing strategy assumed AI stocks would go up in a straight line forever with no bumps along the way.

A levered investment strategy needs to have low volatility pre-leverage. A high volatility, highly correlated portfolio cannot support leverage. These are well-known and understood investing principles.

For someone so “aware”, it is quite stunning he was so blind to such an obvious mathematical risk. He was playing Russian Roulette and apparently didn’t know it.

It’s forgivable that a 24 year old with no investment background might not understand this, but those providing him the leverage most certainly did. How in the world did Goldman, JP Morgan, and Bank of America all not foresee the risk of collapse?

An Alternate Outcome

What should Leopold have done to monetize his idea without risking losing it all to near term volatility?

The simplest answer would have been to stick to only buying AI stocks without leverage and without shorting. That would have eliminated the margin call risk from any near term volatility. Given how much his stock ideas went up in the last two years, he still would have had fantastic returns.

There is a name for this approach. It’s called a mutual fund. Mutual funds (or ETFs) don’t implode and can still make the manager a ton of money, even if less than a hedge fund. Cathie Wood got a lot wrong in recent years, but she still has a business because she doesn’t use leverage so never faced a margin call when her performance went south.

The next option would have been to add the software shorts but without additional leverage. In that case, he could only go out of business if his AI positions went down 50% while his software shorts went up 50% (I’m simplifying here but stick with me).

If he had $100 in AI and shorted $100 of software, then if the AI went to $50 and the software to $150, he would get wiped out. That scenario isn’t out of the realm of possibility but is far lower odds than when you do it with extra leverage where much smaller price movements can force a margin call.

Leverage Without Tail Risk

But let’s say Leopold had clients (and prime brokers) beating down his door to add leverage to really juice the returns. What could he have done to avoid the margin call risk?

He could have made the clients take the leverage instead! He would have operated his unlevered longs and shorts and then told the clients to borrow on their own if they wanted further leverage.

So Client A would write Leopold a $100 check and fund it with $100 of cash. Client B would also write a $100 check but fund it with $25 cash and $75 of borrowings. Leopold doesn’t know the difference and doesn’t care. His fund doesn’t have the leverage risk.

Now, this isn’t something everyone can do. For a broker to give you leverage, they want to know you have the assets to pay it off. If your only asset is an investment in Situational Awareness – which Leopold can’t be forced to sell to meet your margin call – then you’re out of luck. No leverage for you!

But if you have other assets at Goldman – and if you’re investing in Situational Awareness you almost surely have other assets that can serve as collateral – then it’s all good and a margin call would result in you selling your Home Depot or Disney stock or whatever. It would have no impact on Situational Awareness’s liquidity.

Given how all his bets were really just one giant bet on AI infrastructure, pushing the leverage to the LPs and out of his fund would have been the obvious answer. It protects the fund from tail risk and only exposes individual LPs who choose to actively seek additional leverage.

For the insurance readers here, you might be thinking this sounds similar to fully collateralized reinsurance. The vehicle itself will not collapse in a worst case loss. However, we know some of the LPs lever up their capital contribution so they may get wiped out if the ILS has a partial loss.

Insurance investors understand risk management. Unfortunately, 24 year old geniuses with math degrees from Columbia don’t.