@FinanceYF5: Nailed the AI play, but couldn't survive the leverage 1/ A fund that bet on the AI wave and returned 439% in the first half of the year plunged 67% in July, eventually forced to sell most of its stock portfolio to Citadel. Leopold Aschenbrenner may have been right about the long-term demand for AI infrastructure, but he couldn't hold on until the demand...

Got AI Right, But Couldn’t Survive the Leverage

1/ A fund that nailed the AI wave with a 439% return in the first half of the year plunged 67% in July alone, ultimately forced to sell most of its stock portfolio to Citadel.

Leopold Aschenbrenner may have been right about the long-term demand for AI infrastructure, but he couldn’t hold out until that demand materialized.

The cruelest thing about the market is: [Being right on direction can still get you knocked out early.] 👇 https://t.co/ELleqxAthw

2/ The fund is called Situational Awareness, a name taken from Aschenbrenner’s long essay predicting AI’s next decade.

It was heavily positioned in AI infrastructure — chips, storage, data centers, and power — while shorting software companies like Adobe.

In July, AI hardware stocks tumbled while software stocks rose, with both the long and short sides moving against it at the same time.

3/ What truly turned the losses into a life-or-death situation was leverage.

Core holdings like Nebius, SanDisk, Micron, and CoreWeave fell rapidly, and margin pressure followed.

The fund was left choosing between raising additional capital and selling positions — in the end, it sold most of its public stock portfolio to Citadel.

4/ That doesn’t mean Aschenbrenner’s AI thesis has been falsified.

Demand for chips, memory, data centers, and power could still keep growing over the long term.

But a leveraged fund can’t afford to wait for “long-term correctness”: as soon as prices fall to the forced-liquidation line, the broker sells on its behalf — whether the market rebounds later no longer matters.

5/ Even more ironic, Aschenbrenner named the fund “Situational Awareness” — seeing clearly what’s about to happen.

He may have seen what AI will need over the next few years, but didn’t see how the market would move this month.

The hardest part of investing has never been just identifying the endpoint; it’s also ensuring you survive long enough to get there.

6/ Situational Awareness hasn’t completely disappeared — the fund still retains some unleveraged assets and private equity investments like Anthropic.

So a more accurate statement isn’t “$45 billion completely wiped out,” but rather an extremely brutal deleveraging of public equities.

What difference does it ultimately make if you get the trend right but the timing wrong? The market just provided an expensive answer.

Original text:

DeepSeek V4 Flash official API is now in public beta.

Key Agent capability upgrades: Terminal Bench 2.1 reaches 82.7, DeepSWE reaches 54.4 — multiple scores significantly surpass V4 Pro Preview and are approaching Opus 4.8.

It also natively supports the Responses API format and is now adapted for Codex.