# Special Situations Report — Weekly Call (readable notes)

- **Title:** PCE below expectations, Micron (MU) earnings + 16 take-or-pay strategic customer agreements, the Mag7 / hyperscaler capex-burn selloff (~$5.3T), open-source Chinese AI models, Iran ceasefire breakdown + re-talks, Alamos Gold (AGI) follow-up (sell-side calls), Sea Limited (SE) cheap, Rithm Series F preferreds (RITM-F) at ~9%, BTBT / White Fiber (WYFI) at a 43% NAV discount, merger-arb (VRNS, EBAY, WBD, PRA, S&P spin), MSTR/ASST crypto preferreds, BDC redemption gates (APO, ARES)
- **Show:** Weekly SSR research call (premium subscriber recording — no public video)
- **Guest/host:** Jay Singh (founder, Special Situations Report; ex-Goldman Sachs)
- **Date:** 2026-JUN-28 (Sunday call; PDFs title it "6-28-26")
- **Source PDFs (this folder):** `transcript.pdf` (premium transcript), `report.pdf` (written SSR), `topics.pdf` (page-by-page agenda), plus supplemental research Singh references: `GS TMT TODAY 6.26.26.pdf` (Goldman tech/media/telecom), `Standard Chartered June 26th Strategy.pdf`, `markets _ macro(6-28) (tony).pdf`
- **Note:** readable notes auto-extracted from the premium PDFs; fillers (um/uh/false starts) removed, wording otherwise verbatim from the recording. Section headers carry the recording's `MM:SS` cue. No public video, so the per-name table has no deep-links. Auto-transcript garbles corrected to the right entity (Almos→Alamos, Rhythm→Rithm/RITM, "C Limited"→Sea Limited (SE), Migu→Migoo, Whitefiber→White Fiber/WYFI, Wheelock Spring→Wealspring, Banyan Tree→Banxia, Citron→Ed Zitron, Money→Monee, B.D. Vance→J.D. Vance).

---

2026-06-28 - SSN - PCE Lower, MU Earnings, MAG7 Lags, AGI, BTBT 43% Cheap, RITM-F 9% Prefs, SE Cheap

## Agenda - 00:00

We have quite a detailed call. Happy Sunday everyone. We are going to be covering PCE, Micron earnings, semis causing price hikes at Apple and Microsoft. We're going to be talking about the continued Iran peace discussion, which is an ongoing project. We will also be discussing a follow-up on Alamos Gold — I had some calls with some sell-side analysts. We will be doing a follow-up on Sea Limited, which is looking very cheap for a long-term allocation after the recent sell-off. That's ticker SE.

We'll also be discussing Alamos Gold again. Rithm Series F prefs, which are yielding about 9%, available to all retail investors. They sold off over the last couple months with the rate volatility. And we'll be discussing VRNS, which is effectively selling itself.

## Economic Calendar and Earnings - 01:17

On the first page you can see the economic calendar. The most important events next week are consumer confidence, ADP employment change, S&P manufacturing PMI, ISM manufacturing, construction spending change, non-farm payrolls and durable goods orders. But none of these events I think is going to make a big impact on the market, so I'm really not going to waste time going through the calendar. Same for earnings. The most important earnings next week are going to be AeroVironment on the defense side, Nike on Tuesday, Constellation Brands on Tuesday, General Mills on Wednesday, FactSet — which we don't like versus S&P and Moody's; we like S&P and Moody's more — that's on Wednesday. So that's really it. There aren't very many companies reporting next week.

## Iran Peace Talks Update - 02:06

We are seeing futures. We've seen a modest recovery in the S&P 500 futures from the open as, right as futures open, Iran and the US after bombings over the weekend say that they are open to a peace negotiation.

Going forward, if you read the Deutsche Bank piece, this is going to be a wobbly peace negotiation to say the least, and there will be continued volatility as it goes back and forth. But what's almost comical is that we saw Friday after 4:00 PM there was news discussed about the bombing of a Singaporean cargo ship. The US retaliated because it wants to protect the Strait of Hormuz. Then we had missiles directed at US air bases. And now as futures are opening, all of a sudden there's some sort of renewed discussion. It's just quite comical how this war is almost perfectly timed with the market closing and opening. You really couldn't make this up.

## Magnificent Seven Sell-Off and Market Dynamics - 03:41

More importantly, we need to discuss what's happening to the Mag 7. We've seen a serious sell-off in the Mag 7 stocks since the end of October of last year. The Magnificent Seven tech stocks in the United States have sold off to the tune of about 5.3 trillion in market cap from their local highs in October of last year. So Microsoft is down 35%, Meta is down 33%, Alphabet is down 20%, Tesla is down 20%, Nvidia is down 18%, Amazon is down 18%, and Apple is down 13%.

Microsoft shares are poised for their worst monthly loss since 2008, having plunged nearly 19% so far in June. Oracle is actually the worst off out of the group. This sell-off was initially driven by Mag 7 free cash flow being invested aggressively into data centers without a known ROI, instead of dividends and buybacks which shareholders like. Now memory chip inflation has negatively affected Microsoft's Xbox division, which they're actually looking to spin off, and is also now being reflected in Apple's forward margins and price increases.

This huge rally in AI infrastructure is made possible by a massive increase in capex from hyperscalers — the large cloud computing companies that are building physical infrastructure: Google, Meta, Amazon, Microsoft, and Oracle. It is these hyperscalers' willingness to fund huge AI infrastructure capex by spending nearly or completely all of their operating cash flow, tapping debt markets, and raising equity that has resulted in the semiconductor rally. But the market is now questioning whether these companies are investing their capital adequately. While GAAP earnings margins continue to expand, if a company like Meta has to leverage itself and add debt to its balance sheet, or a company like Oracle has to leverage itself, in order to show the market that it is competing adequately in this new AI arms race, then the market won't give them full credit for the forward growth expectations.

The reason why Micron is able to print 85% gross margins and 830% earnings growth in 2026, which we'll discuss in a moment, is that hyperscalers are willing to reach deep into their pockets to fund this massive amount of capex. It's almost like an arms race where there will be big winners and big losers. On the following page you can see the iShares Semiconductor ETF, the SOXX, total return of 107% since just January of this year, which is unbelievable. And the Roundhill Mag 7 ETF, the MAGS, down about 8%.

The sell-off has been driven by free cash flow reallocation, and you can see a very clear graph of Amazon, Oracle, Microsoft, Google, and Meta free cash flows, which have gone from almost a third of a trillion down to zero from the end of 2024 to 2026. On page four you can see the race to spend, actual and estimated capex from the five hyperscalers — forecasted by 2028 to see almost a trillion of spend on capex, which has been driving the semiconductor stocks to all-time highs. At the bottom of page four, the inverse of that graph: the biggest beneficiaries of this capex spend being SK Hynix, Samsung, Micron, Nvidia, Broadcom, and AMD.

On page five, Microsoft stock was down another 3.3% on Thursday, which takes it below its March closing low. Microsoft is down about 28% year over year and its overall drawdown from the high is about 35%, which is probably why Burry partially covered his short on Palantir — down 50% from its local highs — and started to buy leaps in Microsoft. Even Palantir is down 50% from its local highs, which shows you that even the most hyped stocks suffer, and valuations do matter. We talked about Palantir being overvalued in October of last year — it was trading at something like 300 times forward, not even earnings; it was trading at like 300 times forward EBITDA. While it's a great company, that valuation was not justified because there's no way Palantir can grow its revenues at 50% a year every year and its earnings at 100% every year. Now it's basically a quarter-trillion-dollar market cap. It was a half-trillion-dollar market cap with only 5 billion of sales and roughly 2 billion of EBITDA, less than two at the time.

## Special Situations and M&A Update - 09:44

With one week remaining in the first half of 2026 — in fact the second half starts this week — US public M&A volumes are annualizing up about 19% versus 2025 levels. This slowed down during the worst of the Iran war. On Friday morning the FT was reporting that the EU was set to clear Paramount's acquisition of Warner Brothers, provided the company accepts certain remedies, including Sky possibly exiting from its Universal Pictures International JV. In other news, ProAssurance, ticker PRA, announced its sale to Doctor's Company obtained its final approval and is set to close Friday. The approval comes 16 months following the M&A announcement, so it was a long and grueling wait for arbs. It's still a 7% IRR if you set it day one.

S&P is also spinning off its global mobility division, which commenced its when-issued trading on Friday and will trade regular way this week. Shares of the cybersecurity firm Varonis, VRNS, were trading up 10% since the company announced it's weighing a potential sale, which we talked about at the beginning of this call, after receiving interest from private equity. And GME removed its CEO performance award to focus on the acquisition of eBay. So they are looking to continue to pursue their acquisition of eBay. I don't know how they're going to be able to do it because there's no way they would be able to raise enough debt to provide the cash component for the deal. Maybe they offer some compelling equity trade. After that announcement, eBay rallied all the way up to $120 a share. It was trading as low as $80 a share in February and in November of last year, so the market did price in some possibility of a deal. We shorted the stock somewhere probably in the high one-teens, so close to 120, and then it quickly traded off all the way down to as low as 104 on June 22nd and now it's bounced. We covered somewhere in between there. It's bounced back to 108. At some point — I don't know how GME is going to do it, but they probably have some trick up their sleeve to resubmit a bid. You can read the details on these four deals on pages six and seven.

## Rhythm Capital Series F Preferreds - 14:03

On page seven we also discussed these Rithm prefs. As you guys know, Rithm is a diversified asset manager. They bought Oxif a couple years ago. They also have a big mortgage servicing business, which is inversely related to interest rates, and a REIT business. So it's a quite diversified business. It has a 5.3 billion market cap. It is a profitable company, and it's frankly a more diversified and safer company to invest in than RWT — less sensitive to rates overall because it has a natural hedge. The common stock yields about 10.6%.

That's interesting. If the common stock only yields 10.6% and you can buy the preferreds — these are not baby bonds, these are preferreds — at 9%, I would think the preferreds offer a much better risk reward because you have 5 billion of market cap below you. So this is a simple, very simple trade. I call it a common-sense trade. Rithm did about 1.1 billion of operating income, and its operating income was up about 120 million year-over-year. It does about 180 of EPS. So it trades 9.38 on 178 — trades at like five times earnings. And next year its EPS is supposed to grow to 2.26. Doesn't have a lot of exposure to private credit. So the prefs look pretty decent, in my opinion, probably a risk score of two out of five with one being the highest, with treasuries. So I personally think it's one that I'm going to add this coming week. Doesn't require a lengthy explanation.

If you want to look at the ticker and the snapshot of these prefs and the size, they have about 10 million shares outstanding. They were priced back in January of this year. They trade a little bit below par, which is how you get the yield up from 8.75 to kind of a 9.2% yield because they trade below par. And they pay quarterly.

## Iran-US Tensions Timeline - 16:36

On the following page, let's go through a brief timeline. It's so sad that we have to do this every week. The Deutsche Bank piece does a very good job going into the actual details, but I'll give a high-level update. On Monday, Iranians discussed progress on the peace treaty. Oil/USO was down, equities were flat from down 0.5 to 1% the prior night because there was an escalation that night as well. Qatar basically said that parties reached a roadmap to reaching a final deal in 60 days and the market had a bounce. J.D. Vance, before departing from Switzerland, said yesterday was a good day of talks, the Strait of Hormuz is now open, they were setting up a mechanism to keep the Strait of Hormuz open, and Iran agreed to invite inspectors to look at its nuclear stockpiles.

Then the Treasury issued a 60-day temporary general license waiving key Iranian oil sanctions until August 21st, 2026. On a quick read it's a rather wide-ranging license covering exports, shipping, and ancillary services including crude, refined products, and petrochemicals. Then on Thursday it was discussed that despite progress in the talks, Iran still planned to charge an estimated 40 billion in annual tolls for reopening the strait. The proposal was met by resistance from the US and other countries; specifically Marco Rubio strongly rejected this.

Then in the evening on Thursday there's an Iranian attack on a Singaporean-flagged cargo ship. It was effectively the IRGC that attacked this vessel, which was not acceptable given the peace treaty. Hormuz crossings had rebounded upwards of 40 vessels a day as of June 24th before this happened. So after the unnecessary Iranian attack on the Singaporean cargo vessel, Trump announced US military strikes on Iran and threatened that Iran will no longer exist — specifically the IRGC is probably what he meant — if there comes a point when we are no longer able to be reasonable.

The US military conducted strikes on Iranian targets including some radar and satellite sites. They struck a missile and drone storage facility as well and coastal radar positions on Friday. Iran said the cargo ship was attacked because it was using an unauthorized route to transit through the Gulf waterway, whatever, that's just an excuse. After the US strikes, Iran in turn accused the US of violating their interim deal and said it had struck targets linked to American forces including two American bases. That has not been confirmed by the US. US Central Command (CENTCOM) described the American strikes as a powerful response to the drone attack a day earlier. CENTCOM said the US military would continue to provide safe passage coordination and support to commercial vessels. The Iranian Islamic Revolutionary Guard Corps said its navy had struck US military positions in the region without providing any details. Bahrain's foreign ministry said the country had come under attack from several Iranian drones earlier on Saturday, condemning the action as a flagrant violation of its sovereignty. Also on Saturday, the United Kingdom Maritime Trade Operations (UKMTO) said a tanker was struck by an unidentified projectile in the Strait of Hormuz; the vessel sustained damage to its bridge.

The Pentagon reportedly delayed publicly announcing US strikes until after the stock market closed on Friday, with the timing intended to reduce immediate impact on financial markets, reported by NBC. You can see Hormuz vessel crossings by the end of the week actually fell from a peak above 50 all the way down to 24. You can see the Iran Foreign Ministry condemning the attacks and the unofficial translation on page 11 and what they wrote on page 12. Effectively, after all this, it looks like as the futures opened both the US and Iran agreed to talks — that was according to Axios. So that's where we are now. S&P futures are up 0.45%, Nasdaq is up 0.85%, small-cap Russell is up 0.43%.

## PCE Data and Fed Commentary - 22:07

Let's quickly cover the PCE. You can use the CPI and PPI to estimate the PCE. Analysts were quite concerned given the PPI on how the PCE would come out, and frankly it was not as bad as expected. The market did rally very briefly to the PCE data that came out midweek. PCE was 4.1% year-over-year versus estimates of 4.1%. Month-over-month it was 0.4 versus estimates of 0.5. Core was 3.4, which matched expectations, and on a month-over-month basis it also met expectations. So it wasn't that bad. The market rallied on that, then when people realized PCE is still 4.1% and the Fed needs it at 2%, the market gave up the rally. Then the New York Fed — this is John Williams — basically said that in coming quarters they do expect inflation readings to edge down for several reasons.

Inflation is up because of tariffs, energy, and AI-related demand for tech goods, but he thinks it will fall because the tariff effects have mostly played out; a base case is that Hormuz-related supply disruptions are behind us; third, housing inflation continues to slow; and fourth, there's no evidence of labor-market-driven price pressure. So the punchline, according to him: given the elevated level of inflation, it's imperative that we restore our 2% longer-run goal on a sustained basis, and the current stance of monetary policy is well positioned to do that. He thinks inflation will be contained.

In his speech, which I included a link to, he talked about absorbing shocks: despite unpredictable global risks, specifically ongoing economic disruptions from the Middle East conflict, the US economy has remained relatively resilient, which is why unemployment is still relatively low despite white-collar unemployment going up. Then consumer and business health: while households face elevated fuel and energy input costs, consumer spending and business investments remain robust — that's frankly because 10% of Americans do 50% of the spending, and we have a huge AI investment boom. Anchored expectations: inflation expectations remain well-anchored in the midterm; one-year inflation expectations ticked up modestly through May, but three-to-five-year outlooks remain unchanged. The FOMC maintained its target federal funds rate at 3.5 to 3.75 to continue steering inflation towards the 2% goal, and he said there are two-sided risks: the AI boom could push prices up faster than expected, which you've already seen with Apple and Xbox, while Middle Eastern supply chain disruptions still threaten global growth, especially in emerging markets.

## Semiconductors, Hyperscalers, and Market Volatility - 25:44

One interesting thing: semiconductors are biting the hand that feeds, resulting in significant market volatility. It's no secret that the largest contributor to S&P 500 returns in 2026 has been the AI infrastructure trade — all of the AI capex recipients, with key names being Micron, AMD, Intel, SanDisk, and AMAT, accounting for 55% of the S&P 500's return year to date and posting an average gain of 356% year to date.

This huge rally in AI infrastructure has been made possible by a massive increase from the hyperscalers — Google, Meta, Amazon, Microsoft, and Oracle — funding huge AI infrastructure capex by spending nearly or completely all of their operating cash flow, or more than that according to Meta, by tapping debt and equity markets. The reason Micron is able to print its 85% gross margins and 830% earnings growth in 2026 is that hyperscalers are willing to fund it. But instead of rewarding this aggressive spend, the market has been punishing them, essentially biting the hand that feeds the AI infrastructure trade because the ROI isn't really provable yet. This was the biggest risk to the market rally after the inflation and the war, in our opinion, as I've seen over the past 20 years that investors always punish capex spend with unknown ROI versus cash generation and capital return.

The basket of hyperscaler names — Google, Meta, Amazon, Microsoft, and Oracle — is down about 9.5% year to date, trailing the S&P 500 by nearly 15%. These five names represent a 16% weight in the S&P 500, so their weakness is weighing on index returns in an outsized way, which is one reason the equal-weight S&P 500 is up 10.5% year to date versus the S&P 500 only up 7.5% — actually not normal; the last three years it was the opposite. You can see the hyperscaler index absolute and relative performance versus the S&P 500 on page 14.

We must assess how the share-price weakness could impact the hyperscalers' future plans to spend aggressively on AI capex, and thus how hyperscaler equity weakness could challenge the dominant AI infrastructure trade — semiconductors, tech hardware, electrical equipment. We could see hyperscalers continue to spend regardless of near-term stock price, citing an existential arms race and pointing to a future inflection in the earnings payback. Or the market's mood could sour toward this aggressive spending so much that it puts pressure on management teams to find capex discipline, which has negative implications for the AI infrastructure trade. If that were the conclusion, all these names would be down 20-30% in a day. So we find ourselves asking: just how deep do you believe in the ability of hyperscalers to continue to aggressively grow AI capex spending? This may be the most important question the market will have to wrestle with through the rest of this year and next year.

Mag 7 capex guidance has only been accelerating. Just uttering that hyperscalers could pull back or moderate spending during this apparent existential arms race seems sacrilege. But we don't have to see them actually cut capex to see the pace of returns in the AI infrastructure trade slow — we could just see 2027 capex slow versus expectations. The key reason 2026 has been such a powerful year is that analysts significantly underestimated how much hyperscalers would be willing to spend on their own and other people's equity and debt money on data centers, and thus underestimated how much demand would flow to these newly supply-constrained AI infrastructure plays. As you can see in the Goldman Sachs chart on page 15, analysts expected only 36% growth for hyperscaler capex at the start of 2026 — this was expected to be a mere 10% this time last year — but hyperscaler capex is now expected to be nearly 90%. As we turn to 2027, Goldman estimates consensus expects another 22% growth, so capex will be roughly 920 billion by next year. We've seen estimates from other banks as high as 1.2 to 1.5 trillion for 2027 capex, which boggles the mind, meaning all these companies are going to have to borrow debt and issue equity like Google did of over 80 billion to fund their capex spend.

## The Hyperscaler Spending Debate - 31:34

And behind it all there's a price to be paid. This raises a question about how hyperscalers will fund another large increase in capex in 2027. Equity and debt markets have recently been receptive to incremental capital raises, but from the strongest players only. Google was able to issue a 100-year bond with robust demand back in February and raised another record 85 billion in equity in early June with the express purpose of significantly increasing 2027 capex. This is because Google's operating cash flow and balance sheet — it still has negative net debt thanks to a large cash balance — remain robust. But other potential raises have not been as well received. When there was a rumor about Meta raising equity and debt, Meta shares suffered pretty sharply. Oracle's credit default swaps are back on the rise, even though they don't trade much. New cloud players like CoreWeave sank after the company looked to raise 3.5 billion in a dilutive equity round. If Meta were to do an $80 billion equity round, the market would really hammer Meta because it doesn't have a good free cash flow outlook at the moment. And SPCX — the new hyperscaler, SpaceX — which did a $25 billion debt-market round, has seen traders try to short its 2056 bonds. We'll discuss how traders were treating the new bond issue later.

This is a good reminder that the market is now questioning not only the hyperscalers' belief that they need to spend more on AI capex, but also the belief that this capex spend is an attractive use of significant capital. The chart on the next page shows EPS in blue, the PE ratio in black, and free cash flow in gold. It shows that the weakness in Mag 7 shares had nothing to do with earnings being weak — it has all been driven by valuation multiple compression and free cash flow falling. We're using the Bloomberg Mag 7 index because it overlaps with the hyperscaler index but has more history. The Mag 7 is flattered by the exclusion of Oracle, which has the weakest free cash flow, and the inclusion of Apple and Nvidia, which have not spent aggressively on capex (and Nvidia is also a capex beneficiary). The valuation of the Mag 7 has fallen by 10 multiple turns from 32 times to 22 times at the liberation-day lows. For a bullish view, this de-rating has been overdone and the Mag 7 is poised for a rebound as valuations normalize — AKA Burry buying leaps in Microsoft. For a bearish view, the market is now discounting a lower-cash-generative, lower-ROIC, heavier-debt-laden set of companies than in the past. Both may be true to some extent — some hyperscaler stocks look oversold in the near term, but the transfiguration of these business models to be more capital intensive is suggestive of a lower long-term valuation.

How much share-price or valuation-multiple weakness will hyperscaler companies tolerate before they begin to rethink massive 2027 capex plans? And if they do, how much will semiconductor stocks sell off in 2027 and 2028? So what if this whole crusade is a charade? It doesn't mean that it is. We're not suggesting that AI is not exciting and revolutionary and still in the early days of broad adoption that could have significant impacts on the market and economy, but instead noting that there are real debates as to how sustainable the current breakneck pace of AI infrastructure spending is, given questions about the profitability of AI model providers and about market-share loss to the Chinese open-source AI models, which even Microsoft is using now. The flame of this debate grew hotter on Thursday when news broke that OpenAI may be delaying its IPO until 2027. This means it may not have the cash from its trillion-dollar valuation and 60-billion-dollar IPO to fund all the AI capex the market expected. This is important because OpenAI is Microsoft's largest cloud customer and represents 45% of Microsoft's cloud backlog, so it could negatively impact Microsoft earnings.

You can see the Mag 7 forward PE ratio versus the S&P 493 PE ratio in the Apollo graph on page 17. A bullish argument is in reports showing how direct AI revenue is growing rapidly — this is not profitability, just revenue — which Exponential, a research firm, estimates three times faster than prior technology adoption cycles, now starting to clear the depreciation hurdle of AI infrastructure investments already made. Only time will tell, but we think we'll see more volatility before the actual ROI is known, because ROI is not revenue, it's actually EBIT or cash flow. On the following page, AI is scaling three times faster than any other past IT wave like the internet in 1995, mobile apps starting 2007, or cloud in 2010.

## Chinese AI Models and Market Pressure - 37:56

The bearish argument is basically what Ed Zitron says on AI model developer unprofitability, flagging how losses could swell as revenues grow because of the high marginal cost of these business models that need ever more compute to meet demand. This raises a last important point about competition: some models may not be as capable as the US top-of-the-line frontier models, but because of their lower cost — like DeepSeek — and despite their foreign provenance, some simpler AI users are opting for these more cost-optimized models. Michael Cembalest's latest Eye on the Market from JP Morgan displays this. If you look at the cost to run on a US-dollar log scale, all the frontier models like Claude (Claude Fable 5) are almost excruciatingly expensive to use, whereas DeepSeek V4 Flash and DeepSeek V4 Max cost less than 1% of the price of running even Claude Opus 4.8. So I do think you'll see more and more American companies take some of these Chinese models, retool them so our government isn't pissed off — they're open source, you can download them, make changes, and use them for their own needs.

The debate is clearly not settled, but after such a powerful run in the AI infrastructure trade contrasted with weakness in the hyperscalers funding the boom, we should not be surprised to see ample volatility. On page 19 you can see how the AI boom compares with boom-bust cycles of major innovations, trading as multiples of the pre-boom trough — not a really telling graph, but it shows the magnitude and speed of this AI trade. The massive capital being spent on the data center buildout seems a huge gross misallocation relative to the broader economy's needs, according to Semafor, but I don't think Semafor has enough data to make a solid conclusion yet.

On the following page, what is a really big risk — and what you can track — is that more US companies are shifting to open-source Chinese AI models. UBS, the Swiss investment bank, says 60% of companies are now watching AI budgets and moving to cheaper models and open-source Chinese models specifically. The pressure is coming from extreme bills including users spending up to 35,000 a month per employee and teams exceeding quotas by 200% like Uber — they exceeded their entire AI budget in like one quarter — and companies cutting internal AI tools from five to two. Companies are not abandoning AI, they're using model routing, which sends easier tasks to cheaper models and saves premium models for hard reasoning, code, and long-context work. Chinese open-source models such as Qwen, DeepSeek, MiniMax, GLM, and Kimi now fit the enterprise cost curve because they can be run locally or used through cloud catalogs. A snippet of this UBS survey is on page 21.

On page 22, Bloomberg shows an interesting bar chart: US companies' use of Google, OpenAI, and Anthropic was about 70% in June 2025, with only about 10% use of others and about 15% use of Chinese models like DeepSeek, Tencent, Xiaomi, and MiniMax. Fast forward one year to June 2026: usage of US models is only 30%, non-Chinese non-US models something like 20%, and upwards of 45% usage of these Chinese models. I had no idea this was happening to this extent. If this is true, then with more market-share change, why should these companies keep spending as much on AI infrastructure if they're not able to monetize and continue to lose market share to these Asian models? There was an article — not the best written — but some concern in China, specifically two Chinese hedge fund managers warning about an AI bubble.

## Chinese Hedge Funds Warn of AI Bubble - 43:03

Two of China's best-known hedge fund managers — one is Wealspring Asset, whose founder Yang Dong is well known in China for calling the top of the Chinese and global market in 2007 — said global AI stocks have become a super bubble and the collapse may not be that far away. Who knows what he's motivated by, but what we do know is that Chinese companies are also starting to build memory; they're competing with US companies on the cheaper AI models. Maybe he thinks US companies need to cut costs dramatically or US customers will cut back their spend. What he specifically talked about was pressure on breakneck revenue growth at Anthropic, and some of the prices per token Anthropic is charging are just not sustainable for companies to continue buying. At least four other Chinese hedge funds expressed reluctance around AI in May, even around domestic companies, according to a monthly summary of fund views compiled by CSC Financial Group.

The investor letters offer insight into how China's prominent hedge funds are viewing a technology shaking up markets. AI-related stocks have boomed this year with bellwethers like SK Hynix and Micron Technology more than tripling in value, but the rally is suffering from violent pullbacks as investors and their counterparts become more cautious. Wealspring only manages about 1.4 billion of assets; they said they missed a lot of this wave. They didn't anticipate that a mere boom driven by a wave of massive demand could be hyped up so much and result in such high market valuations. What they say is that a lot of these AI companies like Micron don't really have a moat — they are oligopolies, but it isn't very difficult for another company over many years to develop a large memory infrastructure. They're saying that back in the 2015 bull run there was a term called brainless buying, and right now we're probably in a similar state. I think that applies to a lot of the super-speculative, triple-levered, margin-using South Korean and Taiwanese investors. But I do think the fundamentals and earnings of the companies, at least in the near term, have been stronger than anyone expected. Banxia, the other hedge fund talking about an AI bubble, only manages about 300 million. China's AI index is also up about 35% this year, not as high as the US at about 100%, but seven times higher than the rest of the Chinese benchmark, which is up less than 5%. You can see Chinese AI performance at the bottom of page 23; from local highs, the China tech index has sold off quite a bit.

## Alamos Gold (AGI) - 46:12

Now let's take a break from that conversation and talk about Alamos Gold on page 24. We did get a pretty detailed comp set and compared Alamos to some other Canadian gold miners. Remember that Canadian, North American — like US and Mexico — miners are tier-one assets because they don't suffer from the emerging-market risks, for example of having a gold mine in South Africa or Mali, where you'll have issues with power, union riots, which could result in lost production.

So I personally think the fact that Alamos is now trading at the same valuation — P/NAV or earnings or forward EBITDA — as some of the more emerging-market gold miners does make it quite interesting for a long-term allocation. This comp graph includes all the copper miners. We frankly think HBM (Hudbay) is quite interesting and would be buying here as well on the copper side. On the gold miners, on a free-cash-flow basis AGI is around 7%, but it's growing much faster than all the other miners. So while Kinross is actually quite cheap as well, we think AGI — because if they can build their mill to the capacity they need for the Island mine — should be able to increase its revenues quicker than a lot of the bigger gold miners like Agnico Eagle. We think Alamos could grow revenues around 30% a year, and for that reason we think it's quite an interesting trade — and not just a trade, more so an investment.

## Alamos Gold (AGI) Valuation Breakdown - 48:27

You can see our valuation from last week, where we think the stock in US dollars is worth about $48 a share. We did a mine-by-mine analysis. They have the Young Davidson mine, where it had the seismicity, or mini-earthquake, because of unknown reasons — that's probably worth around $7 a share, down from $10 a share, because it does affect about 2,500 tons a day of mining production. The Island mine is worth about $27 a share and that impact is really unaffected; that NAV per share will likely increase once they increase their mill capacity. They also have a mine called Lynn Lake in Manitoba, Canada, worth the same amount as Young Davidson, and then the Mulatos mine. So Young Davidson is only really worth 15% of the company, and of that 15%, maybe 3 to 5% was affected by this event. So the stock really is oversold. They also have some exploration assets in Turkey, and a decent amount of cash — about 1.5 million of cash at 3.6 per share — so the corporate adds about $3 to the corporate NAV, and that's how you get to $48 a share.

## Analyst Calls Regarding Young Davidson Mine - 49:58

We had a call with some Wall Street analysts and talked about the seismicity and the tremors under the Young mine. We talked to RBC, Jefferies, and Evercore, and they basically said the Island mine should be worth about 60% of the NAV, and as that grows people won't be as concerned about the Young mine. The mine collapse happened at the 9410 level, about a kilometer underground, which sounds deep but isn't really. What they need to do is first rehabilitate that specific mine shaft — it's one of three mine shafts, a series of tunnels and drifts. One of the drifts lost access. They'll have to put in some cable bolts and rebar. No equipment was actually damaged, no major equipment, and there are no injuries or deaths. So we don't think the stock should have sold off by a couple billion for, you know, 50 million of damage.

The capex spending according to the sell-side won't be that much. They're going out and taking out excess rock from the drift openings — the highest estimates are around 25 million to put in the rebar, take out the excess rock, and restart that to get that 2,500 tons back.

## Mine Impairment and Recovery Timeline - 51:23

The cleanup, even though the cost isn't a lot, will take months of human labor. So don't expect this to rerate right away. The mine has been impaired — remember they have four mines, but this specific mine has been impaired to the tune of about 2,500 tons a day temporarily versus total production of 5,000 tons per day, and they're hoping to get to 8,000 tons per day very soon, I think by the end of this year. So it's about 2,500 of 8,000; assume 5,000 tons per day by the end of the year instead of 8,000.

The sell-side was modeling 6 to 8,000, but because of the disruption in the rock they don't have access to some of the higher grades for that specific region. This was like 16% of NAV, now it's like 14% of NAV versus 18% on consensus. Production in ounces from the Young Davidson mine versus the whole was about 100,000 ounces, which is 19% of the 570k ounces per year, and apparently only 10% of this mine is offline. Worst case, even if that 2,500 tons per day instead of 8,000 tons per day is offline, you still have a 3 billion NAV at that Davidson mine, and that will likely recover once they finish cleaning it up. The bigger issue some of the larger Canadian pension investors are worried about is: could this company have been too aggressive in their mining grades, and could they have been this aggressive at other mines, which is what caused the tremors. Other issues investors have been frustrated with: there were issues that started after the Argonaut mill acquisition — the Magino Mill. They cut guidance a number of times. I read about the Magino Mill back in 2023; it had some cost overruns and was effectively a startup, so when they acquired it they probably didn't understand how to build a mill. Regardless, they cut guidance, so investors were already pissed off with management. Management had done a really good job over the last 10 years, but over the last two years they haven't performed as well. Once you saw the seismicity event, investors were concerned management wasn't managing safety and risks at this specific mine, so there was kind of a forced-selling event. You need at least two or three quarters of zero operational issues to effectively rerate the stock higher.

## High Grades at Island Gold - 54:32

On the positive note, switching from the Young mine to the Island mine: the grades are very high at Island Gold. They more than displace the Young mine, and from a margin and grade perspective there's a pathway to blend the overall grades higher. The reserve grade at Island Gold is about 10.5 grams per ton, which is five times as high as the Young Davidson mine. That translates into cash costs: only about 988 per ounce versus gold at 4,000 an ounce at Island Gold, versus cash costs above 1,700 an ounce at Young Davidson. So what they need to do is expand the mill and blend these two to get lower cost and higher margins overall. The current mill is operating at about 10k per day. Remember Young is now down at five, and most of the other production is coming from Island at five. If they double the mill capacity to 20k per day, they can increase production at Island, which is much lower cost per ounce, improving overall margins and revenue growth — which is what we're betting happens.

In terms of reserves, Island Gold has about 1.4 million ounces of gold; Young has about 1.2 million ounces, but the quality and cost of extraction at Island is worth much more. Peers include B2Gold (Mali), DPM (Dundee Precious Metals), and Lundin Gold, and OceanaGold in the Philippines is a similar size. Alamos used to trade at a premium to all of them — around 1.2x because it's a top-tier miner — but it's now trading at around the same multiple of about 0.8 times NAV. Historically it traded at a 30% premium to peers, and now it's trading at a discount to peers. Even on forward EV/EBITDA it's trading at a discount. For those reasons, after our calls, we're still confident it's a good name to hold longer term.

## Broader Market Update - 56:41

While I've been speaking, futures have sold off. S&P gave up almost an entire half-percent rally down to basically up 0.1%. The Russell is now negative, and the Nasdaq rally is down to about flat. Brent oil rallied about 2% at the open; the market didn't really seem to care, and now Brent oil is flattish as well. So we'll monitor this. I know why we had the initial rally — the renewed peace talks — but it looks like the market is fading that.

## BTBT and White Fiber Special Situation - 57:30

Now, BTBT and White Fiber. This is a pretty interesting special situation. BTBT is a company called Bit Digital. This was a Bitcoin miner; they transitioned into AI. They own 70% of a company called White Fiber, which provides artificial intelligence infrastructure solutions — basically support for data centers. It offers cloud-based HPC graphics processing units, hosting and colocation services, like DGXX. What's interesting is BTBT has transitioned from Bitcoin to cash and ETH, and they own 70% of White Fiber, which has a $1.5 billion market cap versus a full enterprise valuation of only 790 million for BTBT. And White Fiber doesn't have any debt. So BTBT had sold off with crypto, but its main asset is actually White Fiber. If White Fiber can meet its long-term EBITDA goals and justify its valuation, you can see the NAV chart at the bottom of page 26 — effectively BTBT trades at a 43% discount to its assets, a 44% discount, which is quite interesting.

## Micron (MU) Earnings and Strategic Agreements - 59:03

If you look at White Fiber and BTBT's market caps, you can see them on page 27. We'll quickly talk about Micron earnings. Micron made more profit this quarter than Nvidia made almost exactly one year ago. Next quarter's guidance is way more than Nvidia made in the third quarter of '25. Nvidia did about 35 billion in revenue and 20 billion in profit in Q3 of '25; Micron will do 50 billion in revenue and 35 billion in profit next quarter. That's quite impressive. Micron rallied to about 1228 after hours and maintained some of that rally. Micron's CEO Sanjay Mehrotra said there's no line of sight to when supply will catch up with demand. Shares were up 16% post-earnings. For the fiscal third quarter ending May 28th, sales rose to about 41.5 billion, earnings climbed to about 2511 per share (about $100 a share annualized). Analysts on average had estimated about 35.6 billion in revenue versus that 41.5 billion, and EPS of about $20 a share versus $25 a share. Another sign of Micron's widening profitability: adjusted gross margin more than doubled to about 85% last quarter versus an estimated 81.9%.

## South Korean Chip Market and Zero-Sum Game - 60:42

Micron works with Nvidia to integrate its memory into AI infrastructure. Earlier this month Nvidia's Jensen Huang confirmed his company will rely on Micron's HBM4 memory along with those of its rivals for its next-generation Vera Rubin platform, which requires more than five times as much memory as Nvidia's prior platforms. All three major memory makers have been jockeying for that AI data-center business. Micron's operating income estimate for the fourth quarter of 2026 is about 41.5 billion, which exceeds Microsoft at 38 billion, Google at 37 billion, Apple at 27 billion, Amazon at 22 billion, Meta at 22 billion, and Tesla at only 1 billion.

In terms of Micron's call presentation, the most important thing wasn't the quarterly print — it was the three slides on the strategic customer agreements. For 47 years Micron was effectively a commodity business: memory prices went up, Micron made money; memory prices went down, Micron lost money. Every investor who touched the stock lived and died by the cycle. But that cycle is now a little different. Micron has signed 16 strategic customer agreements — binding multi-year take-or-pay contracts with customers across data centers, consumer devices, auto, and industrial applications. Take-or-pay is the critical phrase: these are not soft purchase commitments or letters of intent but legal obligations with minimum spend obligations. Customers must buy specific volumes over the contract term or pay penalties. The demand is contractually guaranteed regardless of what happens to the broader memory market.

The agreements run from calendar 2026 through the end of 2030 — five full years of committed supply locked in at the exact moment AI infrastructure spending is accelerating into its steepest phase. They do this for about 40% of the business. The pricing structure makes the floor durable: the largest agreements have a ceiling price set at current market prices and a guaranteed floor price locked in through the entire term. That floor is set high enough that even in a deep memory downturn, Micron's gross margins under these agreements would sit well above any peak quarterly margin in the company's prior history. Customers have also agreed to provide 22 billion in cash deposits and financial commitments upfront, literally prepaying billions of dollars to secure their memory allocation years in advance. When fully executed, these SCAs will cover half or more of Micron's total company revenue. Approximately 40% of all revenue will be locked in at fixed or ceiling prices at or near current market levels, with the remaining SCA volume operating within a guaranteed floor-to-ceiling price band. What Micron has built is not just a better quarter but a durable, high-margin, predictable revenue engine on top of what used to be the most volatile business in the entire semiconductor industry — by converting their largest customers from buyers of a commodity into long-term partners with billions of dollars of skin in the game, at least for the next few years.

You can see quarterly revenue for Micron in the blue bars on page 30, DDR/DRAM spot prices on page 30, and NAND flash spot memory which peaked in March and has come down a little but not too much. The three SCA slides are on pages 31, 32 and 33.

Some good news: there are massive bonuses for South Korea's chip workers. A memory chip worker with a base salary of like 52k US is set to receive a bonus of about 410k this year, a complete game changer for South Koreans. SK Hynix employees are expected to receive bonuses of about 454,000. As you can guess, luxury jewelry sales are up 146% in Korea, luxury watch sales up 85% year over year. You can see how Micron's big win is a zero-sum event for megacap tech: memory producers like Micron, SanDisk, Seagate (STX), Western Digital (WDC), and Samsung saw their market caps rise by about half a trillion, while memory consumers like Apple, Microsoft, Meta, Amazon, and Google saw their market caps fall by half a trillion during the earnings sprint.

## Apple and Microsoft Price Hikes - 65:42

We've also seen Apple be forced to raise prices: its base iPad price is up about 28%, the Mac Studio up 25%, iPad Air up 25%, iPad Mini up 20%, and so on. Price changes across the product set — MacBook Neo, MacBook Air, Mac Mini, iMac — are on page 33. This was quite shocking and Apple sold off about 6% because not only are these price increases, it will also likely take a hit to its margin because it's likely not a full pass-through.

Microsoft also announced price hikes for Xbox devices due to memory chip costs just hours after Apple hiked prices. The Xbox Series S 1 terabyte increased from about 150 to 599, the Xbox Series X 1 terabyte increased 150 to 799 — massive price increases. The Xbox Series S 512 gigabyte increased 100 to 499. And these are all just memory prices. Microsoft is doing like a full pass-through.

## Micron and Anthropic Strategic Deal - 67:07

Micron also announced a strategic deal with Anthropic — a multi-year HBM, DRAM and SSD agreement. They're co-designing memory and storage around Claude's workloads. Claude is also going to be deployed across Micron, and Micron invested in Anthropic's Series H, which is interesting. The press release is on page 34. Micron is now Anthropic's investor, customer, partner, and supplier.

## Google Talent Drain and OpenAI IPO Delay - 67:29

Google fell last week, midweek, the most since May of last year. John Jumper — a famous Google DeepMind data scientist and Nobel Prize winner — left to join Anthropic. Another of Google's leading data-center staff members went to OpenAI. Then on June 24th there were rumors that two other senior AI researchers, named Jonas Adler and Alexander Pritzel, were going to Anthropic. This resulted in several billion of market-cap loss at Google. The Nobel laureate was John Jumper, and the star researcher was Noam Shazeer.

OpenAI also delayed its IPO to 2027. Sam Altman is reportedly considering postponing the IPO from September 2026 to later in 2027 given a lack of retail demand. He rejected lower options — he didn't want to bring the valuation down. The company generated about 13 billion in revenue in 2025, now reporting two billion monthly, but it's still well below Anthropic.

## SpaceX, Memory Chip Costs, and Sea Limited (SE) - 68:45

SpaceX has become the latest hyperscaler. An interesting data point: memory semiconductors are so expensive in the US now that US vehicle manufacturers are expected to cut vehicle production by 15 to 20%, or about 600,000 cars, in the second half of 2026. This is affecting cars, phones, tablets, and laptops, so it's quite a big deal.

Let's talk about Sea Limited (SE). We've discussed this name in the past. Sea Limited is a global gaming, e-commerce, and payments powerhouse based in Asia, expanding into other emerging markets around the world like LatAm. It has its own AI platform that is also growing, and overall the business has never been as cheap as it is today. An interesting update is Migoo, one of their new AI platforms, which appears poised as a daily-use AI companion rather than a simple chatbot. Based on the app, the product aims to help users plan their day, digest long emails and files, create content in their personal style, monitor tasks in the background, and act on instructions such as rescheduling meetings or sending alerts. Migoo is trying to sit closer to the user's daily workflow — remembering preferences, understanding routines, acting on context over time. The strategic opportunity, if linked to Sea, is less about building a better Siri or Gemini and more about owning a recurring AI surface across Southeast Asia and LatAm. If Migoo becomes useful as a daily planner, memory layer, and companion, it could capture intent before users enter a marketplace or payment app. Combined with local language, vouchers, merchants, payments, and delivery context, this could funnel demand into Shopee and Monee.

## Sea Limited (SE) Sum-of-the-Parts Valuation - 73:45

While AI adoption cost may slow margin lift — which is important for this thesis to play out — remember they could raise prices over time like Amazon did. If they do that same margin-expansion trick with their Shopee business, their EBIT could effectively triple and the stock could be a multi-bagger versus just the 100% upside we see today. But that will take a multi-year step.

The challenge is this is a crowded and expensive category for the Migoo AI app. Google, Apple, OpenAI, ByteDance, and others already compete for the AI assistant interface, while Google and Apple own the OS layer. So for Migoo to resonate, Sea would likely need meaningful product investment, marketing, promotion, and user education. This could slow near-term margin expansion, although such investments can be comfortably funded from existing positive cash flows. If Migoo gains traction and links into Shopee or Monee workflows, it could help Sea address AI disintermediation risk — a key overhang on the stock — while creating meaningful upside optionality.

The latest sum-of-the-parts model attributes about $29 a share, 16%, to the Garena gaming platform; about $134 a share to the global e-commerce platform across ASEAN and LatAm (split about 116 and 18, given the company hasn't been able to compete as much in LatAm with Mercado Libre, the best company in LatAm for this); and Sea Money has about $42 a share. We take a holdco discount of about $33 a share, divide by the 638 million shares, and compare to the spot 9133 stock price to get about 187 versus 91, or 105% upside in the long term. On page 37 you can see how Migoo's AI app works — it's a very cheap app, probably free, so the goal is for it to be a funnel into its other businesses. Sea Limited is a Singapore-founded internet company with businesses in digital entertainment, e-commerce, and digital financial services, dominant in e-commerce market share in ASEAN and Taiwan. It's grown at over 20% a year over the last few years and is expected to grow at 19% a year through 2030. It owns logistics and has a strong balance sheet with a competitive moat. The reason it sold off is AI and TikTok disruption — people buying directly on TikTok — and unfavorable unit economics in ASEAN, which are now improving. Price drivers: rate shocks, growth de-rating, Garena slowdown, India's Free Fire ban on gaming. There's a relief rally as cost cuts signal the pivot from growth to profitability. First-quarter profit validated self-help and margin discipline with margin expansion. Many forecast 2024-2027 revenue growth of well over 20%; adjusted EBITDA is supposed to grow at over 40%. The cash balance is roughly 10 billion, over 20% of the company's market cap, with no debt. Upside factors: stronger-than-expected user growth, share buybacks, and seller take-rate increases. Downside: weaker consumer spending in the regions (Asia was slowing because of the Iran war), slower user growth especially from increased competition, higher-than-expected credit costs for Sea Money, and new entrants.

## Domino's (DPZ) and Avis (CAR) Updates - 78:15

You can read about Sea's business model on page 40. Quickly — we're not taking a position here, but Domino's has taken a bloodbath. It's raised its dividend almost every single year since I started covering the company in 2016. It's continued to take market share from Pizza Hut, the big loser, from over 40% market share to high 20s; Domino's has gone from mid-30s to over 50% market share in fast-casual delivery pizza; Papa John's has been flat. Despite all that, Domino's stock is down from almost $500 a share to high-200s. Part of that is fear due to the GLP-1s, part is same-store-sales deceleration, but it's still the best delivery-pizza company in the US. A name to flag on the large-cap side.

Avis, which we had some pretty interesting option trades for, received a cash windfall from Pentwater for illegal short-swing profits. You can read that on page 42. The hedge fund had to pay several hundred million in an 8-K filing for selling the stock before the appropriate time, so there's a proposed $650 million cash windfall to Avis. The stock rallied a little on that news.

## European and Chinese Economic Outlook - 79:48

UK service PMIs fell to 48.7 versus 50.1, so the UK looks like it's entering some sort of recession. The French economic outlook is also very downbeat — unemployment's well over 8%, and the prospect of a rebound remains quite low. Business confidence edged up only marginally in June. Business sentiment in services remains very weak and at its lowest since 2021, with deeply negative expectations across all subsectors, which is very bad for the overall European economy. Most worrying is the continued deterioration of the labor-market outlook in June, with the employment climate indicator falling to its lowest level since 2013.

China — the KWEB Chinese tech entered a bear market, with the Hang Seng China Enterprises Index extending its drop to about 20% from its October 2nd peak. So everyone can be wrong, even Appaloosa; I don't know if he sold everything yet. We did see some crazy 12% moves every day in the DRAM ETF — down Tuesday, then up with Micron earnings, then down again. South Korea's tech-heavy Kospi also fell 10% on Tuesday before Micron earnings. According to Dan Ives, investors worried about Tuesday's sell-offs were missing the bigger picture that there's still strong demand for the memory names.

## Oracle, Nvidia, and SpaceX News - 81:47

In other news, Oracle said on Tuesday that the adoption and deployment of AI technologies across its operations has resulted in reductions to its workforce. The company's global headcount shrank to 141,000 full-time employees as of May 31st compared with 160,000 a year earlier, and it's going to fire more. Nvidia raised about 25 billion of debt after the deal was reported more than three times oversubscribed — a decent vote of confidence from the bond market. SpaceX, the new hyperscaler, also raised about 20 billion of debt, and while it was received positively initially — about 89 billion of peak demand on the 20 billion book — the bonds started to sell off a couple days after, and the 2056s were actually quite an interesting short because they're super long-dated. Even a 1% increase in credit spreads, 100 basis points, with 30-plus duration could result in the bonds falling 30 points. Quite an interesting hedge to the overall AI trade, given SpaceX, with its huge data center in Tennessee where Google is paying it several hundred a month to use, is effectively becoming a hyperscaler.

## Trump Meeting with Defense Contractors - 84:06

On page 48, Trump hosted weapons makers in an Oval Office meeting on Wednesday. Relevant names: the Big Five defense — Lockheed Martin, RTX, Northrop Grumman, General Dynamics, and Boeing; small arms and tactical gear — Smith & Wesson, Sturm Ruger, AMMO, and National Presto; drones and AI — AeroVironment (AVAV), Axon, L3Harris; and military shipbuilding — Huntington Ingalls and Textron. We'll get some discussion of what came out of it. It could be good for some of the defense names that have sold off.

## Private Credit, Howard Hughes, and MicroStrategy - 84:56

In other news, Apollo's private credit fund got about 17% redemptions — unreal — and they only honored about 5%. Ares, supposed to be one of the best private credit firms, got about 14% redemptions. So this private credit issue is not over. Private equity bosses are so stressed they're actually taking loans on carried interest as their payouts stall and liquidity dries up. A quick snapshot from JPMorgan on Howard Hughes with a low-80s sum of the parts, on page 51. The Italy relations blow up on page 52. But on page 54 is interesting — some graphs on the STRC MicroStrategy preferreds. MicroStrategy is still over 30 billion market cap; it saw its prefs trade down about 25 points on STRC. Its STRDs were already at a low dollar price because they don't have the variable dividend. Those 10% perps were all the way down to 50, and I believe the dividends are also not cumulative, so they're riskier, but those were down to 50 and we started to buy those. They rallied to 52, not bad. We started to buy STRC; our average for STRC is in the mid-to-high 70s. So if it goes back to par plus the dividend it's over a 30% return.

These are not without risk, so they're small pieces of a diversified portfolio, but we do think Saylor still has the ability to issue stock to pay dividends, and once the market fully understands that you might see a rally in some of these prefs. And the Asasta (ASST), which actually has no debt and is only prefs — whereas MicroStrategy still has debt — those were effectively trading at a 15% yield as well, and you could have bought them in very low 80s; they bounced a little on Friday. But with this war fear coming back, I'm worried crypto will sell off again this week, so I wouldn't be too aggressive.

## Q&A — Redwood (RWT) Analysis - 87:00

Someone on the call asked specifically about Redwood. I've been following Redwood and I own it. We had sold it — I think it rallied well above $6 a share, and then after earnings we took profits, and then we've been adding to it due to this rate volatility in the fives and high fours. Hopefully we get another round trip.

The reason Redwood has sold off is a combination of interest-rate volatility and a headline earnings shortfall from their Q1 report, even though the reaction was very positive to the earnings report in the beginning. People questioned it after the actual call and subsequent days. There was a little margin compression on GAAP profit and some book-value erosion of about 3% to about 7.12. But the stock trades at a big discount to book value — at 481 versus 712, that implies a 33% discount to book value. There are also macro housing headwinds: higher mortgage rates have kept the broader US housing market stuck at neutral, and slower housing turnover and affordability constraints have forced tighter industry margins on jumbo loan aggregation and originations, which is what Redwood is exposed to. But it does trade at over a 30% discount to book and has robust dividend coverage. Despite missing analyst projections, the 21 cents in earnings still fully covers the 18-cent quarterly dividend, about 0.72 annualized. At a 481 share price, this translates into a 15% dividend yield that remains covered by operating cash flows, at least as of the first quarter. It has about 200 million of unrestricted cash. So we still like Redwood — we just really need interest rates to come down for the story to play out well.

## Amazon (AMZN) and Google (GOOGL) Capex and Valuation - 89:20

On the Amazon and Google question, I think some of the valuations already reflect higher capex. Google has had a 180 billion AI pivot, transforming from a pure cash cow into a capital-intensive infrastructure giant. Google management guided toward 2026 capex of about 180 to 190 billion and expects 2027 to be higher, over 200 billion, which is why it had to raise 85 billion in equity. That was a big shock to the market — free cash flow collapsed, and if they're raising that much equity they must believe the ROI on this spend is going to be high. There's just no way to prove that yet. Google isn't building its data centers blindly: as of 1Q '26, Google Cloud revenue surged about 63% year over year to about 20 billion with operating margins at 32.9%, and a big backlog of about 168 billion.

Amazon also shocked the market by outspending everyone with about 44 billion in one quarter alone, on track for about 200 billion annual spend. There's some margin execution risk — early AI infrastructure depreciation is starting to hit. AWS segment operating margins actually fell 140 basis points to 37.7 in the first quarter as it competes with Google and Microsoft. If tech spending experiences a cyclical macro cooling in '26 to '27, Amazon will be left carrying a bag as they invest in their TPUs. But AWS did see a reacceleration, growing about 28% year over year to 37.6 billion in revenue, and it has this hidden chip business — custom silicon Trainium, Graviton, and Inferentia — which has scaled to about a 20 billion annual run rate, quite impressive. So there's some value there and Amazon's not getting credit for it. I think Amazon is still interesting long-term, and Google is interesting long-term, but I wouldn't be adding too much because their capex keeps going up. On this Mag 7 sell-off I would be adding to names like Microsoft, Google, and Amazon — I just wouldn't add too aggressively because it's very difficult to time a bottom in these types of markets.

I hope you guys enjoyed the call. We covered a plethora of topics, more than most calls. The recording will be out shortly, and you should be receiving an email as well. Wish you the best trading during the week.

---
_Readable notes for personal study. Full premium transcript, report & topics retained as PDFs in this folder. Not investment advice._