Jay Singh — Weekly SSR: the "FOMC fumble" and the loss of faith in Warsh, capex spenders ex-Microsoft dragging the index, the Situational Awareness blowup, the Reddit (RDDT) add in the $130s

Reddit sells advertising against its discussion forums. It reported a quarter where almost everything improved — revenue guidance for next quarter came in well above what Wall Street expected, profit tripled, and the amount of money it earns per American user jumped 51% — yet the stock fell 23%. The reason was a single soft number: US daily users came in at 53.2 million instead of the 54 million analysts penciled in.

Singh thinks that's the wrong thing to focus on. Reddit is still adding users overall (up 18% worldwide) and is rapidly getting better at making money from each one, which is the harder half of the problem. He bought more in the $130s and thinks the stock is worth around $200 over the medium term. The model he uses is saved alongside this page.

Sable is trying to restart oil production off the California coast. It nearly ran out of money, then refinanced its debt and sold new shares — which diluted existing holders and crashed the stock to about $4, sending its options prices haywire.

Singh used that panic rather than buying shares outright. He sold "put" options expiring in January 2028 — a contract where he collects cash today in exchange for agreeing to buy the stock at $2 if it falls that far. Because the options were so expensive, the premium he collected means his effective purchase price would be roughly $1.20-$1.35, against a stock trading near $5. So he gets paid to wait, and only ends up owning shares at a fraction of today's price. The bet rests on two things: the refinancing removed the bankruptcy risk, and the Trump administration lets California offshore drilling resume. He calls it a highly risky story — which is exactly why he structured it this way instead of just buying stock.

This is a lawsuit trading at a discount. Fiery — the mobile-gaming company formerly called Skillz — just won a $719 million court judgment against a rival, Papaya Games. The entire company is currently valued at roughly $150 million (about $70 million once you net off its cash), so the award is several times the size of the business that won it.

Even if Papaya can only pay a third of the judgment, the stock "could be more than a double." Singh's range: $20-30 a share in a settlement, over $50 in the bull case, versus $9.16 today. The catch is time and dilution — the appeal could run more than a year, and a 2027 debt maturity may force the company to issue a convertible bond in the meantime.

Korea's stock market fell 44% from its high — not because its companies got worse, but because retail investors had piled into 3×-leveraged funds tracking two stocks (Samsung and SK Hynix, which together are over half the index). When those stocks dipped, the leverage forced automatic selling: 320,000 accounts were wiped out in a single day, and one Irish-listed 3× SK Hynix product lost 96% of its value in a month.

Meanwhile the actual businesses posted record margins and memory prices kept rising. That gap — a market at under 5× earnings on forced selling — is what Singh bought, adding 20 bps of the Korea ETF on July 29, just before it rallied.

TLT holds long-dated US government bonds; it falls when long-term interest rates rise and gains when they fall. It has just hit its lowest price since the 2008 financial crisis, because yields have been climbing all year.

Singh has started a small "tracking" position — a toe in the water so the trade stays on his radar — and says he would buy far more aggressively if the 10-year yield approaches 5%. His logic: if the Iran war ends and oil falls below $70, the 10-year is close to its peak, and long bonds bought at these levels would be the biggest beneficiary.

Redwood is a mortgage company whose share price moves inversely with interest rates. Singh added another 10 bps on July 29 and says he will probably add more. It gave back some of that gain when the 10-year yield spiked after the Fed meeting — but that is precisely the mechanism: "it's a name to add to if you think rates have peaked."

Dynex is a mortgage REIT — it borrows short-term money to hold government-backed mortgage bonds and pays out the difference as a large dividend. Like Redwood, it is a direct bet that long-term interest rates have peaked. Singh added 10 bps alongside Redwood on the same day.

The market has spent a year punishing big tech companies for pouring cash into AI data centers without proof it pays off. Microsoft is the exception the market rewarded, and the difference is discipline: it spent $41 billion in the quarter and still generated $19.6 billion of spare cash, and — unlike Meta — it did not raise next year's spending plan.

The business underneath is also accelerating: its cloud arm Azure grew 43% and passed $100 billion of annualised revenue, and its AI assistant Copilot now has over 30 million paying corporate seats — evidence that the AI spending is actually being sold to customers. One accounting note worth knowing: Microsoft extended how long it assumes its data centers last, from 15 years to 25, which spreads the cost over more years and flatters near-term profits. The stock rose 9% after hours and 13% the next day.

Amazon raised its 2026 data-center budget from $200 billion to $220 billion, and its spare cash over the past year swung to negative $7.6 billion. Normally the market punishes that. It didn't — the stock rose 15%, its best day since November 2022.

The reason is AWS, its cloud division: revenue grew 37% and, crucially, has accelerated for five straight quarters. AWS is about a fifth of Amazon's sales but most of its profit, so as long as it keeps speeding up, investors treat the spending as growth rather than waste. CEO Andy Jassy's framing is that the spending is front-loaded because Amazon is building sites from scratch — the buildings last 30 years even though the servers inside are replaced every five to seven.

Micron is one of only three companies in the world (with Samsung and SK Hynix) that make the memory chips AI servers depend on. Nomura published a forecast this week that Singh calls shocking: the memory market growing from roughly $80 billion in 2022 to over $2 trillion by 2030 — five to seven times what mainstream forecasters assume — driven by chip prices rising from $1.90 per gigabyte to a peak near $18.60 and staying there.

The reason China can't spoil it: its national champion CXMT makes only ordinary memory, not the high-bandwidth kind AI needs, holds under 10% of the market, and costs 30% more per chip to produce. Meanwhile Micron briefly traded at five times earnings during the July panic. If even part of Nomura's forecast is right, memory becomes one of the largest profit pools in the entire chip industry.

CoreWeave is a "neo-cloud" — it buys AI chips and rents computing time to companies that need it. Two things this week confirmed the rental market is tight: prices for AI capacity started bouncing (Singh had flagged these names as interesting on that basis, and they rallied hard), and Meta disclosed it is being offered a meaningful premium for its own spare computing power while still buying from neo-clouds. When the largest buyers are also would-be sellers at a premium, existing capacity is scarce and valuable.

IREN is another operator that rents out AI computing capacity. It is named alongside Nebius and CoreWeave as a beneficiary of the same disclosure: the industry has structurally underbuilt for AI demand, so whoever already has running capacity can charge a premium for it.

Bloom makes fuel cells — refrigerator-sized generators that produce electricity on-site, which matters because new AI data centers can't get grid connections fast enough. The quarter was extraordinary: revenue up 166%, profits nearly seven times higher than expected, and full-year guidance raised well above what analysts modelled. Every major US hyperscaler plus more than a dozen smaller cloud providers has now certified Bloom's units.

Singh owned a small position and trimmed it at $207 after the stock jumped over 30% in a day. That's profit-taking on the price, not a change of mind on the business.

Nebius rents out AI computing capacity like CoreWeave and IREN, and benefits from the same tight market. Singh sold half his position after it popped 27% in a single day — a discipline call on the price move, not a downgrade of the story, which he still cites approvingly.

A textbook round trip in a memory-chip name: bought a small slice at $10.60 on July 28 during the panic (when it traded at five times earnings), then sold the whole position at the top of a 30% weekly rally, just before the market rolled over again on the Friday. "We sold almost perfectly." He is out — not bearish, just flat after taking the money.

Software stocks like ServiceNow had been beaten down because hedge funds were running a popular pair trade: owning chip makers while betting against software. When one big fund blew up and was forced to unwind, that bet reversed violently and software names spiked. Singh owns ServiceNow and trimmed into that spike — selling strength that was caused by other people's forced covering rather than by anything happening at the company.

Warner Bros. Discovery has agreed to be bought, but its shares trade well below the offer price because investors doubt regulators will approve it. That gap has widened to about $5.53 a share — a 21% return if the deal closes — even as the head of the FCC publicly said he expects approval.

Singh's arithmetic is deliberately unglamorous: below $20 if the deal breaks, above $30 if it closes, from $26 today. That's roughly a coin flip on payoff, so he holds a small position rather than a large one — the attraction is that the odds of closing look well above 50% while the spread is priced as if they aren't.

The Wall Street Journal reported that Tesla is considering spinning off its Chinese operations — a step that would clear regulatory obstacles to merging with SpaceX. Musk denied it, but he has denied similar reports before, and there is a logic to it: SpaceX shares have fallen 50% since its IPO, which makes a combination with Tesla stock more feasible. Singh flags it as a live special situation to watch rather than a position. Separately, Tesla appears on the chart of large companies dragging the S&P 500 lower since June.

The cost of insuring Nvidia's debt against default has doubled in two months. The trigger is what Singh calls "a very big circular deal": Nvidia has agreed to guarantee up to $250 billion so OpenAI can lease data-center capacity, is in talks to lend another $350 billion so OpenAI can buy Nvidia's own chips, and signed a $500 billion-plus partnership with Korea's SK Group.

In plain terms, Nvidia is increasingly financing its own customers' purchases — which flatters today's revenue but means Nvidia carries the risk if those customers can't pay. A company that used to be a debt-free cash machine is now underwriting hundreds of billions of other people's commitments. That's what the credit market is repricing, even though the stock itself got cheap (15× forward earnings, down from 25×) in the sell-off.

Robinhood has fallen from almost $120 to about $85 after bouncing from the high $60s, and every operating metric is improving — roughly $2 billion of operating income. But Singh puts fair value near $100, so buying here means paying a premium valuation for perhaps 15% upside.

His answer is to hold, not add: he'd want the stock back at $60-70, or a clear pickup in crypto trading volumes, before committing new money — and he's more cautious on the crypto-regulation catalyst (the Clarity Act) given the political noise around it.

Qualcomm looks oversold on the charts and "slightly cheap," but Singh separates a technical bounce from a real bargain. Compared with its own history the valuation isn't unusually low, and the growth rate behind it is modest — nothing like the memory makers or AI-chip suppliers. It benefits from the AI boom at the margin, but it isn't the way he wants to own that theme.

ASML makes the machines that print the world's most advanced chips, and its stock fell on news that China has begun building a domestic copy. Singh's read is that the scare was overdone: China is targeting about five machines this year and twenty next, against ASML's 131 shipped last year — and they're slower, lower quality, and still rely on Japanese components. Bank of America estimates the whole substitution would cost ASML about 2.4% of sales.

He was tempted, but the discipline held: "ASML didn't sell off enough for us to get too excited." A monitor, not a buy.

DoorDash's economics look terrible at first glance. On a $100 food order it only keeps $13, and after paying for marketing, engineers, refunds and payment processing it's left with about 46 cents of real profit — under 2% of the order.

What matters is the direction: that number was minus 38 cents in 2022. Profit per order has improved by roughly 50 cents in three years. If DoorDash doubles its order volume and improves another 50 cents per order over five years, that's about $7 billion of operating profit by 2030. Singh doesn't own it — he's waiting for a sell-off, because the business is compounding faster than the headline margin suggests.

Alphabet raised its 2026 data-center budget again, to $195-205 billion, and had a quarter with negative spare cash. That's why it sits in the group dragging the index down. But Singh puts it on the disciplined side of the divide with Microsoft — both are still generating positive cash flow overall, and Google's cloud business is growing 82% at a $99 billion annual pace.

The interesting detail is its own chips: industry checks suggest Google wants 12-15 million of its in-house TPU processors deployed by 2028, more than doubling its needs. That is more than its main manufacturer TSMC can supply alone — meaning Intel has to succeed at advanced chip packaging for Google's plan to work. He expects Google to follow Microsoft in becoming "a little bit more mindful of free cash flow."

Meta is the mirror image of Microsoft. It raised its AI spending plan again — to as much as $145 billion, roughly a quarter of everything big tech will spend this year — and its spare cash for the quarter collapsed to $784 million, the lowest since 2022. It then guided next quarter's revenue below what analysts expected. The stock fell from almost $600 to about $530.

The core problem is that Meta has no cloud business, so unlike Microsoft and Google it has nothing to point to that proves the spending is being sold to paying customers. Investors also remember the metaverse — tens of billions spent for almost no return. On top of that sits $165-169 billion of potential legal exposure from lawsuits alleging its products are addictive to minors.

The one genuinely new idea: Zuckerberg said Meta will start renting out its computing power, because buyers are offering a meaningful premium over what Meta paid for it. That's a bullish signal for the companies whose whole business is renting compute.

Apple avoided the AI-spending backlash because it doesn't build data centers — until this week, when a different problem hit it. AI has driven memory chip prices up so violently that Tim Cook called it a "100-year flood," and those chips go into every iPhone and Mac. Apple even tried buying Chinese memory to escape the squeeze and was warned off by the US government.

Sales still beat expectations (up 16% to $109.4 billion), but the stock fell 8% because the profit outlook is now hostage to input costs it can't control, next quarter's growth was guided below expectations, services growth is slowing, China sales missed, and EU regulators are squeezing App Store economics.

Caterpillar has been "a rocket ship," and Singh thinks that's the problem. On the chart comparing today's valuations with each sector's own history, industrials stand out in the red as historically expensive — while communication services sit at the cheap end. Buying an already-expensive cyclical after a huge run, into a slowing economy, is the setup he'd rather be short than long.

Rather than betting against SpaceX's stock, Singh is short its longest-dated bonds — the 6.65% notes maturing in 2056. Very long bonds swing hard on small changes in perceived credit risk, so they're an efficient hedge against the whole AI-infrastructure boom souring. The trade is working: shorted around 98-99, they now trade at 86.

Two related notes: the equity is down 50% from its IPO price, which is why a Tesla merger is even being discussed; and a proposed FAA rule waiving some environmental reviews for launch sites would be a genuine positive for the operating business.

Circle issues a stablecoin — a digital token pegged to the dollar — and essentially all of its revenue comes from the interest it earns on the cash backing those tokens. That is not a business with a moat: anyone holding customer deposits can do it.

Which is exactly what happened. A consortium of banks announced they are building their own stablecoin, and Circle's shares fell sharply. Singh was short the stock last year and closed it: "I wish we had kept that short on."

TBF is designed to rise when long-term bonds fall — a way to bet against Treasuries. A subscriber asked whether to pair it with municipal bonds. Singh's answer was to avoid these products: they reset daily, so choppy markets grind them down regardless of whether the underlying view is right, and he had just spent an hour on how a 3×-leveraged Korean fund lost 96% in a month.

More importantly, the timing is late. With the 10-year near 5% and possibly peaking, this is the moment to reduce a short-bond bet, not add to it — "and if the war really ends, you take it off."