💀 An AI Fund Just Vanished

For the first time in history, the U.S. government forced a leading AI company to pull its most powerful model offline. No court. No warning. One letter on a Friday afternoon, and the most advanced AI on Earth went dark within hours.

It happened in a single weekend. And it happened just as the most hyped AI IPO wave of the decade was taking shape.

That timing is not a coincidence worth ignoring. When Washington can switch off a frontier model overnight, it changes who wins, who loses, and what every one of these filings is really worth.

Most investors will chase the hype and find out the hard way. Our free briefing breaks down what the shutdown signals and the risk buried in the filings — before pricing is announced, not after.

Read the Free Briefing

Hedge funds just dumped more global tech stock in three days than in any three-day stretch since 2016.

That's the year Goldman Sachs started keeping score. Nine years of data, and the record fell over July 24, 27 and 28, according to the bank's prime brokerage desk.

Then on Thursday, everyone found out what the smart money had been running from.

Situational Awareness, the AI-focused hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, sold its entire public equity portfolio to Citadel.

Not trimmed. Not rebalanced. The whole book, handed over in one go.

The trigger was a sharp drop in the value of its leveraged AI positions. When you're forced to sell, you don't get to shop the price around.

Borrowed money has strong opinions about your losses.

A concentrated portfolio can survive a bad week. A leveraged concentrated portfolio can't always survive a bad day. Past a certain drawdown, the margin desk decides your exit timing for you.

And Citadel buying the whole thing at once tells you something about how fast that decision arrived.

The largest position was reportedly the VanEck Semiconductor ETF $SMH ( ▲ 0.3% ), which is roughly the "I'll take all the chipmakers, thanks" trade in a single ticker.

Around it sat the greatest hits of the AI build-out:

Chips, power, storage, data centres. Every single name is a bet on the same idea from a slightly different angle, which works beautifully right up until the idea has a bad month.

The record selling came before the liquidation, not after it.

Hedge funds were cutting tech exposure on the 24th, 27th and 28th. Situational Awareness sold on Thursday. The crowd moved for the door first, and one leveraged fund was still standing in the middle of the room.

That's the risk of a crowded trade. When everyone owns variations of the same position, the exits get narrow at exactly the moment everyone wants one.

If you hold AI-adjacent names, the useful question isn't whether the technology is real. It's how much of your portfolio is making the same bet six different ways.

What would change this read? If Goldman's data shows funds re-levering back into tech over the next few weeks, this was a violent unwind of one crowded position. If the selling keeps going, something bigger is being repriced.

You don't have to scroll every AI thread, track every new tool, or watch every demo. 

The Rundown AI breaks it all down for you — the latest AI news, tools, and tutorials in one free 5-minute email every morning. 

Trusted by 2M+ professionals at Apple, Google, and NASA.

Sign up to start learning.

Five companies spent $182 billion in three months and generated about $5 billion in free cash flow.

Goldman Sachs looked at that gap and got more bullish.

Here's the question sitting under everything right now: is the AI wobble the start of something ugly, or just a crowded trade catching its breath?

Strategist Ben Snider reckons the recent AI selloff looks like every other time momentum investing got too popular. Those episodes usually end in consolidation, then the trend picks back up.

Helpfully, hedge funds and ETF investors have already cut leverage hard. Less borrowed money in the system means smaller tantrums.

But positioning is a story about traders. Earnings are the story about businesses. So how do the businesses look?

As of July 31, companies making up two-thirds of the S&P 500's market value had reported Q2. 64% beat estimates by at least a full standard deviation, one of the strongest reporting seasons on record.

Strip out the chunky investment gains a few tech giants booked and earnings grew 26% year over year. Leave them in and it's 45%.

And this wasn't just five stocks carrying the index on a stretcher. The median S&P 500 company grew earnings 12%, comfortably ahead of the 9% analysts expected going in.

Which raises the obvious follow-up: how much of that is AI, and how much is everything else?

Goldman puts AI infrastructure companies at roughly one-third of Q2 earnings growth, potentially more than half for the rest of 2026 and into 2027. The usual suspects: Alphabet $GOOG ( ▲ 6.88% ), Amazon $AMZN ( ▲ 15.32% ), Microsoft $MSFT ( ▲ 3.02% ), Nvidia $NVDA ( ▲ 2.93% ), Broadcom $AVGO ( ▲ 0.37% ).

The equal-weight S&P 500, which mutes the mega-caps, has kept climbing too, with estimates improving alongside it. The rest of corporate America is fine.

Analysts have already nudged 2027 earnings forecasts up about 1% since Q3 began, with energy and financials getting the biggest upgrades.

One yellow flag: margin forecasts are being trimmed. Tariffs and higher energy costs have been absorbed so far. Analysts are wondering for how long.

Which brings us back to that $182B…

Cloud revenue at Alphabet, Amazon and Microsoft grew 48% year over year, up from 39% the previous quarter. Meta came in as expected.

So the hyperscalers hit the accelerator. Goldman now sees capex topping $1 trillion in 2027, more than $100 billion above where estimates sat before earnings season.

That $5 billion of free cash flow doesn't cover a $182 billion habit. The gap got filled by raising roughly $101 billion in debt and equity.

Goldman reads that as confidence rather than distress. You don't borrow $101 billion in a quarter unless you're fairly convinced the revenue shows up.

The whole bull case now hangs on one hinge: profits. While earnings keep rising, sharp AI drawdowns look like rotation inside a healthy market.

What would change the view? Earnings deterioration. Watch those margin forecasts, because that's where the first crack would appear.

Until then, expect violent swings in AI names that don't say much about the market underneath them.

The US Treasury just bought yen. It hasn't done that since 1998.

Back then, Google was three weeks old.

On Friday, the US and Japan intervened together to prop up the yen, the first time in nearly three decades the two allies have teamed up to buy the currency directly, according to the Financial Times.

Central banks talk. They very rarely spend.

Ugly enough that the yen fell to its weakest level against the dollar since 1986.

That's a little more than a rough patch. It’s more like a currency that has spent 40 years losing an argument.

Reuters reported earlier that day that the Treasury had told several banks it might step in. The market got the memo and kept selling anyway.

Thursday is when it actually started.

The New York Fed rang round the currency-dealing banks and ran a "rate check" on dollar-yen for the Treasury. In FX land, that's the equivalent of your mate asking how much a round costs before he opens his wallet. It almost always means a purchase is coming.

Which it was, less than 24 hours later.

The New York Fed sold euros in exchange for yen on behalf of the US Treasury, per three people familiar with the situation. Goldman Sachs and Morgan Stanley handled the plumbing, two people said. ECB officials were given a heads-up.

Washington reached into the euro pile instead of the dollar pile. Buying yen with dollars means pushing fresh dollars into the world, which is a slightly awkward thing to do when your whole argument is that currency markets are behaving badly.

So they used somebody else's money. Politely. With permission.

The last time the Treasury did this at all was 1998, when it bought yen after the currency hit an eight-year low. That intervention has been the reference point ever since, mostly because nobody expected to need a new one.

Because a joint intervention is a confession.

It means Tokyo decided it couldn't fix this alone, and Washington agreed strongly enough to spend money on it. Governments do not co-sign each other's currency problems for fun. It's an admission that the usual tools stopped working.

For your money, the knock-ons run wider than one exchange rate. A weak yen has been the free fuel behind the carry trade, where investors borrow cheaply in yen and park the money in higher-yielding assets everywhere else. Change the price of that fuel and you change the maths on positions that have nothing to do with Japan.

It also raises the obvious question: if the US is willing to intervene once, is Friday the floor, or the first attempt at one?

Markets are about to find out in public.