# Special Situations Report — Weekly Call (readable notes)

- **Title:** The "FOMC fumble" and the loss of faith in Warsh (30-yr to 5.21%, a "zoo steepener", gold rallying with real rates), capex spenders outside Microsoft dragging the index down, the Situational Awareness $45B AI-fund blowup and Citadel's rescue of the book, the Reddit (RDDT) add in the $130s after earnings, the Tesla (TSLA)/SpaceX merger rumour, Nomura's $2T-DRAM-by-2030 forecast, the Korean margin-call crash, Bloom Energy (BE) and the big-tech earnings round (MSFT/AMZN good, META/AAPL bad), Fiery (ex-Skillz) and Sable Offshore (SOC) as deep-value special situations, plus the Warner Bros. Discovery (WBD) arb spread widening to 21%
- **Show:** Weekly SSR research call (premium subscriber recording — no public video)
- **Guest/host:** Jay Singh (founder, Special Situations Report; ex-Goldman Sachs)
- **Date:** 2026-AUG-02 (Sunday call; the deck/transcript title it "2026-08-02 - SSN")
- **Source URL:** the premium recording is distributed via the SSR Discord + email (no public YouTube video; no Discord permalink captured for this week)
- **Source PDFs (this folder):** `transcript.pdf` (premium transcript), `summary.pdf` (the condensed report + actionable items + economic calendar), `Agenda.pdf` (the full weekly strategy deck — the call walks it page by page: p2 the economic calendar + earnings, p3 special situations/M&A + the FOMC, p5 mortgage applications, p9-12 the FOMC notes and the 10-yr/30-yr intraday spikes, p13 capex spenders vs the S&P, p15 the Nvidia debt/guarantee piece, p16 the Morgan Stanley AI supply-chain heat map, p17 the Nomura DRAM forecast, p19 the Korean crash, p25 TLT's lowest close since the GFC, p26 the 3× levered SK Hynix ETP, p28 the Kospi drop, p29 Trump on space regulation, p30-32 earnings-driven returns/buyback windows/PE-vs-history/top-10-by-market-cap, p33-34 Iran, p35-36 the China lithography piece + the semi heat map, p37 Morgan Stanley's token-economics model, p40 Bloom Energy, p44-45 the VIX/FOMC-day cross-asset piece, p57 the Situational Awareness timeline, p60-61 Apple), plus the third-party macro pieces uploaded live during the call: `Amundi Macro Views for August.pdf`, and (referenced, not saved here) HSBC "A Cruel Summer", Standard Chartered "Now for the Hard Part" (33 pages on inflation/bonds/the Fed/energy), the State Street Q3 Investment Management Outlook, a Deutsche Bank piece on the Japanese market, and a Vanguard fixed-income perspective. `Reddit Valuation Valuation and Risk Framework.xlsx` is the Reddit model he says will be re-sent to subscribers.
- **Note:** readable notes auto-extracted from the premium PDFs; fillers (um/uh/you know/false starts) removed, wording otherwise verbatim from the recording. Section headers carry the recording's `H:MM:SS` cue. No public video, so the per-name table has no deep-links (the Ref column is the section time as plain text). Auto-transcript garbles corrected to the right entity and listed here: **"SSN" = SSR** (Special Situations Report); **"Besant" = Bessent** (US Treasury Secretary); **"Leopold Ashenbrenner" = Leopold Aschenbrenner**; **"wonderkin" = wunderkind**; **"Fiery, which used to be Skills" = Fiery (ex-Skillz, SKLZ)**; **"Hot 8" = Hut 8 (HUT)**; **"onsemi" = ON Semiconductor (ON)**; **"Sports Radar" = Sportradar (SRAD)**; **"Peace Sky" = Paramount Skydance** (the WBD acquirer); **"Kimmy / Kimi K3" = Moonshot AI's Kimi K3**; **"Open Claw" = OpenClaw**; **"Niko Securities" = Nikko Securities**; **"Wood Capital Management" = Sowood Capital**; **"Iren" = IREN Ltd**; **"Rebecca Chong"** left as spoken (the tech-unwind note's author). Internal garbles/inconsistencies in the source, left as spoken and flagged here: CXMT's DRAM share is read as **"about 69%"** where the report and the surrounding argument make clear it is **6-9%**; **"either CXMT needs to come down, or Microsoft and SK Hynix and Samsung need to rise"** should be **Micron**; **"Korea's 2023 year-to-date price return"** should be **2026**; China's strategic petroleum reserve is called **"1.0-odd trillion"** barrels (the report repeats it; the real figure is orders of magnitude smaller); Microsoft's fiscal-2027 capex is given as **"1.175 billion"** where the context is **$117.5B**; Meta's raised capex guidance is stated three different ways in the same section (**"135 to 145"**, **"125 to 145"**, **"130 to 145"**) — the report uses $125-145B; and the Nasdaq's worst-July comparison mixes **1986/1985/1987**. The transcript footer's "Date of Transcript: 2024-07-26" and the "2026-01-26" filename are stale template artifacts — the real call date is **2026-08-02** (Sunday).

---

2026-08-02 - SSR - FOMC Fumble, Capex Spenders Outside of MSFT Drag Indices Lower, Situational Awareness AI Blowup, RDDT Buy After Earnings, TSLA Merger with SpaceX Possible

## Introduction & Market Overview - 00:00

Happy Sunday guys. Hope you're all doing well.

We are focused on earnings this call. We had the biggest week of earnings over the last week. This week, we also had, lo and behold, ahead of futures opening, another call for peace from Trump. Trump alleging that Iran, before the US attacks their energy infrastructure, is asking for a peace treaty. They're obviously denying it, so we'll discuss that as well.

We'll also discuss the FOMC meeting. We call it the FOMC fumble. And the FOMC fumble resulted in the 10-year rallying as the market lost faith in Warsh. This also resulted in one of the few times in history where gold has actually rallied as rates have rallied, because of a loss of faith in the US dollar and a loss of faith in the Fed. And that's effectively one of the biggest risks in the market right now, the strength of the 10-year.

As we speak, futures are rallying about 40 bips on the S&P 500, 70 bips on the Nasdaq, and 50 bips on the Russell as oil is down 4% on this new peace discussion. And as you can see, as oil is down, the 10-year is also down about 5 bips, back down from 473 to 469. So because of that, you are seeing a knee-jerk reaction as shorts likely covered that were expecting further bombings over the weekend.

Again, this Iran piece is not certain. It was likely that we tried for another ceasefire ahead of midterm elections and that's what we're seeing now. We very well could see further attacks next week after the market opens, so don't use this to increase your risk appetite too much.

Not only have we seen the 10-year yield be a risk to risk assets, we also saw the Japanese 30-year rise over 4% for the first time in several decades. And I personally think that there is a risk of a September hike being priced into the market, and Warsh may have to acknowledge that given the market effectively called his bluff after the FOMC meeting.

We will also discuss adding to our Reddit position after mixed user results but earnings that blew it out of the park. And we'll discuss a special sits update with a discussion of Tesla selling its Chinese division, EA, TXNM and AES.

And we'll also discuss upcoming earnings this week, which is obviously not as important, but it's still a pretty jam-packed session.

The majority of the market's rally this year has been supported by earnings. And contrary to what people believe, earnings is actually the strongest we have seen in several years. And as I'm speaking here, I'm going to upload the PDF, as I've realized it's not uploaded yet, so that you have the PDF and we can follow along on page one as we go through this. And as I'm speaking, I'm also going to be uploading several macro pieces. Starting with HSBC, titled A Cruel Summer, which is a short read.

And I'll just upload that as I'm speaking, and I'll upload several other pieces that I've read that I don't talk about in the strategy piece, that you should read separately. More information you can digest. You can even summarize these in AI to get a better, just a quick overview. But this HSBC piece is the first one.

The second one we'll upload is Standard Chartered, titled Now for the Hard Part, which is a 33-page piece on inflation, bonds, the Fed, and energy. So I'll upload that as well.

And I'm not going to be able to send these in email, I have to send them separately because they're too large. The third piece we are uploading is the State Street quarterly investment management outlook. This is for the third quarter. State Street is one of the largest ETF issuers in the world, so it's a nice data pack.

Forgive me for uploading anything more than once, just trying to be efficient here.

I have Deutsche Bank on the Japanese market. The 10-year has been a big driver and the 30-year in Japan has been a big driver of a risk-off at certain points in time over the past year.

And just a couple more pieces before we go back to our scheduled programming.

Amundi's views for August macro views.

I'm adding a Vanguard piece on fixed income perspective. So I'll likely add some more, but for now, let's just get back to our scheduled programming.

## Upcoming Earnings & Economic Calendar - 08:17

So now that you have the PDF, we can jump to page two.

And we can look at the economic calendar for next week. The most important items are US manufacturing PMI, ISM manufacturing — we'll keep an eye on ISM prices paid — construction spending, factory orders, JOLTS job openings, which is quite important on the labor market, durable goods orders, US services PMI, the ISM services index, and the change in non-farm payrolls.

At the bottom of page two, you can see the earnings for this upcoming week. We will also talk about the major earnings last week during this call. But the upcoming earnings this week are Marriott International, the hotel business, Sportradar pre-market, Avista Corp. After hours we have Palantir on Monday. The Clorox Company, BWXT, Whirlpool, Jazz Pharmaceuticals, Snap, ON Semiconductor and TKO.

On Tuesday, we have Pfizer, which has been a difficult value name for the past couple of years due to its backlog. Caterpillar, which has just been a rocket ship. I think it's probably overdone here. McDonald's, which has obviously been negatively affected, with a lot of consumer names, due to GLP-1 drugs and less people eating out.

Wayfair on Tuesday, Hut 8. Wix, which has rebounded nicely after the software sell-off again. They use too much of their cash to buy back stock, which let short sellers hit the company again. It has rebounded as a lot of hedge funds that were long semis, short software, covered that trade. A lot of software names rallied last week like ServiceNow, Wix, etc., which we own both of.

And Energy Transfer, one of the biggest energy MLPs where we own common and prefs, is reporting on Tuesday morning pre-market. We also have Merck. After the close, we have the AI GPU giant AMD reporting, and SpaceX is reporting after the close on Tuesday. You also have Arista Networks, which we've owned in the past. Zeta, Opendoor, Booking Holdings, and Hecla Mining on the silver side, which we own.

On Wednesday before the open we have Eli Lilly, Shopify, which I think is an interesting growth name. Uber, which is also getting very cheap. Walt Disney, Riot, Circle, which sold off hugely after banks came together to build their own stablecoin. As you know, almost all of Circle's revenue is just interest they earn on their stablecoin reserves, so it's not a very defensible business model. We were short that last year. I wish we had kept that short on. Lumen Brands, Unity pre-market, and then Novo Nordisk, which has underperformed Lilly.

After the close, we have SanDisk on the semiconductor side, so a big week for semis with AMD on Tuesday after the close, Wednesday SanDisk after the close. Also Western Digital. And then a name that we've added to recently, AppLovin, on the growth side as it onboards more e-commerce clients. I think AppLovin is actually quite cheap relative to its growth on a PEG ratio, PE to growth. ELF after the close on Wednesday, TTM Technologies, MercadoLibre, which we've owned on and off for several years now. Very impressive LatAm growth story. Have a small exposure there. Redwire, Beyond Meat, which I'm surprised isn't bankrupt yet, and Albemarle.

On Thursday, we have ConocoPhillips pre-market. We have D-Wave Quantum, DataDog, which has been one of our best cybersecurity and software name longs. Celsius, where we've trimmed Celsius. We also have Dine Brands, Fiserv, and Cheniere LNG reporting pre-market on Thursday. After the close, we have DraftKings, which is suffering a bit from just rising competition. Airbnb after the close on Thursday. Rigetti Quantum, the Trade Desk, which is getting quite cheap on the algorithmic advertising side. We have Red Cat after the close and MP Materials, rare earth metals.

Friday before the open, we have Oklo. Then we have Vistra Energy, which we have a position in. Wendy's, which has become a meme stock like GME. Under Armour on Friday and Canopy Growth on the marijuana side.

So again, while not as big of an earnings week as last week, it is quite a busy week. And again, all questions, you can type in the SSR Q&A tab. If you're unable to type any questions there, please let me know. I want to have an interactive Q&A session.

## 10-Year Rate Risk - 14:08

So, EJW, we'll definitely get to this question on the 10-year climbing to 4.7%. It was at 4.73%, but because of this discussion of peace that Trump has put out there right before futures opened, the 10-year has come back down below 4.7. But if the 10-year crosses 4.9% a major risk-off event will be triggered. I don't think the 4.9% level really matters. It's anything close to 5%, whether it's 4.9, 5, 5.1, 5.2. Obviously the higher it goes, the more the market will fall. My calculation is for every 10 bips rise in the 10-year, the market should fall about 2%.

And that accelerates close to 5%. So once we get to 5%, every 10 bips above 5%, I think the market should correct by 2%, and I think that Bessent is very focused on this. In a worst-case scenario, the Fed could do something like an Operation Twist and issue short-term bonds to buy long-term bonds. I doubt they'll do that anytime soon, though. And the easiest way for the long end to come in is for Trump to actually commit to a peace treaty with Iran so that oil prices fall below 70. And I think that's what he's trying to do over the weekend. But obviously Iran isn't playing ball. You need two to taco in this situation. You can't just have a taco from Trump, after, I think, the 170 bombings in two days, followed by additional bombings throughout last week, and threats of bombing energy infrastructure, and the bombing of bridges. Obviously Iran now has to swallow their pride and agree to whatever peace treaty is being brought up again. So it's very, very unclear how things will play out. We'll get back to that discussion.

## Special Situations / M&A - 16:15

In terms of special situations, merger arb, spin-off, etc., on page three: through the end of July, US public M&A volumes are up 17% year-over-year. Now that's despite higher rates, despite all the geopolitical uncertainty, and it's because of a handful of bigger deals. There were 11 transactions announced this past month over $1 billion in enterprise value each, $52 billion in total. Our flows — the big funds — are focused on Electronic Arts closing next week, and certain names like Roku, UNF, VAL, and AES.

In terms of the headlines, the Wall Street Journal reported that Tesla is weighing the separation of its Chinese business ahead of a potential merger with SpaceX, but claims were dismissed by Musk. Musk has dismissed claims like this in the past, but there could be some truth to this, obviously, with SpaceX down 50% from the IPO price.

In other news, TXNM reported Q2 numbers, AES received China approval for the deal close. FCC Chair Brendan Carr expressed more conviction in the Warner Brothers–Paramount Skydance deal closing, even after the spread widened out. So the spread has actually widened out for the Warner Brothers–Paramount Skydance deal to about $5.53 per share, so about a 21% spread. The fact that that's that wide and the chairman of the FCC is saying he thinks the deal will close, I do think makes this spread — with the shares around $26 a share — we did see a little bit of a rally on Friday. These were kind of in the mid-25s. It's a decent spread to have a little bit of exposure to. I think if it breaks, you go under 20, and if it closes, you go above 30. So it's not one to add a lot of, it's like a one-to-one up-down type trade. But if you believe the FCC chair and you believe the probability is well above 50%, then the spread is quite interesting.

## The FOMC "Fumble" - 18:40

So on page three, we also talk about how the Federal Reserve has mucked up this latest FOMC meeting.

Year-ahead consensus predictions rarely get things perfectly right, but it's fair to say that despite looking very promising for the first two months of the year, the consensus call for lower interest rates looks particularly bad. The 10-year US Treasury yield has risen sharply against expectations that it would fall, and instead of further trimming its policy rate, the Federal Reserve may shortly begin hiking again. And obviously, the 10-year rise has been historically bad for tech stocks, bad for crypto, bad for gold. Real interest rates are also negative for precious metals. And this 10-year rally, given how benign inflation has been outside of memory and outside of oil especially, as rents and real estate prices have been coming down — I do think that if we get control of the war, then we think that inflation may be short-term.

But anyway, if inflation is not contained then a September hike is definitely on the table. So it's a very binary situation. Instead of further trimming its policy rate, as the market expected at the beginning of the year, the Federal Reserve might have to begin hiking, and the adverse scenario of the war not ending. The effects of this mess are all around us. We see a flatter yield curve, a frustrating lack of progress in the housing market recovery — in fact, mortgage applications have been coming down — and a further souring of consumers' moods, just to name a few. Changes to the central bank's framework and communication often have great implications for investors. These reforms under Warsh, as they are put into place, may drive more of the type of bond market volatility that occurred last week. This could leave long-term interest rates higher and contribute to slower growth in the rate-sensitive parts of the economy.

So what did we learn from the July Fed meeting? Coming into last Wednesday's Federal Open Market Committee meeting, investors were putting the odds of a rate increase at roughly one in three. That may not sound like much, but one can count on one hand the number of Fed meetings in the past 20 years whose outcomes were so uncertain going into the meeting. The evolution of investor expectations for Fed policy from multiple cuts to at least some chance of a hike by July illustrates the interest rate odyssey of 2026. By the way, I don't know if anyone's seen the movie Odyssey. Phenomenal. Too long — three hours is too long — but given how terrible most movies have been, I thought it was quite entertaining.

So as it turned out, the Fed did not raise rates. Notably three of the 12 FOMC members dissented from that decision, favoring an immediate increase. But the market-implied odds of a hike by the next meeting fell from over 100% to below 60% the morning after the decision.

In his comments following the meeting, Chair Warsh did not share any thoughts on the economy or policy outlook beyond what was in the committee's brief and unchanged statement. The most interesting part of the press conference was Warsh's repeated mentions of the rise in real interest rates, among other market moves since the last meeting, as perhaps one reason the committee did not feel it needed to adjust policy. We'll cover that in some depth below.

The case for hiking versus the case for waiting. The September FOMC meeting is only a few weeks away, and it's one of the biggest sources of risk that we face in today's market.

And despite a lack of guidance from Chair Warsh or the FOMC statement, markets continue to suggest that a hike is more likely than not. The best case for hiking interest rates is that inflation has been too high for too long. Indeed, core PCE inflation has been above the Fed's 2% target since March of 2021. Even the new chairman, who has been reluctant to say much of anything about the economy at this point, has made it clear that the Fed is fixated on bringing inflation down. While energy-related inflation has been responsible for the bulk of new price pressures this year, proponents of higher interest rates cite other, less transitory drivers such as continued wage pressures, which are not a real thing, and AI production bottlenecks as a reason to hike.

Indeed, core services inflation excluding housing, often referred to as super core for its stickiness and lagged response to tighter policy, remains well above even its highest level from pre-COVID years, but not above COVID levels themselves. Core services inflation is connected to the labor market because of its close connection to wage growth, while noting that nothing in the labor market currently screams overheating. Nothing in aggregate weekly payrolls suggests that slightly higher interest rates will drive a significant further slowing in the labor market. So basically the labor market's not overheating, but it's also not slowing down. With inflation above target and unemployment barely above 4%, it's a hard time to be a dove on the FOMC.

The argument against raising rates begins with the fact that the cumulative tightening of 2022 to 2023 is still wreaking havoc on rate-sensitive parts of the economy like commercial real estate and the housing market. Mortgage rates just hit their highest level in a year and mortgage applications for purchase are still barely half of what they were in the late 2010s, let alone the post-COVID boom years. You can see the mortgage applications chart on page five.

Third, this year's inflationary shock has come primarily from energy supply shortages, not a surge in overall demand. As a result, wage growth after inflation has fallen essentially to zero. This has historically not been an environment that calls for a tighter monetary policy, as the dark bars below indicating recessions show.

On page seven: to be candid, the questions of whether the Fed should hike or will hike over the next several months are hard to answer. Warsh and company will have another six weeks of economic data to consider before their next meeting. Soft core inflation prints like the one we just got for June could keep the hawks at bay, while unexpectedly strong consumer or manufacturing data could make it harder for the doves to advocate for more patience. If forced to make a call, we would say that the major data in the coming months should marginally weaken the case for a hike at the next meeting, supporting a hold stance from the Fed.

So what is the bond market telling us? Of course, the Fed is not the sole arbiter of whether monetary policy is too loose or too tight. The bond market has a say as well and often doesn't wait to hear from central banks before it chimes in with its own view. Consider the yields on long-dated inflation protected securities, or TIPS. They're almost at 3%, the highest they were since 2008. The sharp rise in long-dated yields since 2021 has coincided with the end of zero interest rate policy and a shrinking Fed balance sheet. Together, these represent a significant tightening of financial conditions, one that has continued through the month of July with a 30-year TIPS yield approaching the highest yield ever.

Higher real yields reflect investors' collective view that bonds must offer higher returns to compete with other assets, but they also weaken growth in home borrowing and capital spending. Now, it's likely that Warsh and the other eight members of the FOMC majority last week felt that rising yields had given them some cover to delay tightening policy.

In other words, the market had already done some of the work for them. Playing the ball instead of the referee, as he put it. This approach may work provided rates continue to move in the direction the Fed prefers, but it's also likely to lead to more volatility than would otherwise be the case as markets overreact to data to get the Fed's attention. Long-term rates are, in part, a reflection of the market's expectation for short-term rates. And therefore, that spike we saw in rates was a reaction to Warsh delaying hikes.

For further evidence of tightening in the bond market, we can look at the slope of the yield curve. After steepening for almost two years, the curve has flattened this year on increasing expectations for Fed hikes and marginally less optimism about economic growth. A flattening yield curve is a sign that monetary policy is restricting growth and financial activity. Until it infamously registered a false positive in 2022, an inverted curve has been a very reliable forecaster of recessions.

The messy market reaction to the FOMC requires a quick aside. As the graph above shows, the yield curve steepened immediately following the latest FOMC meeting, specifically following Chair Warsh's comments about allowing markets to follow data without considering how the Fed might react. He also hinted at a potential change to the Fed's inflation framework next year, which could weaken the credibility of the current 2% PCE inflation target. The curve underwent a zoo steepening where short-term interest rates fell on dovish hopes, like the two-year falling, while long rates rose on credibility concerns — i.e., bulls and bears pulling in opposite directions. Historically, this is a sign a central bank risks losing its credibility and it bears — forgive the pun — close watching. And like I said, I do think that this is one of the biggest risks to the market right now.

The war effectively coincides with this inflation risk, but I think the bigger risk, war aside, is that we get some rise in inflation and/or perception that the Fed has lost its ability to control inflation. And that would be an absolute risk-off in the market. And it would actually almost force Warsh to acknowledge that he was wrong in September. And I think that's one of the reasons why, after the strong tech rally, and every single bounce, the market continues to get faded — because everyone's eyes are on the bond market, not only the US bond market, but the Japanese bond market. In fact, one of the risks that we might face is that the Fed might actually have to extend swap lines to the Bank of Japan; they may have to help support the Yen at some point.

You can see the US Treasury yield slope, the 2s10s, on the following page actually. And the Fed has seen its credibility waver before, but it has been nearly 50 years since Arthur Burns lost it completely.

Last week's volatility aside, we remain more concerned about the effects of monetary tightening, whether intentional by the Fed or not, than we are about the risks the Fed will let inflation get out of control. Moreover, Treasuries are not the only financial assets showing the effects of tightening financial conditions.

Stock valuations have fallen mainly for reasons other than monetary policy, it must be said, and corporate credit spreads have crept up from their lows. IG credit spreads have been so low that I haven't found any value outside of prefs in bonds. And high-yield spreads have also been relatively low despite this private credit fear in the market.

Now, while spreads have continued to rise from these low levels, they are still nowhere near attractive. And I continue to believe that the most prudent thing to do is to avoid high-yield bonds and IG bonds, and to focus on areas where there is alpha potential and spread differentials above high yield, like in prefs and in closed-end funds where you can actually buy things at a discount to par. Or at a very high yield relative to the average spread in high yield, which is around 300. So I do believe strongly that the US bond market is not very attractive relative to equities, which are also expensive. I think emerging markets, I think that certain prefs, and I think that a number of event-driven names and value stocks that have been left for dead are the most interesting way to generate P&L for the rest of the year. There are also a lot of growth stocks and GARP stocks that have sold off, that are becoming growth at a reasonable price, that we'll continue to highlight over the course of the year. And names like Fiery, which used to be Skillz, which just got a $719 million judgment awarded by a judge to recover from Papaya Games. Now, it's unclear how much they'll settle for, it's unclear what Papaya Games is worth, but Papaya Games has done several hundred million in revenue over the last couple of years, and well over a billion in revenue over the last five years. And what that means is that even if they settle at something like a third of that 700 million — this is a 70 million enterprise value, 150 million market cap. The stock could be more than a double after the appeal. And while the appeal could take more than a year, with the stock at 9.16 today, it effectively means that in the bull case, the stock could be worth over $50 a share, and in a settlement case, could be worth well over $20 to $30 a share. And again, there's timing issues, there's also the fact that the company has a debt maturity in 2027, which means that it may issue a new convert, maybe dilutive. But I do think that we'll continue to find names like that.

Another interesting name where we had sold puts is a company called SOC — Sable Offshore. We had sold some puts around 84 cents. That was after Sable Offshore sold off pretty dramatically. And they refinanced their debt. And that refinancing of debt and the issue of stock was dilutive. We didn't have a position at that time. Sable Offshore sold off to like a four handle and the options volatility blew up completely. If you look, we went further out. We went to basically Jan 28, and you could effectively sell the three strike puts. If you believe that the company no longer had bankruptcy risk because they termed out their debt, and that the Trump administration would overrule current regulations and allow drilling offshore in California, then at the price you're creating the company it was effectively de-risked. So what we ended up doing was we sold the two strike puts, going out to Jan 28, and you effectively created the company around like 1.20 a share. And because the vol was well over 120, those puts are probably still trading around 65 cents — yeah, they're marked around 65 cents. So you can create the stock at like 1.35 and it's quite a decent yield to take in that much premium for a year and a few months. And so we also own some of the common shares around four, and there's some news over the weekend that Trump is going to effectively — I don't know if it's actually fake news or it's been approved or confirmed — but it looks like there is going to be some support for Sable. And even if it's not confirmed, it makes a lot of logical sense given where energy prices are going, for there to be more drilling in the US. I mean, if you look at where the sell side is: Jefferies, Lloyd Byrne is the analyst at Jefferies. He believes that you'll see drilling resumed, there are several pieces that you can read. He did lower his target on the stock after the equity raise from 24 to 11, but he still believes in the story. It is a highly risky story. We just believe that because they've actually refinanced the debt, we've created that risk low enough, created the stock at 1.20 versus where it is in the overnight market at 5 bucks a share.

## Capex Spenders and Market Impact - 40:04

Now, going back to the macro discussion very briefly. Treasuries are not the only financial assets showing the effects of tighter financial conditions. We talked about momentum stocks falling to multi-year lows. We'll talk about that later in this piece. While the market isn't concerned about a recession, they are gently reminding us that regardless of what the 12 FOMC voters decide over the balance of the year, the markets do get a vote. The era of radical Fed transparency is clearly over with Chairman Powell, rest in peace. The regime that has replaced it has yet to make any concrete policy changes or even any indication as to when or why future changes might occur. This puts the market firmly in the driver's seat, which Chair Warsh seems to prefer.

One lesson of 2026 is that Fed policy is not always the primary driver of interest rates in any given moment. For the time being, the global economy and financial markets, including bond markets, remain hostage to the developments in the Middle East and knock-on effects of an impaired global oil supply. Over the longer run, the data will tell us whether policy is too tight or too loose, and the uncertainty about the Fed's plans for interest rates should start to feel more normal for investors. However, the impact of higher rate volatility on the economy — fewer mortgage applications, less business investment and higher Treasury borrowing costs — may start to add up soon. You can see my full notes for the FOMC meeting and the reaction and the sharp spike in the 10-year yield and how short the policy statement is on page 9, on page 10, on page 11, where you can see that 10-year intraday spike, and page 12 where you can see that 30-year go to 5.21, resulting in that fear-led sell-off after the FOMC meeting. Now, mortgage rates are actually more correlated with the 10-year given mortgages have a 10-year duration, even though they're 30-year in maturity. So that 10-year rise is resulting in mortgage rates being at the highest levels of the year. And you can see that core PCE report from Bianco Research saying that core PCE has rebounded by almost 80 bips, almost a full percent, since the April 25 lows, after tariffs kicked in and now the war has kicked in.

Now on page 13, we change the subject temporarily and talk about how capex spenders have dragged down the S&P 500 this year. Microsoft, Amazon, Alphabet and Meta, whose enormous capital spending plans have made them relative laggards and therefore a drag on the market-cap weighted S&P, have been a topic of discussion as they've gone from free cash flow positive to free cash flow negative. We won't spend too much time on this, but it is an ongoing theme as Meta increased its capex and the stock sold off almost 10%.

Apple was safe until it missed earnings this week on spiking memory costs, and we'll touch on that later as well. Companies receiving capex money have generally done better than companies spending it, although the semiconductor trade also unwound with a $45 billion fund unwind which we'll talk about later, run by Leopold. And funnily enough, that fund was named Situational Awareness when obviously this 24-year-old wunderkind did not have situational awareness of shorts smoking him out.

Chip makers, memory suppliers, data center equipment companies and power infrastructure providers are booking revenues. Hyperscalers are paying today for capacity whose financial return may take years to prove. While operating performance, i.e. EPS, has improved, free cash flow has collapsed because of infrastructure spending rising faster than eventual cash flow return on capital. This matters because mega-cap technology stocks historically commanded premium valuations of 10 PE multiples higher than they are now, because they were highly cash generative and relatively asset-light. That asset-light nature is changing dramatically as they become more capital-intensive. The four largest hyperscalers are reportedly on course to spend over $725 billion this year and much, much higher next year. You can see in the cross-sectional graph in the middle of page 13 that in terms of the share of the S&P 500 moves since the start of June, Alphabet, Amazon, Microsoft, Meta, Oracle, Tesla have been dragging the market lower.

The Mag 7 optically on EPS look very cheap, but on cash flows, they don't look as cheap. The revenue tends to arrive later than the spending, and you're starting to see revenue growth due to the cloud backlog. But all the spend is going into GPUs and AI accelerators, several hundred billion a year, servers and networking, data center construction, electricity generation, grid connections, cooling systems, and land and long-term leases. The suppliers generally recognize revenue when equipment is delivered. By contrast, Microsoft, Amazon, Alphabet, and Meta must subsequently generate through Azure, AWS, Google Cloud, advertising and AI product revenue an adequate return on those assets. And we'll discuss cloud revenue growth later in this piece.

Demand may be real, but that does not automatically mean the economics will be attractive. While Satya Nadella argues that there's a 30% ROIC on AI spend, not every company is going to achieve that level of ROIC. Competition could push down AI inference prices before hyperscalers have earned back their investments. And depreciation is a delayed earnings problem. Capex initially appears on the balance sheet rather than being expensed immediately. So earnings are subsequently affected through depreciation over the useful life of servers and data center equipment. As each new investment cohort is layered on top of earlier spending, depreciation keeps rising, even if capex eventually stops accelerating. That means reported earnings could face a multi-year margin headwind if the cash has already been spent. Goldman Sachs expects the AI capex boom to reduce mega-cap technology ROE, with consensus estimates implying an average decline of roughly seven percentage points of ROE next year.

Capital intensity reduces ROIC and ROE, using the simple equation you can see below. So suppose a company earns 80 billion and has 300 billion of invested capital. The ROIC is 80 over 300, which is 26.7%, very high. It then invests another 200 billion in AI infrastructure, but near-term earnings only rise by 10 billion. So the ROIC is now 90 over 500 billion, which is 18%. Earnings have increased, but the return on the enlarged capital base fell sharply. Especially as these companies like Google raise equity to fund their capex, they really need ROIC to rise very, very rapidly to make up for that type of action. Lower prospective ROIC normally warrants a lower valuation multiple, especially when investors previously valued the company as an asset-light compounder.

Another technical is that there's less money available for share buybacks, which have fallen precipitously for the Mag 7, which were doing most of the tech share buybacks. A dollar used to build a data center cannot simultaneously be used for share repurchases, dividends, acquisitions, or debt reduction. In fact, debt has been rising. There's going to be $5 trillion of debt issued in the IG markets for AI-related projects. Buybacks have been particularly important for mega-cap tech companies over the last few years because they reduce share counts and support EPS growth. Capex therefore hurts shareholders in two ways: lower free cash flow and slower buyback-driven EPS accretion. The Nasdaq and the S&P 500 are market cap weighted. Hyperscalers are amongst the largest constituents, so even moderate underperformance by the hyperscalers has an outsized index impact. This is one reason why investors have rotated towards businesses with stronger near-term cash distributions and better capital discipline. Morgan Stanley notes that the equal-weighted S&P 500 has outperformed the conventional market cap-weighted S&P 500 in 2026, with investors favoring cheaper, more cash flow-oriented companies over mega-cap growth names. And you can see that hyperscalers have a 20% SPX weight, negative 10% return, and a negative 2% index contribution for the year. In fact, if it wasn't for the hyperscalers, the S&P 500 would actually be higher by 2%. The remaining companies are actually up 12% for the year and have contributed 9.6% to the index. So the total S&P return has been 7.6% net.

Thus, the rest of the market can perform strongly while the largest capex spenders subtract several percentage points from the headline index. This helps explain why the equal-weighted index can outperform the market cap-weighted index. Morgan Stanley is raising its 2027 and 2028 hyperscaler capex forecast by 9% and 10% respectively again, now modeling $1.2 trillion and $1.4 trillion in annual spending. Google raised its 2026 capex guidance from 180 to 190 billion to 195 to 205 billion, posting negative free cash flow for the quarter and saying demand still outpaces its capacity investments. TSM raised its 2026 capex forecast as well. It raised its guidance from 52 to 56 billion to 60 to 64 billion and is increasing its full-year revenue growth outlook from more than 30% to above 40%.

ASML raised its 2026 revenue guidance for the second consecutive quarter, first from 34 to 39 billion euros to 36 to 40 billion euros. And now to 43 to 45 billion euros. It's also planning a roughly 30% capacity increase for low-NA EUV and DUV immersion systems by 2027.

Meta raised its 2026 capex guidance from 115 to 135 billion, to 125 to 145 billion to reflect higher component costs and additional data center spending. That's the reason why Meta underperformed so much last week.

OpenAI raised its projected compute spending through 2030 from 600 billion to 750 billion, a 25% increase, while shelving its $60 billion IPO. It's one of the reasons why NVIDIA gave the company a $250 billion guarantee, which resulted in NVIDIA CDS — insurance on NVIDIA debt — rising quite dramatically last week. Microsoft continues to say that demand exceeds available capacity, even after delivering additional capacity earlier than expected. And Kimi suspended new subscriptions after demand for Kimi K3 overwhelmed its available compute capacity. This was the Chinese model.

So Morgan Stanley, as you can see in the bar charts below, has 2026 capex at 779 billion, 2027 capex at $1.23 trillion, which is unbelievable. And 2028 capex at $1.4 trillion.

## NVIDIA Debt Risk & Guarantees - 52:00

Like we mentioned earlier, on page 15: NVIDIA debt risk jumps on a $750 billion AI push. The cost of insuring NVIDIA's debt surged to a record after reports the company is discussing more than $750 billion in AI infrastructure financing tied to OpenAI. I think people misconstrued that. It's not like they're actually spending that or supporting that. It's just that they're guaranteeing up to 250 billion to help OpenAI lease computing capacity for its massive 10 gigawatt, $500 billion data center hub in Ohio, which has been developed by a Softbank subsidiary. And there have been separate talks involving NVIDIA potentially financing 350 billion of OpenAI's processor purchases for that same project. So basically giving it financing for it to buy GPUs from NVIDIA. So this is a very big circular deal.

And then there's this SK Group partnership where NVIDIA announced a multi-year partnership with South Korea's SK Group valued at over 500 billion and encompassing mutual procurement and memory chip supply. So there have been all these big circular financing deals which are finally scaring the market, because NVIDIA was effectively a cash cow with not a lot of debt and a very clean balance sheet, but now it's getting quite involved doing hundreds of billions of these guaranteed deals — and so its CDS was as low as 40 bips, credit default protection, in June, and it's now at about 80 bips.

## ASML and Chinese Competition - 53:35

After news that the Chinese were building their own version of ASML, people realized that that would only take about 5% market share over the next couple of years. So Bank of America Securities reiterated its buy rating and price target on ASML. ASML didn't sell off enough for us to get too excited, although I was thinking about buying it as well. BofA called the recent sell-off an overreaction. The bank estimates a Chinese shift to domestic DUV tools would cut next year's sales by only about 2.4%. It said that ASML's EUV leadership remains unmatched and views its current valuation as attractive.

## AI Supply Chain Overview - 54:23

Morgan Stanley published this really interesting AI infrastructure supply chain value heat map. And you can see they separated it from the owners and operators — the hyperscalers, Meta, Alphabet, Amazon, Microsoft, Alibaba, Tencent, SpaceX, and Oracle — to the DC REITs and operators, so all of the data center REITs. And then the private equity-backed asset managers that have data centers like Blackstone, Macquarie, Brookfield. And then the enterprise tier-two clouds like IBM, SAP, Salesforce, DigitalOcean, and then the neo-clouds, which are Nebius and CoreWeave, which rallied a lot last week. We said that they were quite interesting as AI lease rates were bouncing.

Now, you have the semi-production companies including the IC design companies, NVIDIA, AMD, Qualcomm, Hewlett Packard Enterprise, Marvell, Broadcom. Then the GPU names, NVIDIA and AMD, and the power supply names, the server brands, liquid cooling companies like Vertiv, grid and on-site renewable storage, the OSAT companies, CPU companies, Intel, Marvell, and AMD, which have had massive runs, passive components, ethernet companies, power electronic companies. You can see all the different parts of the AI supply chain — grid infrastructure, memory and storage, obviously Micron, Western Digital, SK Hynix, Pure Storage, Seagate, the cabling companies, the fuel cell companies, the PCB subsidiaries, semi-design service companies. So it's quite an interesting infographic.

## Nomura DRAM Forecast - 56:08

And then on page 17, we were quite shocked to see Nomura's latest memory or DRAM forecast. So they're projecting DRAM revenue to explode past $2 trillion by 2030. So yeah, that's trillion with a T, from around 80 billion in 2022 to over $2.06 trillion in 2030. Here's the full demand picture that they published. Production is expected to grow from 29.3 billion gigabytes in 2022 to 128.4 billion gigabytes by 2030. Shipments are expected to rise from 24.7 billion gigabytes to almost 126 billion gigabytes. Utilization rates are extremely high still, often above 100% in later years, and inventory even turns negative in absolute terms. That is a clear sign of structural tightness in the market.

So what they're saying is that the sell-off in AI was more technically driven; the actual demand hasn't really changed. The price story is just as dramatic. After the big crash in 2023 when prices fell to $1.90 per gigabyte, Nomura sees a strong recovery. They forecast $13.70 per gigabyte, roughly a 6x, in 2026, peaking near $18.60 in 2027, almost a 10x in memory prices, and then settling down at 16 to 17 by 2030, so not a big drop. The combination of much higher volume and these sustained high average selling prices is what drives this massive $2 trillion plus revenue number. This is not a cyclical recovery. The forecast assumes AI and data center demand will keep running ahead of supply for years. We'll see if that's true. Wafer capacity is growing but not nearly fast enough to match the increase in demand. High-bandwidth memory, or HBM, is the biggest driver here — it brings higher density, higher prices, and much higher profitability than regular commodity DRAM. For some context, most mainstream long-term forecasts still see the entire DRAM market well under 300 to 400 billion by 2030, and that could be off by a factor of seven.

Nomura is basically saying that it could be five to seven times larger than those conservative estimates if AI infrastructure spending keeps going at its current pace. The implications are huge for the big three players, Samsung, SK Hynix, and Micron. If even part of this comes true, memory chips could become one of the biggest profit pools in the whole semiconductor industry, which we're already seeing. Capex will need to speed up, but the pricing power looks very favorable in this scenario. Of course, these are aggressive assumptions — AI spending could slow down, new capacity could come online faster than expected, or Chinese producers could close the gap quicker — but the overall direction is clear. The old idea that DRAM is just a pure commodity cycle is being completely rewritten by AI. A $2 trillion DRAM market by 2030 is a big number that Nomura is putting out there and is describing in the red font in the DRAM supply-demand chart at the bottom of page 17. You see the wafer capacity, wafer output, utilization, shipments, inventory, inventory weeks.

## Korean Market Sell-Off & Margin Calls - 59:19

And then the following page, you can see that Goldman is quite negative on its CTA flows. They think that if the market sells off, these trend-following funds over the next month might have 71 billion to sell. Nasdaq, before the bounce last week, was on track for the worst July in 22 years on the AI sell-off, and four US hedge funds being forced sellers, and obviously margin calls in Korea as well, which leads us to page 19.

Korean stocks were down well over 44% from recent highs on Wednesday before rebounding 18% after the Leopold liquidation the following day. Kospi forward PE fell below 5 at the lows, from as high as 12 last year, simply because of earnings growth and the market selling off. Korean retail investors through Wednesday suffered a 2.15 trillion won loss in leverage ETFs. While that's roughly 14.5 billion, investors age 20 to 30 made up about 62% of the accounts that were liquidated. On July 13th alone, over 320,000 accounts were margin called and were liquidated.

To prevent a financial crisis from evolving into a social crisis, the South Korean government launched a nationwide unified debt consultation hotline at 1375. They also have installed cameras on bridges to avoid suicides. It got pretty, pretty hairy, and very sad. Korean stocks were down 11% on Tuesday alone before rebounding later in the week. And South Korea hosted an emergency meeting on Wednesday regarding the stock market crash.

Korean stock valuations post-sell-off by name, like Samsung, SK Hynix, and a lot of these are AI related. We found this table with RSIs, EV to sales, PE ratios, etc. PEs got quite cheap on a forward basis for a lot of these companies, especially in memory.

The GS high beta momentum basket was down 8% Monday, extending a two-day decline of over 15%, so it was a very, very difficult trading market. GS spec sales basically said that AI data center is clearly coming off elevated bubble-type levels with high beta momentum giving back all year-to-date gains, down 35%, which is quite shocking if you are a high-risk trader. By Tuesday, Goldman's high beta momentum pair was down almost 41%, one of the worst drawdowns outside of COVID since the great financial crisis began. GSPRHIMO is the index you can find on Bloomberg, where you can see we were down 40% in momentum stocks year to date. The Nasdaq 100's monthly drop before the Leopold rally, of 9.5%, would be its worst July since it fell 11% in July 1986. The index bounced back 5% in August 1985, but was then down 2.2% for the remainder of the year. From the end of July 1986 to July 1987, it was up 35%, however. You can see the Nasdaq rollover, still up for the year, but it has given up about half of its gains. Job growth has also been relatively weaker, which is why I'm not that worried about wage inflation, but the ADP data last week showed US private hiring has slowed for the fifth straight week, averaging just about 15,000 jobs per week through July 11th, less than half of the 35,750 weekly pace seen earlier in May. At the same time, Visa reportedly is cutting 7% of its workforce, or about 2,600 jobs, adding to signs that labor market momentum is cooling. Also on Wednesday, retail dumped — they actually dumped at the lows. It was the biggest single stock dumping day since COVID, followed by that huge bear market rally the next day. Leveraged ETF AUM has almost been completely unwound to pre-May levels in Korea. And inflows into leveraged products have also stalled ahead of regulatory checks. On the following page, on page 25, we saw that we hit the lowest close for TLT, the long bond ETF, since the great financial crisis.

## SK Hynix Earnings - 1:04:03

In terms of SK Hynix earnings, we'll get back to those in just one second. I'm just going to get some water.

So, SK Hynix, which is one of the three biggest memory players in the world, reported a slight earnings miss. Q2 26 revenues came in at 54.6 billion, still a very impressive number versus overhyped expectations of 57.7 billion, which was still up 257% year-over-year and 51% up quarter-over-quarter. While operating profit came in at 41.6 billion versus 44.2 billion expected, up 557% year-over-year. Operating margin hit a record 76% with gross margins at 83% and EBITDA of 44.4 billion, or 81% margin. Pricing remained extremely strong with DRAM ASP up 30% Q over Q and NAND ASP up 50% Q over Q. While Q3 guidance calls for DRAM bit growth of about 10% Q over Q and NAND bit growth in the low single digits.

SK Hynix also said that HBM4 shipments began in Q2 with a full ramp planned in 2H, with HBM4E samples being supplied to a major customer in the first half of this year. Net profit surged to 64.6 billion boosted by 43.5 billion of investment asset gains, meaning net income exceeded revenue due to 42.8 billion of non-operating profit. SK basically mentioned that they have finalized long-term agreement negotiations with about 10 customers just like Micron did, including key clients, and are continuing to add discussions with major industry customers. SK sold off in part because management didn't sound enthusiastic enough about capital returns, but they had just printed the ADR. Under US securities rules, they were effectively constrained from announcing material new information not included in the prospectus for 25 days after the offering. So they're limited in what they could actually say.

In terms of quick math, SK Hynix at the after-hours lows, including the 48 billion of cash, the enterprise value is roughly 650 billion US. The sell side was looking at 221 billion of income next year, not including their 15% stake in Kioxia, and 95 billion of net income they're expecting to generate in Q3 and Q4 of this year that you would add. So the stock was trading at less than three times EPS before that additional income and their Kioxia stake. Keep in mind that SK Hynix is the global leader in HBM, which is crucial for the AI build-out, and China can't do HBM alone. CXMT does commodity DRAM, not the HBM DRAM that you need, which means that China cannot flood the market with HBM and bring down prices. There is a significant demand-supply imbalance for another two or three years due to the oligopoly of SK Hynix, Micron, and Samsung, which means that SK Hynix will continue to report record-breaking results and print gobs of cash.

Also last week, the South Korean Finance Minister apologized for his rushed view of single-stock leverage ETFs. Given the Korean gambler mentality, single leverage ETFs made up a very large percentage of the stock market. And Minister Ku Young-chil apologized to a lawmaker on Wednesday after two consecutive circuit breaker halts in Korean stocks, saying that the government regretted introducing single-stock leverage ETFs without sufficient deliberation. Ku said that officials are internally studying market stabilization measures and that further adjustments to rules on single-stock leverage ETFs are possible to address volatility. The Kospi plunged as much as 12% intraday, following an earlier session drop of 6%, before rallying later, narrowing around an 8% decline and then rallying the following day. SK Hynix fell as much as 17% after earnings before trimming losses to only 8%. There's basically an Ireland-traded 3x levered SK Hynix ETP that's actually down 96% since mid-June, basically in a month. And that's why I've always told you guys that leveraged ETFs are a trap, especially the single-stock ones.

You can see that screenshot of the 3x levered Irish product on page 26. The South Korean market basically saw the highest 90-day rolling realized volatility in history over the past few days. South Korea's entire stock market just basically triggered a trading halt as billions were vaporized in minutes. But here's why the businesses behind the crash are actually stronger than ever. Samsung Electronics and SK Hynix together account for roughly half of the entire Kospi's market value, over 60%. So when those two stocks swing violently, the whole index swings with them, and much of the recent volatility has been amplified by the heavy selling in leveraged single-stock ETFs tracking exactly those two names. The trigger this time was because of China — because CXMT just had a blockbuster Shanghai debut and made it the most valuable listed company in mainland China, which spooked investors about the memory duopoly's long-term competitive position in Asia. But CXMT is not the threat that headlines make it sound like. It holds only about 6-9% of global DRAM shipments, mostly in China, compared to Samsung's 36% and SK Hynix's 29%, and Counterpoint Research estimates that CXMT would need at least 15% global market share to be competitive long-term, a level that isn't projected to be reached even in 2028.

Most of its output is basically basic commodity DRAM, not the high-margin HBM that AI servers actually need, and CXMT's manufacturing costs remain over 30% higher per bit than industry leaders, despite lower wages. So it has no real ability to undercut or displace Samsung and SK Hynix in the segment that matters the most. In fact, current DRAM supply is so tight for AI demand that even CXMT has stopped competing on price and is riding the shortage to raise its own margins rather than steal share through discounting. This is simply a valuation and positioning reset, not a breakdown of the AI story. Korean equities had simply become one of the most crowded AI trades globally, so it didn't take much to trigger heavy profit-taking once sentiment turned.

Samsung actually reported a blockbuster forecast of a 19-fold jump in quarterly operating profit around the same time its stock was falling, and memory pricing continued to strengthen. Now, some may argue that this was because of the AI-driven margin calls in both Korea and in US hedge funds.

Korea's year-to-date price return has been driven almost entirely by rising earnings expectations. And its earnings expectations jump is by far the largest of any market on Earth, well ahead of Taiwan, global semis, and every developed market listed. Both companies sit at the center of the global memory supply chain that every AI data center depends on, and their $520 billion joint investment into four new memory plants in the US shows that they're doubling down on long-term HBM and DRAM capacity rather than retreating. Memory pricing has continued to strengthen even through the sell-off and Samsung's own profit guidance points to some of the strongest quarterly results in the company's history. Now, CXMT trades at 30 times while the incumbents trade at 4 to 6 times, and that's likely not sustainable. Either CXMT needs to come down, or Micron and SK Hynix and Samsung need to rise a little bit.

Korea's Kospi is basically down 40% from the peak after the SK Hynix earnings we talked about. It did bounce a little last week. You can see that sharp drop on page 28. You can also see that global chip stocks were tracking the worst monthly sell-off since 2022 before the Leopold bounce.

On page 29, you can read about Trump on space regulation. The administration is proposing a change that will allow the FAA to waive certain environmental review requirements for commercial launch sites, which should help Elon — rocket launches, spacecraft re-entries, according to the Wall Street Journal. The move could speed up approvals as annual launches and re-entries are expected to rise from only 214 this fiscal year to more than 500 per year over the next decade. SpaceX, Rocket Lab, Blue Origin, Stoke Space could benefit, though the proposal still has to go through a public comment period.

Household equity exposure before the volatility of last week was near all-time highs — household equity exposure data we have going back to 1945 according to Nikko Securities. And so this over-positioning, from a technical perspective, did show that people were overexposed to equities.

On page 30, you can see that earnings have driven the bulk of 2026 returns, so it's not as speculative as 2021 or as 1999, but the market's still been very volatile. In terms of corporate buybacks, Goldman is estimating that 31% of S&P 500 companies are in their open window to do stock buybacks. By the end of next week, it should be about 53% of companies, which means that half of companies can now buy back their own stock, which is usually a positive technical.

On the following page, on page 31, you can see PE valuations relative to their history. You can look at communication services being relatively cheap. But names in the red like industrials being historically expensive. So one of the reasons why I think Caterpillar is overdone.

Then the following page, you can see the top 10 companies by market cap, how they've evolved over the years. In 1990 it was Exxon Mobil, General Electric, IBM, AT&T, Signal, Altria, Royal Dutch Petroleum, Merck, P&G, and Bristol-Myers Squibb. Today it's NVIDIA, Google, Apple, Microsoft, Amazon, Broadcom, Tesla, Meta, Micron, and Eli Lilly. And Apple's been quite impressive as it's been in the top five since 2015.

China is basically rejecting US tariffs tied to forced labor allegations after the Supreme Court turned down US tariffs. The US tried to put on new tariffs by alleging labor violations, and obviously, all these countries are going to challenge that.

On the following page, on page 33, the chance for Iran negotiations on Monday essentially faded in the beginning of the week as bombings accelerated, but by Sunday, peace rumors surfaced again, which were denied by Iran. At the beginning of the week, Trump said that the US could return to very strong military action if negotiations fail, adding that he will give diplomacy not much time. Talks are focused on reopening the Strait of Hormuz and restarting broader nuclear negotiations. Trump was scheduled to meet Israeli Prime Minister Netanyahu on Tuesday. And you effectively saw oil prices and yields fall on Monday and then they picked back up. Hormuz insurance jumped 1% to 9% as well, which is a very big jump. And then we saw IRGC attacks on July 28, 2026, where US Central Command said that IRGC forces launched multiple ballistic missiles in a surprise attack on US forces stationed in the Middle East, but all those missiles were intercepted. CENTCOM said that all Iranian missiles were successfully intercepted and that US forces remain on high alert, ready to respond. Oil was up 5% after hours on the 28th, even though it went down 4% on earlier reports that Egypt, Qatar, and Pakistan were working on bringing the original MOU between the US and Iran back. There have been no updates on the original MOU coming back yet.

And on the 29th of July, in a Wednesday Fox News interview, Trump promised a hard-hitting US response to the surprise attack on American troops at the Jordan base, reportedly involving Iranian ballistic missiles or proxies. He stated that we are going to beat the f-ing shit out of them, signaling imminent action in line with administration plans. Markets dipped on the news about a potential escalation while coverage highlighting ongoing regional tensions and prior US casualties in similar incidents grew into the spotlight.

Oil imports are still very low from China. China's helping save the world market by just importing less because they had so much in reserves — they had like 1-odd trillion in oil reserves — and as they, from May to July, cut their imports by more than half, that's been helping the oil market absorb some of this lack of supply. Also places like Vietnam with a three-day work week where they already cut back consumption because they had no choice, and that's weighing on demand as well. So offsetting some of this supply shock.

China basically lied to the US about not supplying Iran with missiles — who would have thought? Iran is expected to receive an initial shipment of up to 400 Chinese-made MANPADs within weeks for 60 to 70 million. And I doubt that this is going to stop soon. Bombings continued over the week with the US contemplating power and energy targets in Iran before pausing on Sunday, which I have Trump's tweet on page 34 to explain. Effectively he says that the US is locked and loaded and ready to go against the IRGC. Despite this, we've been asked by Iran and other Middle Eastern countries to hold off any attack, that the perimeters of a deal have been agreed to. Basically other Middle Eastern countries don't want to get bombed. So what Trump is saying he wants is the total opening of the Hormuz Strait and an end to Iran's nuclear threat. I don't think Iran wants to give up either one. So we'll continue to monitor this.

China has begun limited production of homegrown immersion DUV lithography machines, which is bearish for ASML slightly. An unnamed state-backed Shanghai company is targeting roughly five systems this year and 20 by next year with initial deliveries planned for SMIC, Hua Hong, and CXMT, basically all Asian companies. For comparison, ASML shipped 131 immersion DUV systems last year, so this would be less than 5%. The machines use mostly Chinese components, though some key components still come from Japan. They also trail ASML's systems in performance and build quality and may require months of testing before entering chip makers' production lines. So this is not an immediate threat to ASML, but it marks progress in China's efforts to localize critical chip-making equipment as Western restrictions tighten. China's domestic EUV program remains at the prototype stage and is likely years from producing working HBM chips. So this is from The Information.

So again, the semi sell-off on Monday was driven by a mix of China's lithography headlines, memory mystery, and the Fed jitters — not because demand for AI infrastructure has suddenly disappeared. So you can read the details of that on page 35; on page 36 is a heat map of names like NVIDIA down 5%, Micron and AMD down 5 and 7%, on the big sell-off day last week. Google's capex was very aggressive in 2028, but now I think they're going to follow Microsoft and just be a little bit more mindful of free cash flow, but we'll see. Google for now is a very aggressive target in 2028 to deploy its in-house TPUs based on channel checks. Google plans to have 12 to 15 million TPUs in 2028. Entering 2028, Google's TPUs will enter the V9 generation with four compute dies, suggesting their capacity consumption will more than double in 2028 and 2027.

Although we do not have a detailed allocation yet, we think that it's difficult to reach Google's target with TSMC alone and that Intel supply is a must by 2028, which creates another unknown. Currently, there is still a lot of noise on Intel's EMIB progress, especially its learning curve improvement. So we'll see Intel becoming more positive in capacity expansion, which is positive to the Google supply chain.

Intel's EMIB capacity will reach 10 to 12K by the end of 2026 and the plan is to double its capacity to 24 to 25K per month by the end of 27. Intel has already announced its capex increase, although there's no detailed breakdown of its capex spending. The CFO highlighted that a great portion of the new capex is earmarked for expanding advanced packaging capabilities such as the EMIB. We think that Intel's major supplier Ibiden's upcoming earnings call and potential capex increase in the next two years will be another highlight.

Now, data centers have been profitable according to Morgan Stanley. Their high conviction view is that demand for compute is likely to significantly exceed supply for many years to come. They are fundamentally bullish on the rate of improvement in AI capabilities, the benefits of AI adoption and associated capex. The ROI on AI capex is quite attractive. Their token economics model suggests both large LLMs and more efficient models generate strong ROIs on the underlying AI infrastructure. You can see that on page 37. Their intelligence factory model projects 58 to 90% margins for token sales depending on GPU. So if you look at the Feynman data center, you're looking at 85 to 90% margins, the Rubin data center, almost 80% margins, and the Blackwell data center, almost 60% gross margins. So Morgan Stanley is saying that the GPU cloud business is extremely profitable across hyperscalers. GPU rental businesses are generating 70% incremental margins or a 30% ROIC. So you look at revenue, OPEX, incremental EBIT, and then tax and NOPAT, so that shows a 31% ROIC. And if you look at the following page, AI chatbots like ChatGPT and Claude have triggered a scramble for high-bandwidth memory. And even the Chinese models used in the US will trigger more demand for memory because even if your inference is cheaper, your demand for memory is still going to be the same. J.P. Morgan's Schulsky's internet stock sentiment index — what are the names that hedge funds are consensus long and which ones are the names that outperform and underperform. I think the battleground is quite interesting with DoorDash, Shopify, Reddit.

## AI Update - 1:26:59

Before we get into Jensen Huang's meeting with Howard Lutnick on AI, I'm just going to get a little bit more water.

So, in terms of the AI update, while as we're speaking by the way, the S&P is up 50 bips. Nasdaq's up a little bit more, 80 bips. And the Russell is up 54 bips.

So, we will touch on Hood, Big Cat, and we'll do that in the Q&A.

So on the AI update, Jensen Huang reportedly met with Commerce Secretary Howard Lutnick as NVIDIA faces growing scrutiny over chip exports to China. The meeting comes while the Commerce Department investigates potential violations tied to NVIDIA's Blackwell shipments to China. We've always talked about how there have been way too many shipments to Singapore, and where have those shipments leaked? Singapore doesn't have enough data centers to support all those GPU shipments.

Huang is also meeting with lawmakers from both parties as Washington prepares a broader AI framework covering advanced models, open source AI, China, expected this week.

OpenAI and Anthropic are reportedly lobbying the Trump administration ahead of an August 1 deadline for deciding which frontier AI models should face government evaluation. The companies want rival models like Meta and SpaceX AI included as well. Qualifying models would be subject to review at least 30 days before release if they raise cybersecurity or national security concerns. OpenAI and Anthropic are also warning about Chinese open-source models allegedly built in part using their outputs, citing privacy, IP theft, and cybersecurity risks.

Moonshot AI is reportedly looking for more NVIDIA Blackwell chips as it discusses Kimi 4, a Chinese model expected to be significantly larger than its 2.8 trillion parameter Kimi K3, per The Information. K3 was reportedly trained partly in China using Blackwell chips accessed through multiple Chinese cloud providers, with Moonshot linking separate chip servers across providers and data centers because no single provider had enough capacity. Inference is now another constraint, with Moonshot pausing new subscriptions within 48 hours of launch as demand overwhelmed capacity. The company reportedly uses NVIDIA H20 chips for inference and recommends at least 64 chips to host K3.

The report also says Alibaba trained its 2.4 trillion parameter Qwen 3.8 Max using NVIDIA chips including Blackwell. So obviously these Blackwell chips are making their way into China, either through cloud or directly, and that's a US security concern.

## Bloom Energy Earnings - 1:32:15

Bloom Energy, which we took a small stake in, surged 13% in after-hours trading after it beat earnings and then it sold off all the way back down to 160, and then it was up like 30% the next day after the Leopold sale of equities to Citadel, which we'll talk about in a moment.

It crushed 2Q 26 earnings with revenue growth of 166% to 1.07 billion versus 826 million expected. And adjusted EPS jumping 680% to 0.78 versus 0.41 expected. Product revenue surged 215% to 935 million, while adjusted gross margin expanded 604 basis points to 34.3% from 28%.

Bloom also raised its fiscal 26 guidance. Now expecting revenue of 3.9 to 4.2 billion versus 3.73 billion before, and adjusted EPS of 2.55 to 2.85 per share versus 2.16 expected.

Management said that all major US hyperscalers plus over a dozen neo-clouds, AI labs, and data center operators have approved Bloom's power solutions for AI factories, calling Bloom a new standard for AI on-site power. In terms of the annual forecast at Bloom Energy, it sees revenue of 3.9 to 4.2 billion, up from 3.4 to 3.8 billion. Sees adjusted gross margin at 34% now versus an estimate of 33.3%. Second quarter results you can see in the middle of page 40. Adjusted EBITDA of 253 million for the quarter versus an estimate of 152 million was quite impressive.

## Tech Unwind and Market Bottom - 1:34:10

The tech unwind note by Rebecca Chong. Both Goldman and Rebecca Chong, I think they did call the bottom in tech stocks quite accurately. And maybe it's because they knew about the liquidation of the famed hedge fund wunderkind, Leopold, who ran Situational Awareness.

We talk about capitulation selling and retail capitulation. Semiconductor multiples had fallen to multi-year lows. And again, we should be using average PE multiples, but we're using the forward PE, which might be at peak margins — but you look at SK Hynix, it was at four, Micron was at five, SanDisk was at five before the rally at the end of last week. Qualcomm was at 15, NVIDIA was at 15, down from 25.

One thing we saw interesting ahead of SpaceX earnings next week was that SpaceX long-term bonds — we mentioned these were shorts. It's hard to short these bonds unless you're institutional, but the SpaceX 6.65% bonds due 2056, I think we said there were shorts around like 98, 99. They're all the way down at 86.

## Germany-China Trade War Tensions - 1:35:52

Let's briefly talk about a German-China trade war. So Germany's effectively been in a recession for the last three years because China's been taking market share in terms of industrial manufacturing and auto. I mean, BYD cars and Chinese EVs are just so much cheaper than EVs made in Europe. And in order to avoid continued job losses — you saw 100,000 job losses at Volkswagen over the last two weeks — the Germans need to do something to protect their industry.

So German officials are working behind the scenes to identify Chinese economic vulnerabilities that they could exploit if the European Union finds itself in a trade war with the world's largest economy, like the US did. The German government is informally mapping China's weaknesses by analyzing trade flows, supply chains, and company-level data. Right after this, we'll talk about Apple, Amazon, Microsoft earnings, etcetera.

The exercise is aimed at identifying where China remains dependent on German and European technology, specialized components, and industrial know-how, and where Beijing could face pressure in a potential economic confrontation. Initial findings include German companies Trumpf SE and Carl Zeiss AG, key suppliers of technologies used by ASML Holding NV to manufacture machines that make advanced semiconductors, as possible vulnerabilities in China's supply chain.

The report may also include semiconductor firms such as Siltronic AG, Aixtron, and Suss Microtec, whose highly specialized technology supplies critical positions across the supply chain. Chancellor Friedrich Merz has recently hardened his stance on China, concluding that Germany risks being caught between the strategic rivalry of the US and China unless it develops greater resilience and leverage of its own.

EU leaders last month tasked the European Commission, which handles trade matters for the bloc, with presenting a slate of new trade defense measures to confront what they deemed to be an unsustainable deficit with China that exceeds 1 billion per day. Yes, there are dependencies, but these dependencies are mutual. That also applies to issues such as rare earths. Merz said this month, when asked how to deal with China, "We can only withstand unfair competition if we defend ourselves against it."

The effort to prepare contingency measures against China's weaponization of critical raw materials, as it did last year in the case involving Nexperia, forms part of 34 non-public measures formulated in Germany's National Security Council aimed at strengthening resilience and reducing strategic dependencies. You guys can read the rest of this, but it's quite an interesting piece that shows China's trade surplus with Germany hitting record highs. You can see the German suppliers of strategic technologies to China that might be pulling back. The list of that is on page 43.

Other significant sectors outside of high technology that employ millions of Chinese workers include steel, chemicals, plastics, textiles, toys, and household appliances. Those are areas that could be targeted by Germany.

## The Fed vs. Geopolitical Risks - 1:39:26

And the piece on page 44 is also quite interesting. I'm going to go through it very briefly. It talks about how the war in Iran — the market's been looking past this and the VIX has been relatively subdued, and it was really only after Warsh's speech that we saw a VIX spike. So we had five months of war with the VIX never closing consistently above 17.5 in July, but one FOMC meeting under presidential pressure resulted in a VIX spike to 21. And told you exactly where the real risk to the market lives.

So you start with what happened before 2:00 in the FOMC meeting because it makes what happened after 2:00 legible. On Wednesday, Iran and the US were shooting at each other. Iranian strikes hit Saudi energy infrastructure. American and Saudi forces struck back jointly. Tehran had just rejected Oman's proposal for shared management of the Strait of Hormuz. Crude was on its way to a 6.7% single-day gain, and the VIX was actually down, drifting under 18.4, comfortably inside the same behavior this newsletter has documented through every phase of this war. The shock routed to oil volatility, to freight, to rates, while broad equity gauges shrugged. At mid-morning on the FOMC day, the war was doing what five months of war has always done to the index: approximately nothing.

Then the Federal Reserve held interest rates steady. Three committee members dissented in favor of a hike, and Kevin Warsh gave a press conference. By the close, VIX was at 21, up 13.5%, its first close above the entire war restart band. The Dow closed down 1,153 points, 2.2%, its worst day since April 2025. The S&P was down 1.52%, the Nasdaq was down more than 10% off its high, a correction. The 30-year Treasury was up 10 to 12 basis points to 5.21%. This was its highest level since 2007. And mark this one: the dollar was down against nearly every major currency while yields surged. Very strange.

Oil was up 7%, but it was a footnote. And the Federal Reserve was the headline. CNBC's summary of the day was blunt: "Stocks fell mostly because the bond market signaled the Fed could be falling behind the inflation fight."

So here is the claim we tend to make. The FOMC day's tape shows that the threat to this market was never really the war, it was the erosion of the institution — the Fed — that anchors almost every asset price in it. A president publicly pressuring the central bank, a chair caught between the pressure and a board of other FOMC members pricing hikes, and the long bond charging 2007 prices for the uncertainty. That configuration did more damage in two hours than Iran managed in five months. The market has now told you its hierarchy of threats and it's time to take it seriously. That's why the bond market, like I said at the top of this piece, is the biggest threat to the market.

And it's why all these rallies on Trump jawboning peace — he knows that the war is a risk and inflation is a risk. And frankly, if the market perceives that he's losing control of the Fed, you have to short everything.

So you can see that VIX spike versus other VIX spike days on page 44. An attack on the Federal Reserve has no address. The Fed's credibility is not a sector, it is a denominator. It sets the discount rate under every cash flow, the risk-free and every risk premium, the anchor of the currency whose cash flows are denominated in it. You cannot rotate out of the denominator. A threat to it does not get routed, it gets priced as a correlation, which is precisely what FOMC day looked like. Stocks down, bonds down together, with a safe haven bid landing in gold, up two-thirds of a percent to 4126 with yields surging, which is not how gold usually trades when yields are surging. It's how gold trades on a credibility issue.

That is why the same market that absorbed 140-target strike night without the VIX closing above 17.6 broke on a rate hold. The war is a shock the system has an organ for; it understands the Fed is the system itself. Skeptics will say yields rose because oil rose, but yields rose well after oil rose. They rose after the FOMC discussion. You can see how the 30-year Treasury, the 10-year Treasury, the US dollar, the S&P, gold, and WTI traded all on page 45.

Now, when long yields rise on an ordinary inflation repricing, the dollar rises with them. Higher expected rates attract capital due to interest rate differentials. Yields up, currency down, stocks down in one session is not how reserve currency markets trade on data. So the market was actually challenging the reserve currency status as well as challenging the Fed's credibility.

This is how emerging markets trade on politics. The signature of investors charging a sovereign more while trusting it less. In a war restart piece, we wrote that the rates channel was the fragile point, that a break arriving through it would be a correlation story, and that readers should watch the curve, not the tanker count. The market fired through the long end with the dollar confirming the diagnosis.

Look also at where on the curve it fired. The two-year barely moved, vendors split between down four basis points and up one. The 30-year surged 10 to 12. That bear steepener is the whole political economy of this moment drawn into data points. The short end of the curve, which is where the Federal Reserve interest rate and the two-year are, was priced for a Fed whose next move is capped by presidential pressure. And the long end is basically charging everyone — pushing mortgage rates higher, every mortgage, every corporate borrower, the Treasury itself — for the credibility cost of that arrangement. Politics wins the front end of the curve, but the country pays at the back end of the curve.

On July 10th, mid restart, the VIX term structure was a staircase. One-day vol was at 9.9, nine-day vol was 11.15, spot was at 15.02, and the front future was at 17.55, the three month at 18.57. So you had an 8.7 point upslope from tomorrow to autumn. The vol curve had a relaxed center because dispersion had twice proven it absorbs the war, paying up steadily for it later because the genuinely unresolved risks — the Fed's reaction function, the inflation end game, and the September Fed meeting — all live out the curve.

By Monday, the 27th, pre-Fed, the staircase had flattened into a ramp. One-day vol was 15.43, nine-day was 17.62, spot was at 18.58, and three month was at 20.5. The slope collapsed from 8.7 points to five, which means that the deferred risk now had a date. The market pulled the later risk it had been paying for into the front of the curve and parked it on Wednesday at 2:00 PM. The market, in terms of options, was already worried about this meeting. Bond markets came into pricing roughly 40% odds of a hike against a president demanding cuts. Every options desk could see the collision scheduled.

And on the 29th of July, spot closed at 20.66, above every single point on Monday's curve except for the three month. The payment came due. Whether Wednesday's future settlement shows full front backwardation or a hair of contango, the shape's message is identical. The risk the surface spent all of July prepaying for arrived, and it arrived wearing a Federal Reserve lanyard, not a uniform. Meanwhile, the skew index fixed between 144 and 147 for the entire month. Never budged, because the tail was never priced for Iran. The center just rose to meet the tail. The surface has been telling you which risk it feared, and yesterday the center agreed that the biggest risk is the Fed.

## Kevin Warsh's Impossible Seat - 1:48:16

So Warsh has an impossible seat. Kevin Warsh gave his second press conference as chair, and the market's verdict — 10 basis points on the 30-year — showed that he failed. Consider his position. Two days before the meeting, the President of the United States said that rates should be lowered, that America should have the lowest interest rates in the world, while praising Warsh as fantastic, after appointing him, and then describing Warsh's own board as very political, with perhaps bad intentions, adding the phrase that will follow this Fed chair around: "Kevin's fantastic, but he's got a board."

The White House framing is explicit: the chair is our man, the institution is the obstacle. The NEC director called his tenure already a home run and said that "We trust his judgment," while insisting in the same breath that everyone expects him to be independent. When independence becomes a talking point, it has already become a question. Inside the building, three of Warsh's colleagues dissented hawkish. They wanted to hike, against a president demanding cuts. With markets pricing in 40% odds, you would get one in roughly eight percent for September.

Warsh split the difference. He held, offered no forward guidance as a matter of stated philosophy, and told the room that tightening in financial markets is doing some of the Fed's job, that the bond market would — and in effect, sold the story. The bond market then told it: a 30-year at 5.21%, the highest since before the financial crisis, is the market doing the Fed's job at the Fed's expense, charging a term premium for a central bank that looks caught between its own hawks and the president, communicating less precisely when its independence is most in doubt. A chair can survive being disliked by the markets. What a chair cannot survive is markets concluding that the policy path is being negotiated somewhere other than the committee room. Wednesday's steepener says that the conclusion is being drafted.

## Fed Transmission Mechanism Failure - 1:50:20

So the transmission that no longer transmits. There's a deeper layer, basically saying that one of the biggest risks is whether inflation can be tamed. And one of the reasons why the Fed actually can't control inflation is because the top 10% of earners in this country do 50% of the spending, and they're not as affected by interest rates as the bottom.

The theory of rate hikes is that expensive money curtails borrowing. Curtailed borrowing curtails spending and cooled demand tames prices. That chain assumes the marginal dollar of demand is borrowed. It no longer is. The top 10% of American earners now account for roughly half of all consumer spending, which is a record share according to Moody's Economics. And I believe that's up from as low as 35% 20 years ago.

This rich cohort is functionally rate immune. Their mortgages are fixed, locked at pandemic-era rates the Fed can't reach. Their spending is keyed to portfolios and home values sitting near records, not priced for credit. And higher interest rates actually raise their income on the cash and bills they have. So for the asset-rich, the tightening is actually a pay raise. So it's a boomer stimulus.

The households the Fed rate can actually squeeze, the borrowing 90%, are not the ones driving most of the demand. The evidence is on the Fed's own letterhead. The federal funds rate has sat between 3.5 to 3.75% for seven straight months, and yesterday's statement conceded that inflation still remains elevated. The old channel is not gripping.

What still grips is the long end — mortgage rates, corporate borrowing costs, and the discounting of every long-duration asset. And as supply of credit has been increasing with trillions of dollars of AI spending and a private credit crisis, it's only a matter of time before financial tightening across the capital markets gets worse.

The long end doesn't answer to the FOMC, it answers to credibility, to the market's belief that the institution setting the anchor is independent, competent, and durable. Which means that the one transmission channel this economy has left is precisely the one that presidential pressure on the Fed degrades. Attack the institution and you're not merely risking bad optics, you're detonating the last working part of the machine — and on Wednesday, at 5.21% on the 30-year, you watched the detonation get marked into the market.

So the two tensions were: first, oil was up 6.7% and some of the long-end move is legitimately war inflation. I will not pretend the decomposition is clean, but the dollar is the tiebreaker. Oil inflation repricings do not weaken the currency against the entire G10. And five months of oil shocks never moved the 30-year like this. The market had a war premium already fully built in. What it added on Wednesday was an institution premium.

Second, one day is one day. The VIX at 20.66 is not the VIX at 31, where round one peaked. A week of calm Fed speak could pull the center back into the teens and make it a one-day tantrum. The framework does not claim that the regime has flipped yet; it claims that the market has now shown you, side by side in a single session, which class of threat it can route and which it cannot, and priced the difference at 10 Dow points to one.

So the falsifiable markers on the record: the US was trading like an EM bond market — long yields up, dollar down, stocks down together, clustering on the Fed and independence headlines rather than the data. The 2s30s steepener extending, a sustained VIX regime above 25 whose catalysts are institutional. The combination means that a correlation repricing is underway and the dispersion era playbook is subordinate to it.

Now on the refutation, taking the other side, there could be a Warsh re-anchoring through restored guidance, a hike the board's hawks clearly own, or visible daylight between the Fed and the White House, followed by the 30-year retracing, the dollar firming on yield strength, and the VIX settling into the teens. If that happens, yes, the FOMC meeting was a warning shot the institution absorbed. And I will write that piece too. Watch the 30-year and the dollar and the 10-year. I think the 10-year is actually more important than the 30-year. The tanker count was never the tell.

## Microsoft and Meta Earnings - 1:55:33

Now let's talk about Meta and Microsoft earnings calls. If you look at Microsoft, it crushed earnings. Satya was saying that expect free cash flow to be positive in 2027 despite the increase in capex, which was extremely positive for the AI buildout that it funded by operating income. Free cash flow was 19.6 billion reflecting higher capex, but it was still very positive. Quarterly capex was 41 billion, roughly two-thirds were short-lived assets, primarily CPUs and GPUs required for AI. It expects capex spend will be roughly 50 billion for next quarter, which was in line. Capex guidance should be 117.5 billion for 2027, roughly the same, not increased like Meta. It extended the estimated useful life for its data centers from 15 to 25 years. It will be amongst the first cloud providers to deploy next generation racks based on AMD's Helios and NVIDIA's Vera Rubin. Customer demand continues to exceed available capacity.

Now Meta was the opposite. Meta actually increased capex to 135 to 145 billion from 125 to 145. It justified this by saying it's receiving offers of significant premium for its existing compute, which should be positive for neo-clouds like IREN, Nebius, CoreWeave. It expects a significant portion of compute like the one gigawatt data center in El Paso to develop internal models. Meta has multiple ROI-positive uses for additional compute across its core business. Meta finally said, we believe that overall industry capacity is going to remain tight for the foreseeable future, given AI demand compute growth. Industry has underbuilt historically for the wave of AI adoption, making existing capacity extremely valuable, according to Susan Li.

Now, Microsoft and Google have had largely sustaining AI capex buildout while remaining free cash flow positive. Meta has actually reduced its cash flow to negative, but flags compute demand far exceeding supply. All three hyperscalers' capex is largely in line with Google hiking capex figures. The AI sell-off seems extremely overblown now given hyperscalers are either keeping spending flat or rising. Compute scarcity is visible through every single hyperscaler earnings call.

In terms of Microsoft earnings, Microsoft posted strong results for its quarterly earnings on Wednesday, beating expectations and showing strong growth in its cloud computing platform and a boost in paid AI users. It earned 90 billion, or 4.81 a share, in the second quarter. That's up 18% from the same period last year. Analysts surveyed by FactSet expected it to earn 4.24 a share, which is much lower. Microsoft Cloud revenue was 60 billion, up 27% year over year. That reflects the demand across Microsoft's cloud computing platform, Azure, which we'll show you a graph of the cloud growth, as well as its first-party and third-party AI applications and services. Azure and other cloud services revenue increased 43%. For the full fiscal year, which closed out at the end of June, the company brought in 331.8 billion in revenue, unbelievable. Azure revenue surpassed 100 billion on a run rate basis for the first time, and Microsoft 365 Copilot reached over 30 million paid seats, reflecting confidence customers are placing in Microsoft to power their AI transformation.

The concurrent Azure and Copilot growth indicate that Microsoft is winning both fronts, according to Michael J. Wolf, founder and CEO of Activate Consulting. Microsoft is doing so by supplying cloud infrastructure for enterprise AI while monetizing the AI tools embedded in — one second here.

Anyway, Microsoft stock rose about 9% to 426 in after-hours trading. Then it was up 13% the following day as people were relieved. Satya Nadella showed ROIC on hyperscaler capex using an app he built, on 357 billion of cash capex, to estimate the ROI across the industry. And so the market was also relieved to see that.

In terms of Meta earnings, Meta's revenue guidance also missed market expectations as its free cash flow plunged. So Meta stock fell all the way to like 530 from almost 600. Meta Platforms gave a disappointing quarterly revenue forecast. The social media giant also reported its lowest free cash flow in years, signaling ballooning expenses for AI. Its AI capex could be 145 billion this year. Meta relies on its broad internet advertising business to finance these investments, but shareholders are growing increasingly antsy over Meta's ability to bring sales and profit growth at a fast enough pace. Investors balked earlier this year when Zuckerberg increased projected capital expenditures on AI, and many of them rue Meta's heavy spending on virtual reality in the metaverse, where it had to cut everything. That cost tens of billions of dollars with almost a zero return.

On a call with investors on Wednesday, Zuckerberg offered repeated reassurances that his focus on AI will ultimately pay off. Basically said that I get this is a big investment and it's a big bet. We see the technology working. We're happy with the trajectory of the lab. I'm excited about the products that are coming, and we believe that this is going to be a big thing. Third quarter revenue will be 61 to 64 billion, Meta said, with the midline of that range below the average estimate of 63.2 billion, so there's also a revenue forecast miss, outlook miss. Meta is spending hundreds of billions of dollars to compete against American tech rivals like Alphabet, OpenAI, and Anthropic in a race to develop leading AI models and products. It is one of the tech industry's heaviest spenders when it comes to AI data centers, but it has yet to show any type of ROI like the cloud backlog you see at Google and Microsoft.

It did announce a partnership with BlackRock on a $14 billion complex in El Paso. It's building another data center in rural Louisiana that is expected to cost upwards of 250 billion, which is unreal. In part because it doesn't yet have a cloud computing business and its AI products have at times been considered less competitive than other AI labs' work, Meta has faced recurring investor skepticism that it will ever recoup its spending. Meta announced several new AI business lines in recent months, including a consumer chatbot subscription and a pay-to-use AI model for developers, although those are in early stages. The cloud opportunity is what's interesting. Basically, Meta said that it would start a cloud computing business to compete with Amazon's AWS, Microsoft's Azure, and Google.

Zuckerberg will sell computing power to other companies. He said that a substantial amount of Meta's computing power currently goes towards training its own AI models, which is necessary for being a leading AI lab. He also said that Meta has a large number of offers from companies interested in buying its computing power at a meaningful premium over what Meta spent to acquire it. That has created an opportunity, he added, saying that Meta must now think through the trade-off of selling the computing power it has versus continuing to use it for its own products and services, which suck. These calculations are happening at the same time Meta is also buying computing power from independent data center operators called neo-clouds as well. "There is just nowhere near enough compute for all the demand," he added, when he talked about the idea of Meta being both a buyer and a seller. Bloomberg earlier reported on Meta's plans to develop a cloud computing business in the prior week. Meta adjusted its full-year capex to 130 to 145 billion, lifting the bottom end from the previous projection of 125.

The company's free cash flow in the second quarter fell to only 784 million, which is the lowest level since the third quarter of 2022, right before it did its big cost-cutting plan. Much of its investment is fueled by advertising that Meta runs on its flagship networking products like Facebook and Instagram. Meta reported revenue of 60.8 billion for the second quarter, above the 60.3 billion people expected. The biggest US tech firms plan to spend 725 billion on capex and Meta's about a fourth of that. In terms of legal fees, Meta is also facing a big lawsuit, like 165 to 169 billion of penalties. Thousands of individuals and US schools are suing Meta and other major social media companies over allegations that their products are addictive and harmful to minors, which we all know. Earlier this year, a jury found Meta and Google liable for a young woman's mental health struggles, giving her 6 million in damages, which is wild.

## DoorDash Unit Economics - 2:05:49

DoorDash I think is quite an interesting company. Someone posted the DoorDash unit economics and it shows you how thin the residual is that they earn relative to the gross dollars sloshing through the system. So if DoorDash touches $100, i.e. the marketplace GOV, only 13% of that is theirs to spend, and of that 13, they have to spend on marketing, engineers, refunds, insurance, and payment processing. So they're really only left with 80 cents of EBITDA plus SBC, or 46 cents of EBITDA, which versus an average $32 order, they're making less than 2% per order. But as they tweak the business and as it grows, that 2.5% EBITDA margin could easily grow to a 5% EBITDA margin, and that's the path to which — if you see a sell-off in DoorDash, it could be a very interesting buy. After 13 years, more than 100 billion of GOV and market leadership in the US, DoorDash earns less than 50 cents per order pre-SBC.

Of course, the slope matters far more than the level here. That number was -38 cents per order as recently as 2022. And adjusted for the 1 billion of stock-based comp, the underlying cash economics per order are meaningfully better and inflecting, although I don't know if I would take out the stock-based comp. Per-order profit increased by roughly 50 cents in the last three years. If DoorDash can double the aggregate order count and increase per-order profit by another 50 cents in the next five years, that would be 7 billion in operating profit by 2030.

## Dan Niles on the AI Speedbump - 2:07:35

Now Dan Niles is a tech expert that did successfully call a short-term bottom in the AI names last week. Given a couple of his friends misunderstood the short tweet in the midweek, he basically said that Hynix was an acknowledged data point for those calling the top of the AI cycle. I think what he believes is this is an AI hiccup. And he thinks it's important as an investor to not just focus on what is supportive of your investing case, but even more to focus on what is directly against it, especially if it is a universally loved idea like AI.

So, having said that, he basically says that he believed that it was a speed bump, and he's been writing about the speed bump since June 20th. But a speed bump can be ugly. Starting in both 1995 and 1997, there were drawdowns of over 50% in the semiconductor index, but the semiconductor index still finished up roughly 850% by the end of the 1994 peak of the internet buildout, in March 2000, despite this. Also, back then those drawdowns had fundamental drivers. In 1995, Windows 95 did not lead to an upgrade cycle expected, and companies were sitting on a bunch of DRAM inventory. Intel wrote off over a billion of DRAM inventory as an example. There's no excess memory inventories on balance sheets today, but it could happen again. In 1997, the Asian currency crisis forced a slowdown in demand. The Iran war and continuation of the Ukraine conflict are both disruptive, but there's no general slowdown in demand being caused by them that you can see. From a technical standpoint, I believe liquidations and margin calls in both retail accounts and hedge funds, which typically run with leverage, over the past couple weeks is leading to a technical bottom.

From June 22nd, the peak of the momentum trade, the S&P is only down 2.1%, the Nasdaq is down 6.6%. Both the SOX index, the tip of the spear in the AI trade, is down 28.6%. Morgan Stanley's momentum index was down 38%, and their TMT momentum index was down 53.5%.

So from a fundamental standpoint, the advent of agentic AI started arguably at the end of January 30th with the formalization of OpenClaw. This requires 10 to 100 times more tokens than chat-based AI. No question that token minimization is going to be at the top of the list — so the token minimization that's happening at places like Microsoft using Chinese models was the main reason for the speed bump, but there are also other concerns that will arise along the way. On the following page you can see personal spending and GDP misses, which were positive for the market because less chance of a hike. Good news is bad news as we've seen in the past. Samsung's earnings couldn't have been more bullish. It was very, very positive on inventory levels for DRAM and NAND being low. They're witnessing continued robust demand on the HBM memory side. They have unmet demand from this year likely to carry over into next year. So I believe that was quite straightforward.

## Citadel's Rescue of Leopold's Situational Awareness Fund - 2:11:21

Now, what caused this big AI sell-off? So Citadel bought out Leopold Aschenbrenner's AI fund portfolio after massive July losses. So Situational Awareness, which was only launched in 2024, soared with leveraged bets and got as big as 45 billion in AUM using investors' money. It hit over 400% returns through June of 2026. Mid-July's rout in tech and chip stocks erased those gains, dropping the fund 67% for the month, amid four times leverage, which many people didn't know about, and forcing the bulk sale of its public equities to Citadel on July 30th at discounted prices. Aschenbrenner wrote to investors that the firm let them down but stressed the AI opportunity ahead. The fund retained a $10 billion private book, including private stakes in Anthropic, and stays up 80% for the year.

The 24-year-old ex-OpenAI researcher ran a fund to 45 billion with eight people, was up 439% net through June. By the end of July, he sold this entire public portfolio to Citadel. Ken Griffin described the exact failure of this two years ago, on public video, when he asked why portfolio managers wash out. Basically said that it happens when your portfolio is extraordinarily concentrated, you have large positions, you cannot demonstrate a clear, concise competitive advantage on why you own those positions. And there are some people that, with full information, are unable to help themselves and get better portfolio construction because they have loss aversion, they have personal biases, they're not good at trading, etcetera. He was also asked the opposite question in the same interview: why Citadel keeps working? His third and final answer was, "It's experience. It is the price paid in losses and the pain that converts that into wisdom." His leadership team, they've been through a lot of difficult moments together, through periods of turmoil and crisis like the GFC where they blew up their bond book.

Citadel started November 1990. Situational Awareness started in 2024. Long AI infrastructure and short software at roughly 4x leverage. Both legs went against it in the same three weeks. Leopold wasn't really wrong about AI, he was just wrong about the concentration and construction of his book. He had no concept of risk management and frankly, he had no situational awareness at all. Just one second.

So after news of the Situational Awareness liquidation hit the market, we had a sharp short covering rally in tech stocks. Now Dan Niles's comments on the Situational Awareness liquidation as well: "Forced liquidations due to excess leverage typically mark near-term historical bottoms." Now again, there could be another sell-off if this Iran war continues or the 10-year breaks out, but in the short term, near-term historical bottom. The fact that you had basically 30 billion in stock from a peak either fall or effectively change hands to Citadel, along with the fact that there were three other funds that had pressure, not just Situational Awareness, and that Citadel effectively fixed the deleveraging issue and the margin call issue, caused everything to rally.

The 10.7% rally in the Morgan Stanley momentum index after the FOMC day beat all but the 11.1% gain in April of 2001 during the dot-com bust. All the other moves in the top 10 occurred either during COVID, which was three times, the GFC, two times, or the dot-com bust, three times. For the most concentrated Morgan Stanley TMT momentum index, the gain of 19.1% crushes the prior 11.9% gain seen in December of 2000 during the dot-com bust. Five of the other top 10 gains were seen since November of 2025 during the recent meteoric rally. There were three more during the dot-com bust. While some giveback is certainly likely in the days ahead, given the ferocity of this one-day move, he believes that you have seen the near-term bottom due to forced liquidations. He's hopeful that the speed bump that he started to talk about since June 20th is behind us. Getting oil prices back to $70 with a decline in bond yields would certainly increase the odds even more, which we all now know.

He believes that we are still in early adoption of agentic AI, and the 10 to 100x increase in tokens needed since January relative to chat-based AI means that we're going to need a lot more compute.

So you can see, the peak size of the Situational Awareness book was 20 to 45 billion and now it's down to 13.16 billion before the sale to Citadel. The core failure was that the book was too concentrated, four times levered, and its software shorts were squeezed. There's one month of pressure, six days from performance letter to full exit. It wasn't a fraud. Sowood Capital Management was 3 billion, credit relative value. There was margin pressure on subprime stress. It took less than one month to do a deal and they substantially sold everything to Citadel as well at that time. Amaranth Advisors was 9 billion of AUM, huge concentration in natural gas futures. It took one to two weeks for the entire book to blow up, a several billion dollar book. The energy book was transferred to Citadel and JPM at a steep discount. You see a trend here where Citadel takes over broken books. LTCM blew up in 1998, relative value, fixed income arb. It took weeks to months to blow up and the Fed orchestrated a consortium rescue. Archegos was absolutely wild. It was a $50 to $100 billion family office, which concentrated equity into multi-prime TRS with a margin cascade. They basically just over-levered, and their business model was flawed. So at one point, Leopold contends he was actually up 490% and is still up 80% year to date.

The letter also reported blaming short sellers who were targeting the firm. We'll skip that part. And so the timeline on page 57 is that on July 1, Leopold's hedge fund Situational Awareness reaches 45 billion, up over 450% year to date. On July 10th, the SK Hynix IPO marks the top of the AI sector. Then from July 10th to July 20th there was a major sell-off across all AI stocks, mostly down 30% plus in weeks.

From July 10th to 20th, the short positions start going up causing more losses. He was short Adobe and other software stocks. He sends investors a letter on July 24th calling the sell-off a buying opportunity and asks for more money. This was a red flag. On July 27th, Citadel Securities actually says the Fed is going to do a surprise rate hike, which pushes the tech market even lower. And then the market sells off after Citadel's report. Banks start to margin call him and then Citadel comes in and buys his entire portfolio for a massive discount.

Leverage is the main reason people go broke in the stock market and this is a lesson for you to be careful. And leverage is also in the form of options.

## Cloud Giants Growth Update - 2:19:42

Cloud giants growth update: if you look at AWS, Azure, and Google, you'll see the 169 billion run rate, growing 37% year-over-year. Azure is also 124 billion run rate estimate, growing at 43%, and Google Cloud is 99 billion run rate, growing 82%. So you can see those in the three graphs below. This is from Clouded Judgement, Jamin Ball.

## Apple Earnings and Supply Chain Issues - 2:20:12

Apple is down 8% on supply chain issues and memory shortages, which is a big deal. It was trying to source even Chinese memory and move away from Micron. It was then warned by the US government. Even though revenue will rise 9 to 11% in the fourth quarter, that's lower than estimates of 12% growth. The shares were down 7% pre-market on Friday despite sales topping estimates and climbing 16% to 109.4 billion.

Apple has been struggling to secure enough computer processors and encountered fast rising memory costs and it basically underestimated what demand would be. Cook has called the memory issue a 100-year flood. He doesn't expect it to happen again, but he did look petrified this week. Apple also said that services growth would decelerate in the September quarter and it warned of an impact from regulatory changes to its App Store business model in the European Union because of fines. In addition, the tech giant's results showed weaker than expected revenue from China and services through the June quarter. China sales were around 18.8 billion, short of the 19.6 billion it expected. Still, total sales topped estimates, climbing 16% to 109.4 billion. Earnings rose to 2.02 a share versus average predictions of 1.89. So overall earnings were quite decent. You can see the stock market reaction on Apple on page 60 and then on page 61.

## Amazon Earnings Surge - 2:22:04

We talk about Amazon earnings — they surged after the company reported cloud computing revenue accelerated for the fifth straight quarter, easing investor concerns that it won't produce a return on huge expenditures to meet booming demand for artificial intelligence. Revenue was up 37% to 42.4 billion at AWS, which generates about a fifth of the company's revenue and most of its operating profit, which is great. Amazon raised its projection for 2026 capex to 220 billion from the previous estimate of 200 billion, but its operating earnings growth resulted in it rallying, whereas Meta hasn't monetized yet.

Like big tech peers, Amazon is spending heavily on data centers and chips in an effort to take advantage of rising demand for AI and cloud computing services. The company reported spending more than 53 billion on PP&E, including proceeds from store sales.

Amazon raised its projection for 2026 capex to 220 from a previous estimate of 200, but because its underlying business is so strong, it didn't sell off by any means. Those outlays for Amazon sent free cash flow negative, with Amazon reporting an outflow of 7.6 billion for the trailing 12 months at quarter end, but because the operating profit growth is so strong, people ignored it. Jassy said that Amazon's investments would moderate over time because it's currently building data centers from the ground up. These sites will last 30 years while equipment inside will have to be updated every five to seven years. So Amazon shares rose almost 15% as the market opened in New York, for the biggest intraday gain since November of 2022. So Microsoft and Amazon did well, Apple and Meta sold off because of the memory squeeze and overspending on AI without an ROI.

## Reddit Earnings Review - 2:24:05

You can see Reddit's earnings. We bought shares in the 130s. Shares were down 23% after earnings even though they had a very strong quarter for revenue. Daily active users were a little bit mixed and that's why it sold off. These are very, very strong P&L results. The company's growing top line at an extremely fast rate. Third quarter forecast was 860 to 870 versus an estimate of only 829. EBITDA of 385 to 395, so a mid of 390 versus an estimate of 367. Advertising up 64%, other revenue up 23%. Net income 252.8 million versus only 89 million last year and estimates of 196 million. R&D expenditures were lower than expected, 231 million versus 237 million. Daily active users were up 18% to 130.3 million versus an estimate of 130.07, so slightly below that. International daily active users were 77 million versus 75.97 expected. It was really the US daily active users which were 53.2 million versus estimates of 54, so slightly lower than estimates. But US average revenue per user was up a lot, 51% year over year at 11.85 versus an estimate of 10.50. So their advertising rates are going up and their total daily active uniques was up 18%. So overall, I think the business is still doing fine.

## Japanese 30-Year Bond Breakout - 2:25:40

On the following page, you can see the Japanese 30-year breaking out above 4%. Now, I did want to make a note that after Reddit's results, we did add more. We think it's quite cheap in the 130s. And the stock is probably worth around 200. I think that it is a medium term to above 200. Let's quickly go back and look at some of these questions. So we did answer EJW's question on the 10-year. Dr. Inflation Steve is asking, with wage growth inflation and corporate earnings through the roof, at what point does this break? I don't think wage growth is going to continue to accelerate with the labor market weakening. We have been replacing GDP growth with AI spend, but that's not sustainable. I'm not as worried about inflation. I think that you would need a reacceleration of inflation outside of energy for the market to really freak out. But with core PCE going up several months in a row and now up almost a percent, I think if we see inflation prints go up further, outside of oil, that would be a big problem.

## Q&A — Oil and BDC Attractiveness - 2:27:10

David is saying, do you find oil or BDCs attractive at this price? No, not yet. Not yet because private credit default rates keep going up.

## Q&A — Analysis of Robinhood (HOOD) - 2:27:26

So let's look at Hood for Big Cat real quick. 2 billion of operating income, stock is around 85.56. It's sold off from almost 120. Bounced from the lows in the high 60s. I don't think — and because of the Clarity Act, I'm more skeptical because of all the bad news around Trump. Metrics for Hood have been higher across the board, but it's not super cheap. I think it's more of like a hold. I think the stock's probably worth somewhere around $100 a share, so just modest upside from here. I think the stock would need to fall kind of 60, 70 to be more interesting to me, just given the valuation is already a premium valuation. So I would just hold off on Hood until crypto also inflects higher. Not too excited by it.

## Q&A — Shorting Treasuries and Munis - 2:29:03

I call is short Treasury long Munis via TBF. Again, I don't like leveraged products, so I think if the war goes away, I do think the 10-year is close to peaking. But if the war doesn't end, I do think that I started to buy TLT, a tracking position, and would add around five, and that's where I think you probably start taking off the short now a little bit. And if the war really ends, you take it off.

## Q&A — Thoughts on Qualcomm (QCOM) - 2:29:38

In terms of QCOM, I think QCOM is slightly cheap. On a fundamental basis, it looks oversold from a technical perspective, but it's actually not very, very cheap relative to historical valuations. If you look at the earnings growth, it's really not as high, but it does benefit from the AI boom.

## Q&A — Fed Hike Impact on Preferred Stocks - 2:30:02

If the Fed hikes rates a quarter percentage point in September, it does provide a buying opportunity in preferreds, although I don't think preferreds would even sell off that much unless there was a risk that you could see more than one hike in a row. I'm waiting for Marcus 13 to type a question as well.

## Q&A — Citadel's Strategy with the Acquired AI Book - 2:30:23

So what does Citadel do with the massive AI book they just acquired? Well, remember that it was a long and a short book. So a lot of hedge funds had degrossed and they had already, for the last month, there had been consistent pressure on these AI-focused funds and a lot of the hedge funds were already degrossing, and I'm sure Citadel had already been building a short book. And basically, by buying this big book, they probably offset some of their shorts. And I think that now that they've bought the book, they could easily hedge the book, or just absorb it in their various TMT pods, as they're probably underweight tech before they bought this book. So I'm actually not sure. I really don't know their positioning. I'm assuming that because most hedge funds had been cutting back semiconductor exposure very rapidly, this allowed them to basically resize their semiconductor book ahead of a rally into 3Q earnings, because demand for memory and GPUs is not going away. But if, for example, they're caught the wrong way and the 10-year does rise rapidly, it could result in Citadel then hedging or dumping into the market as well. But at least in the near term, we don't see that happening. Even if they do it, they'll do it over time, to lock in some of the quick gains that they've already made. But that's a good question.

And so, we were very, very careful this week because of the Fed, not to add too much risk. We did have, if you go to the alerts tab, we did have a couple wins. And then on Reddit, where earnings — where the user numbers missed — we did add. We bought 10, 15 bips of Reddit in total. We were out of SanDisk after a massive rally. We sold almost perfectly after it was up 30% on the week before the Friday AI sell-off. We trimmed half of our Nebius, which was up 27% early in the day, and we trimmed Bloom Energy at 207. So that's all on the alerts. We did add 20 bips to the Korean market July 29th before the rally. So that was also a good trade. And we had added 10 bips to SanDisk at 10.60 on July 28th, which was also really good timing. We added 10 bips to Redwood. We'll probably add more to Redwood. We added to Dynex 10 bips. Redwood gave back a little bit of its gains because of the 10-year. But I think it's a name to add to if you think rates have peaked. We trimmed some of our ServiceNow, which rallied a lot. So we've traded the market quite reasonably. During the week we'll have additional names to add to the portfolio, additional sells. I hope this call has been helpful. We'll have during the week some additional ideas for you in the alerts. We'll resend the Reddit model as well. And I hope you have a good trading week. This call recording will be up very shortly. Thanks, EJW.