Jay Singh — Weekly SSR: weak payrolls cover 25% of the short book, the "so-far-so-fake" paper gains behind 47% S&P earnings growth, and a week of adds — T1 Energy, Uber, Reddit, Nokia, Redwood, Datadog, Alphabet

T1 Energy makes solar panels in the United States. Its selling point is where the parts come from: most solar cells in the world start with polysilicon made in China, and Washington is deciding whether to put tariffs or import quotas on it. Anyone who can promise an American-made panel therefore has something customers will pay up for — T1 says it will be above 60% domestic content by 2027.

That promise turned into a real order mid-week: a contract to supply 641 megawatts of panels to Clearway Energy Group. The stock jumped 15% on the news and kept going. Singh had bought at $4.70 and it finished the week near $6 — "a great trade."

The argument is about cash, not growth stories. Uber currently produces about $10 billion a year of spare cash after all its bills, and analysts expect that to reach $15 billion by 2028. Against a $139 billion company, that is a jump from roughly 7% to 11% of the price returned as cash each year — unusually high for a business still growing revenue 10-15% annually. It also just got upgraded to investment grade and is buying back $20 billion of its own stock.

Two engines drive it. Uber One, the subscription, has passed 50 million members who spend about three times more than non-members, which lowers what Uber has to spend to win each customer. And the advertising business inside the app is now running at $2 billion a year at high margins — the same playbook Amazon used.

The standard objection is robotaxis: if Waymo, Tesla or Zoox make human drivers unnecessary, Uber's network loses its value. Singh doesn't dismiss it, he just thinks the timeline is much longer than the share price implies — "that's going to take many years."

AppLovin sells advertising inside mobile apps. It reported a quarter that missed expectations by less than 1% and the stock fell by double digits. The number that actually scared people was cash flow, which came in 27% light — but that turned out to be the timing of international tax payments, not lost business.

The other explanation management gave is worth understanding: AppLovin's ad engine is an AI model that gets steadily better at matching ads to users, and that improvement ran slower than usual during the quarter, with the step-up arriving right after it closed. Meanwhile guidance for the next quarter was unchanged, meaning the company is squeezing more revenue out of the same customers.

Singh bought the drop, and his broader reason is defensive: advertising shown inside an app you already opened is much harder for an AI chatbot to disintermediate than advertising that depends on someone typing a search query. He thinks the stock can double in two years from the mid-$300s. An open SEC inquiry is the live risk.

The follow-through on last week's featured idea. He had bought more Reddit at $137 after the stock fell 23% on a quarter that beat on everything except US daily users. Days later it was back at $160.

TLT holds long-dated US government bonds, so it rises when long-term interest rates fall. He had started a small position the week before with a plan to add as yields approached 5%. The weak jobs report did the work: yields fell, and he says the add caught "the peak in the 10-year almost perfectly."

Redwood is a mortgage company whose share price moves opposite to interest rates. He first bought in the $3s, watched it rally above $5, and has now added again in the mid-$4s — it ended the week in the high $4s.

The appeal is arithmetic: it trades below the accounting value of the assets it owns, and it pays a dividend equal to about 15% of the share price each year. So you get paid a large income while you wait for rates to come down, which is the same bet as the TLT position expressed through a stock.

Nokia is a telecom-equipment maker that the market treats as a slow, cheap, unexciting business. Singh bought it as what he calls an option: you pay very little for the possibility of a re-rating, and lose little if it never comes.

The template he names is Intel — a stock that traded very cheaply, then pivoted its story toward AI chips and silicon and rallied hard on the narrative change rather than on immediate profits. He thinks Nokia is "a sleeper name" with a new AI story that is cheap relative to the market it could address.

A preferred share is a hybrid: it trades like a stock but behaves like a bond, paying a fixed dividend that must be paid before common shareholders get anything. Buying them below their face value pushes the effective income rate up.

This is a small income add to the part of the book Singh calls the ballast — several hundred high-yielding preferreds and small bonds that generate cash regardless of what equities do. He bought these at a yield above 9%. He does not name the issuer on the call.

Datadog sells software that monitors whether a company's own software and cloud systems are working, and flags security problems. Singh owned it, watched it double in the spring, sold it — and has now bought it back because it fell 17% despite reporting good results.

His reason for owning this corner of software rather than the rest: security and monitoring tools are hard for AI to replace, because the value is in the plumbing and the trust, not in generating text. Morgan Stanley reaches the same conclusion independently, putting Microsoft and security-software names on the winning side of AI in every scenario it models.

Google's chief scientist Jeff Dean announced he was leaving to start a new company, and the stock fell 5% — about $250 billion of value. Singh bought into that, on the simple view that no single researcher is worth a quarter of a trillion dollars: "it's just one data scientist."

The business supports it. Google's cloud arm grew 82% year on year, up from 63% the previous quarter, and expanded its profit margin faster than Amazon's cloud did. He kept the add small only because the stock had already run.

The caveat he raises in the same call is about accounting, not operations. Of the $9.11 per share Alphabet reported, only $2.88 came from actually running the business. The rest was paper gains on its stakes in Anthropic and SpaceX being written up in value — money it never received and could easily have to write back down.

Vistanc is what's left of CommScope after it sold most of itself off. The plan was always to hand the proceeds back: it already paid a $10 per share special dividend, and this quarter it confirmed another $5, funded by the $1.85 billion sale of its Ruckus unit that closed on July 1. Add it up and roughly $15 a share — $3.4 billion — has been returned, leaving a company with no debt, no preferred shares, and over $2.5 billion of book equity.

What remains is Aurora, the broadband equipment business — and that is where the quarter disappointed. Memory chips have become so expensive and scarce that Vistanc couldn't build enough of the boxes cable operators use to upgrade their coaxial networks, so it cut its full-year profit forecast by $25 million. Being dropped from the Russell indexes added more selling on top.

Singh added a small 5 basis points at $10.35, and is deliberately keeping it small: the special dividend is confirmed, but the memory-cost problem hitting the remaining business is the same problem hurting Apple, and he doesn't want size in it while that runs.

Barrick is one of the world's largest gold miners and a core long in the book, reporting Monday before the open. It sits alongside the two gold names he was actively adding to during the week — Alamos and Kinross — on the view that Chinese buying of gold has turned back on and central banks keep accumulating.

The trigger for adding was a flow signal rather than a company event: Chinese gold-backed exchange-traded funds recorded 14 consecutive days of money coming in, after months of outflows following January's price spike. Chinese retail demand returning, on top of steady central-bank buying, is what he wants to see before adding to miners. Alamos "did very well at the end of last week."

The second gold miner he added on the same signal — Chinese gold ETF inflows resuming — alongside Alamos, with Barrick already held.

Sea runs e-commerce, gaming and payments across South-East Asia. It is a core position and reports Tuesday; SSR has published its own financial model on the company to subscribers. A related tailwind he flags in the same breath: he believes China is "overdue for a stimulus announcement," which would lift the whole regional consumer complex.

Shopify provides the software small and medium merchants use to run online stores. It beat on everything that matters: revenue $3.58 billion versus $3.45 billion expected and up 34% year on year; the total value of goods sold through its platform ahead of forecasts; and operating profit up 68%.

The number Singh watches most is monthly recurring revenue — the predictable subscription fees, as opposed to the variable cut of sales — which came in at $221 million. Growing that is what turns a transaction business into a compounding one. He owns it, and notes the sell side is very bullish though price targets are conservative.

Atlassian makes the project-tracking and collaboration tools software teams use — Jira and Confluence. It was the book's biggest winner of the week: earnings came in at $1.87 per share against $1.50 expected, and the stock rose 32% before the open and nearly 40% intraday. "This is not a small beat."

Grab is the Uber of South-East Asia — ride hailing, food delivery and payments, run out of Singapore. It just reported a record quarter, raised its full-year outlook and added another $750 million to its buyback.

The reason Singh finds it interesting is the balance sheet against the price. The whole company is worth about $11 billion, but roughly $4 billion of that is cash it already has, so the operating business is being valued at about $7 billion while producing around $700 million of profit before interest, tax and depreciation this year, on revenue growing 22%. The stock is down about half from its 12-month peak, and the buyback now running is worth 5-7% of all the shares outstanding — $1.7 billion repurchased since 2024.

He has no meaningful position yet: "it could be an interesting one to take a look at."

Nano Dimension is a 3D-printing company that has more cash in the bank than the entire company is worth on the stock market. It holds $433 million of cash with no debt, while the shares value the whole business at about $340 million. Subtract the cash from the market value and you get a negative $80 million — the market is saying the operating business is worth less than nothing, because it burns money.

That used to be fair. What changed is that activist shareholders forced management into a three-stage clean-up: sell the non-core divisions (Markforged went to Stratasys for $42.5 million, other product lines for up to $12.5 million), cut staff, and break the headquarters lease — removing $38 million of future commitments and lowering the annual cash burn by about $25 million. Stage three, now running, is a formal review of selling the company outright, merging, or handing the cash back.

Singh is honest about why the discount hasn't closed: a company still losing $9.6 million a quarter slowly eats the cash before shareholders see it, and this management has a history of spending its cash pile on expensive acquisitions instead of returning it. So it's a watch, not a position — "it only becomes a compelling trade if phase three results in concrete capital returns or a total sale."

Backblaze sells cheap cloud storage and backup. It rose about 50% on a quarter that beat on every line — revenue, the business-to-business storage segment, and profitability — and raised guidance.

The bigger news was a $335 million, five-year agreement with CoreWeave, one of the AI data-centre operators. For a company this size, a contract that large changes the earnings trajectory outright. Singh's own caveat: nobody yet knows the exact shape of it — "we still have to investigate what that 335 million is."

Micron makes memory chips, and this year it is single-handedly holding up the S&P 500's earnings. Its contribution was $25 of index earnings per share — roughly a quarter of the entire index's earnings growth for the quarter — on profits up 850% in 2026, with another doubling expected next year.

The supply picture got tighter again: Micron, Samsung and SK Hynix have sold every chip they can make in 2027, and customers are being allocated only 60-70% of what they asked for. Industry watchers now expect 2027 to be the worst of the shortage.

The thing to keep in mind is that this is still a commodity business. Because so much of the index's earnings growth now rests on it, any slowdown in AI data-centre spending shows up amplified — "which would push the Nasdaq a lot lower."

The cleanest statistic in the call: Microsoft is the only one of the big cloud companies still generating spare cash for shareholders after paying for its AI build-out. Every other hyperscaler is now spending more than it earns and funding the gap with debt.

It also did the two things the market rewards at once — held its spending forecast flat while guiding revenue above expectations — and its cloud arm, Azure, accelerated from 39% to 43% growth with 45% guided next quarter. The stock rose 16% in a session and added $450 billion of market value in a day, the largest single-day gain on record.

Two things to watch. Copilot, its AI assistant, is used by only 30 million of 450 million paying Office subscribers — the upside case, but also proof adoption is early. And its biggest cloud customer is OpenAI, which represents 40% of its committed future cloud revenue, tying Microsoft into the same circular AI financing loop as everyone else.

Lumentum makes the optical components that move data around inside data centres at light speed. The reason it comes up is policy: the administration is drafting a rule, expected via the FCC this year, banning imports of new Chinese data-centre components on espionage and supply-chain grounds.

If that lands, the main Chinese supplier of these parts — Zhongji Innolight — loses access to the US market, and the volume has to go somewhere. Lumentum and Coherent are the two American companies named as the direct beneficiaries.

The other US maker of data-centre optical components named as a winner from the pending ban on Chinese parts. Same mechanism as Lumentum: a rule that removes the largest Chinese supplier from the market hands its share to the two domestic alternatives.

Zeta sells marketing software that helps large companies target customers. It closed up 12% at a 52-week high as analysts raised targets — KeyBank went from $27 up from $22 — on early traction for Athena, its newer platform, particularly among its largest customers (those spending over $1 million a year), which is where the growth compounds fastest.

This is a merger arbitrage: Warner Bros. Discovery has agreed to be bought, and its shares trade well below the agreed price because the market doubts the deal closes. The gap is the return if it does.

Four things improved in one day. Britain's competition regulator cleared it at the first stage, satisfied that Universal, Disney and Sony provide enough competition. Britain's Culture Secretary separately declined to block it on public-interest grounds — the political risk Singh worried about most, because UK ministers have historically been the ones to object. Quarterly profits came in as expected despite weaker sales, because costs were cut. And the chief executive of Regal Cinemas wrote formally in support, following an op-ed from AMC's boss — together roughly 40% of US cinema attendance, and nearly half for blockbusters.

He puts the chance of completion at 70%, and with one of the widest spreads left in large-cap deals, that still makes the risk-reward attractive.

Palantir sells data-analysis software to governments and large enterprises. Its quarter was exceptional by any measure: revenue up 93% to $1.94 billion, and $1.22 billion of spare cash generated — a 63% cash margin, which is almost unheard of at that growth rate. The "rule of 40" combines growth rate and margin, and anything above 40 is considered excellent; Palantir scored 155.

The stock rose about 40%, its best day since April 2025, and in doing so inflicted large losses on Michael Burry, who was short.

AMD reported revenue up 50% and profits up 246%, with its data-centre business more than doubling, and the market ignored it — spooked instead by higher planned capital spending. Singh's note is simply that the reaction didn't match the numbers.

What matters next is Helios, AMD's new rack-scale AI system, which starts shipping in the fourth quarter alongside Nvidia's Rubin. Bernstein expects AI server shipments to compound 15-22% a year through 2028, which is the demand AMD is scaling into.

Caterpillar had been a short for the house as a bet on peak industrial valuations. This quarter went the other way: sales of $20.5 billion against $19 billion expected, a record order backlog, and full-year growth guidance raised to the mid-to-high teens.

The driver is worth noting because it links back to the AI theme — the strength came from generators and power equipment for data centres. The fear had been that this demand was cooling; the results said it isn't. Michael Burry, also short, was on the wrong side.

Shake Shack jumped 12% after Starboard Value's Jeff Smith revealed on television that his activist fund had built a stake worth several hundred million dollars. Activists typically push for board changes, buybacks or asset sales, which is why the market pays up on the announcement.

Singh's discipline here is the transferable part. He does not chase the pop — he waits for the stock to drift back to what's called the unaffected price, meaning where it traded before the announcement, and buys there. That way he owns the shares ahead of anything the activist actually accomplishes, rather than paying in advance for it. He also notes he didn't think the stock was cheap to begin with.

JD.com is one of China's largest online retailers, so its results function as a read on whether Chinese consumers are spending. Singh is watching it for that reason rather than as a position — and because he believes China is "overdue for a stimulus announcement," which would be the catalyst for the whole Chinese complex.

Nebius rents out AI computing power — it buys graphics chips, installs them in data centres, and charges others to use them. Singh bought it during the AI sell-off and it rallied, but it has given some of that back on a specific fear.

An insider note circulated saying Meta has excess computing capacity it may sell into the same rental market. If a company that large starts renting out spare capacity, rental prices fall — and rental prices are the entire business model for Nebius and CoreWeave. On top of that, Michael Burry has picked Nebius as a way to bet against the AI trade broadly, which adds selling pressure independent of results.

CoreWeave is the largest of the AI compute landlords, and it signed two deals in a week: a $335 million five-year storage agreement with Backblaze and a multi-year deal with Solidigm, SK Hynix's memory-storage arm. Both signal it is building out capacity for committed demand.

It carries the same single risk as Nebius: if Meta dumps its roughly 400,000 spare graphics chips onto the rental market, the price CoreWeave can charge falls. That is the one scenario that breaks the otherwise favourable outlook for these operators.

EchoStar was the long leg of a paired trade: own EchoStar cheaply as a way to hold space exposure, while shorting SpaceX shares into the period when early investors were allowed to start selling. That trade is closed at a small profit — and closing it was the right call, because SpaceX has since fallen about 50% and dragged the whole space sector with it.

He hasn't gone back in. The reason is mechanical rather than fundamental: large blocks of previously restricted shares are still becoming sellable across the sector, and that supply has to be absorbed before prices can stabilise. It stays on the watchlist.

The Electronic Arts takeover completed mid-week — Saudi Arabia's sovereign wealth fund, Silver Lake and Affinity Partners bought it for $53 billion, closing out the arbitrage position.

Singh's observation is about the price the buyers got, not the trade. They paid only about a 10-15% premium to where the shares traded last year for a company with more cash than debt and roughly $3 billion of forward earnings before interest, tax and depreciation. On an all-cash basis the return looks thin, but once the buyers add borrowed money to the balance sheet — standard in a private-equity deal — it becomes attractive. "They got a relatively steal."

Dominion has agreed to merge with NextEra, and the arbitrage depends on regulators letting it happen. Virginia's governor has now announced she will formally intervene in the proceedings — which does not kill the deal, but adds a political actor with an interest in extracting conditions. Singh's read: "a little bit concerning."

The buyer in the same transaction. Virginia's governor intervening raises the odds of delay or conditions attached to approval, on a deal that had been progressing cleanly.

A memory-chip maker riding the same shortage as Micron: revenue nearly quintupled year on year to $8.97 billion, with record gross margins of 84.6% — extraordinary numbers for a chip company.

The stock fell anyway, because guidance for the next quarter of $10.3-10.8 billion came in below the $11.1 billion analysts wanted. It is a clean illustration of a market that prices the future, not the past. The company also announced a $14 billion buyback. Singh had already sold his position at the prior week's peak.

Not a position, but a signal he thinks is worth reading. Delta's chief executive exercised his stock options and sold $19 million of shares after the stock rose 30% in three months.

The run-up happened because airline shares rally when investors expect the Iran war to end and fuel prices to fall. So the person with the best view of the business chose to sell into exactly that optimism. "He doesn't seem to believe that it's a good risk reward at this moment."

A small speculative holding kept for its stake in White Fiber, the data-centre subsidiary reporting this week. Singh describes it as "more of a speculative AI position" rather than a core one.

Nvidia sits at the centre of a money loop that Singh keeps returning to. It is considering guaranteeing $250 billion of OpenAI's data-centre spending — money OpenAI would then largely spend buying Nvidia chips. OpenAI in turn accounts for 40% of Microsoft's committed future cloud revenue. Microsoft accounts for 22% of Nvidia's revenue. And Nvidia accounts for 17% of Micron's. "How circular is that?"

Nothing here says the demand is fake — the shortages are real and Elon Musk just committed SpaceX to building its computing exclusively on Nvidia hardware. The point is that each company's revenue is increasingly another company's capital spending, funded by borrowing. If credit markets tighten, the whole ring contracts at once.

Amazon's cloud business had its best quarter since the pandemic, accelerating from 28% to 37% growth — a nine-percentage-point jump — with margins improving. That was enough for the stock to rally 15% even though the company guided both revenue and profit below expectations.

The catch is the same one as Alphabet's. Amazon holds a stake in Anthropic, whose valuation went from $183 billion to $965 billion during the year, and it books that increase as profit even though it never sold anything. For those gains to repeat, Anthropic must keep raising money at ever-higher valuations. If it doesn't, reported earnings fall in 2027 without anything changing in the actual business. Jeff Bezos also sold another $4 billion of stock in early August.

Meta committed the cardinal sin of this earnings season: it guided next quarter's revenue and profit down while raising how much it plans to spend, and it still has nothing to show customers are paying for the AI capacity — no cloud service, no way for outside developers to pay to use its models. The stock fell 8%, and would have fallen further without a forced-selling rally happening the same day.

The new information is worse for other people than for Meta. According to a note from an Nvidia employee, Meta has around 400,000 graphics chips it does not need and will have to rent out. That is enough supply to push down the price of renting AI computing power — which is the entire revenue model of Nebius and CoreWeave. It is the single specific risk that breaks the otherwise favourable case for those operators.

Apple had been the place investors hid from the AI spending frenzy — it isn't building data centres, so it looked safe, and the stock hit an all-time high intraday during the week.

Then it fell 7% on guidance that came in below expectations for both revenue and gross margin, and the reason was memory. iPhones, iPads and Macs all need memory chips, and the shortage driving Micron's record profits shows up on Apple's side as higher costs and constrained supply. The same trade that makes Micron a winner makes Apple a loser.

Hims & Hers sells prescription treatments directly to consumers online. Its problem is that a large part of the growth came from selling compounded versions of GLP-1 weight-loss drugs — copies made outside the normal branded supply chain, in a regulatory grey zone — alongside a partnership with Novo Nordisk that has since produced litigation.

Singh's characterisation is that this makes the stock unusually volatile rather than mispriced, and the SSR report lists it under names to reduce or avoid.

Super Micro builds the server hardware that AI data centres run on, so it should be a direct beneficiary of the build-out. The reason Singh flags it negatively is unrelated to demand: the company has had accounting problems and is under investigation. That is a governance risk that no amount of order growth compensates for.

Oracle is the warning light in the AI build-out. Because the big cloud companies will spend more on data centres in 2027 than they earn, they have to borrow the difference — and the bond market prices how risky that borrowing is through credit default swaps, essentially insurance against the company failing to pay.

Oracle's are "blowing out," meaning that insurance is getting rapidly more expensive. Capital markets are still open and funding the spending, Singh says, "but at an increasingly higher price." Oracle is where that price is rising fastest.

SpaceX has fallen 40-50% since its listing and taken the rest of the space sector down with it. The quarter missed: revenue of $7.8 billion, up 92%, but a loss per share worse than expected, even as Starlink subscribers doubled to 12 million and the order backlog reached $47 billion.

The structural issue is funding. Musk wants to grow computing capacity from 2 gigawatts to as much as 10 gigawatts by the end of 2027, all built on Nvidia hardware — "he'll have to raise so much debt in order to do that."

There is a second-order consequence that runs through this whole call: Alphabet owns a stake in SpaceX and booked its rise to a $1.77 trillion listing valuation as profit. Now that it has fallen back, Alphabet has to unwind part of that paper gain — which is exactly the 2027 earnings problem Singh spends the middle of the call on.

This was a two-hour trade, not an opinion on the company. A hedge fund blowing up forced a wave of buying to close out its bets against Doximity — a "short squeeze" — which sent the stock up as much as 240% before the market opened, on no company news at all.

Singh shorted it while it was still up 80%, around $37, on the view that a move that size with nothing behind it has to give back. It fell to $31-32 and he covered for more than 20% in a couple of hours. The lesson embedded in it: an extreme squeeze in a low-conviction name is a tradeable event in itself, and the exit is defined by the fade, not by a target price.