№ 031Market UpdateJul 31 20269 min read

CreditMacro

Leverage Is Expensive Again

The Fed made liquidity scarce, and flushed the leverage in AI. Crypto already learned this lesson last October.

By Akram Ayyash

At some point before the opening bell yesterday morning, a hedge fund that was up 439% on the year sold everything it owned.

Situational Awareness, Leopold Aschenbrenner's AI fund, unwound its entire public equity book in a single block, longs and shorts together, after margin calls from Goldman Sachs, JPMorgan, and Bank of America. Citadel bought the portfolio. The fund started in late 2024 with roughly $225 million and had grown to somewhere around $20 billion by midsummer, running about four times leverage against AI infrastructure and semiconductor names. Those positions fell between 35% and 47% this month. Assets are now near $10 billion.

Nothing changed about artificial intelligence in July. What changed is the price of borrowing money to bet on it.

The Fed Tightened Without Moving

The FOMC held the target range at 3.50% to 3.75% on Wednesday for the fifth consecutive meeting. Nobody expected otherwise. The vote was 9 to 3, and the direction of the dissents is the entire story: Hammack, Kashkari, and Logan all wanted to raise rates.

Chairman Warsh has been consistent about the rest. The statement was noticeably shorter than the ones the market had been trained on, forward guidance is largely gone by design, and he has called inflation "a choice."

But here's the real story. You can call it higher for longer, or you can call it liquidity is not coming back soon. The reality is that on Wednesday the Fed delivered a rate hike without actually hiking rates.

The Bond Market Already Knew

The 30-year Treasury closed at 5.21% yesterday, its highest since 2007, after trading above 5.24% intraday. Pre-crisis levels, reached in a week the Fed did nothing at all.

The shape matters more than the level. The long end has been steepening away from the front end while the policy rate sits perfectly still. The market is demanding more compensation to hold duration, and taking it without waiting for permission.

Equities spent Wednesday working the same problem out in public. There were three intraday reversals in the S&P 500 on Fed day, totaling something like $2.9 trillion of market capitalization. No new information arrived in those three hours. Because the Fed decided to stop being explicit on guidance, the market was confused. But the market had to react somehow, and it did, it just did it badly, at speed, and in front of an audience.

Capital Can Say No Again

A bond yield is the price of leverage. It is what a lender charges to let somebody else operate with borrowed money, and for fifteen years that price was set close to zero.

It is not close to zero anymore. Investment-grade bonds now yield roughly 1.45 percentage points more than the S&P 500 earnings yield, the widest that gap has been since 2002. Creditors have repriced what they charge to fund corporate balance sheets, and they have repriced it above what owning the equity pays.

Now an investor collects a higher contractual yield while sitting senior to every shareholder, without underwriting a single assumption about margins or adoption curves or what a model can do in three years.

Every asset competes against a real alternative again. The ones that suffer most are long-duration and levered at the same time, because both halves of that description just got repriced. AI is exactly that: a long-duration, highly levered asset.

The AI Selloff Is a Credit Event

Scott Goodwin put it more directly than anyone else this week: there is too much discourse about AI and not enough about leverage. What is underway, in his framing, is a classic head hunt of the most levered players in the equity and convertible markets.

That reframe explains the tape: AI models did not get worse in July, AI financial health did.

Follow the cash instead. The AI buildout is, in balance-sheet terms, a transfer: the hyperscalers write the checks and the chipmakers and data-center operators cash them. One company's capital expenditure is another company's revenue.

Amazon, Alphabet, Meta, Microsoft, and Oracle are now on track to post combined negative free cash flow for the first time on record, down from a peak near $260 billion two years ago.

Which means the companies that spent two decades as net creditors to the financial system have become borrowers in it. Hyperscaler issuance is running near $194 billion this year against roughly $108 billion in all of 2025.

The bond market is not asleep at the wheel. It is charging for this. Spreads have widened, order books have thinned with cover ratios falling from nearly five times in February to below two times this month, and S&P cut Oracle to BBB-, one notch above junk. Investment-grade technology companies are discovering that their cost of debt has to compete with their cost of equity. Nobody has had to think that way since 2007.

Wall Street Collects

Which brings us back to yesterday morning's block trade.

I spent the better part of a decade on funding desks, first at BNP Paribas and later at Citadel, and the most useful thing I learned there was that a margin clerk is not a market participant. He does not have a view on your thesis. He has a schedule, and when your equity and his schedule disagree, the schedule wins. Being "right eventually" is not a defense that has ever worked.

Situational Awareness was up 439% net through June 30. On paper it was one of the top-performing funds anywhere. Then its concentrated positions fell 40% in three weeks against four turns of leverage, and none of the performance mattered, because the arithmetic of a margin call does not care what your returns were last quarter.

This is an old story wearing new clothes. Long-Term Capital Management had two Nobel laureates and a portfolio that was, in the long run, largely correct. Archegos had a concentrated book and prime brokers who each thought they understood their exposure. In both cases the trade was not what killed it. The financing was. Different assets, different decades, identical mechanism: concentration, leverage, and a dependence on the continued generosity of somebody else's balance sheet.

Goodwin published his note about a head hunt of the most levered players roughly twenty-four hours before the most levered player in the trade was hunted.

The retail version of the same story played out in Seoul, where the KOSPI fell about 34% in July, its worst month on record, with seven circuit breakers and roughly ₩250 trillion erased. Regulators responded by moving to curb retail access to leveraged ETFs. That detail tells you what actually broke. Not a view on semiconductors. Borrowed money held by people who could not meet a call.

So What Is Crypto Doing?

Here is what I expected. If equities and AI names are down 5% to 10% on a leverage unwind, crypto should be down 20%. That is the whole reputation of the asset class. It is the high-beta expression of risk appetite, the thing that falls hardest when anything falls at all.

Then I checked prices. Bitcoin and the major altcoins were basically unchanged on the day, which is not what a market in the middle of a deleveraging is supposed to look like. Widen the lens and it gets stranger. Bitcoin was up roughly 11% on the month, and Ether about 22%. So what gives?

In my view the answer is the same one running through everything above. The story is leverage, and how violently the market reprices it. Crypto already read that chapter last October.

October 10, 2025 was the Global Financial Crisis moment for crypto. Massive leverage was flushed that day, and the repercussions are still being felt. Bitcoin is down about 26% this year and roughly 48% from its October 2025 high. Crypto did not dodge this drawdown, it went first. It feels like right now there is very little left to force. Every seller who could be made to sell already had been.

The volatility data is the receipt. Over the last sixty days, Bitcoin's annualized realized volatility has run around 40%. Semiconductors are closer to 57%. Bitcoin is currently less volatile than the semiconductor complex, which is not a sentence anyone would have written in January.

I argued in June that Bitcoin is best understood as a store of volatility rather than a store of value, because it delivered three to four times the risk of the S&P for every dollar committed. That is still true against the S&P. It is no longer true against the part of the market where the leverage now lives. The risk did not leave the system. It changed address. I wrote earlier this year that capital has somewhere else to be. AI is sucking the air out of crypto, and it is sucking the leverage too.

A word of caution though: none of this means crypto is out of the woods. Market depth remains roughly 40% below pre-October levels, and desks describe leverage quietly rebuilding into soft spot demand. Paying the leverage tax once does not exempt anyone from paying it again.

Who Else Is Levered?

That is the only question worth carrying out of this week.

Yesterday the market rallied. Microsoft reported its fastest cloud growth in four years and rose about 16%, semiconductors jumped nearly 7%, and the Nasdaq recovered a piece of the month. Some of that is a genuine answer on revenue. Some of it is just what happens when the largest forced seller in a trade finishes selling and the pressure lifts.

I would not mistake it for the all-clear. On the same day the equity market got its answer, the 30-year yield rose to a nineteen-year high. Equities were reassured. Credit was not.

We know about one fund because its prime brokers forced the issue in public. Four turns of leverage on a concentrated book is not a strategy anyone runs alone, and leverage stays invisible until it is urgent. The hyperscalers are borrowing at a pace nobody has seen from them, into a bond market with thinning order books. The chipmakers on the other side are booking the proceeds as revenue. Somewhere in that circle are balance sheets nobody has stress-tested against a 5.21% long bond.

A fund blowing up is contained. The loss stops at the equity and the prime brokers take the rest. Credit does not work that way. If the financing of this buildout is more entangled than it looks, through vendor arrangements, prepayments, and a supply chain that has learned to borrow against purchase orders, then the next unwind is not a fund. It is a chain.

The sharks are circling in three places at once. Wall Street, where the margin desks have found something to hunt. Silicon Valley, where free cash flow has stopped being a birthright. And as far as Korea, where a retail market just found out what borrowed money costs when the tide goes out.

Crypto has been in this water before. It has the scars to prove it.

So what about AI?

AI has become too large a force in markets to keep treating as a sector story, and most of what I want to say about it does not fit in a market update. I will be spending more time there over the next few weeks, with a couple of deep dives in the works.

Drop me a line:

What do you think? Do you like this? Do you not like this? I would love to hear your thoughts, so please reach me at akram@span.blog

Enjoyed this? Subscribe for the weekly deep dive, or drop me a line at akram@span.blog.