Jay Singh — Weekly SSR: big Mag7 earnings week, Nvidia's $250B guarantee for OpenAI, the Betterware de Mexico (BWMX) add, an AppLovin (APP) add, and a temporary Iran peace the Taco Index called to the week

Betterware sells household goods — kitchenware, storage, cleaning products, personal care — in Mexico through a direct-selling network: an army of independent distributors who sell to friends and neighbors, Tupperware-party style. Last year it actually bought Tupperware's Latin American business, which is why the numbers are moving so fast.

The setup Singh likes is a rare one: the business is getting dramatically better while the stock gets cheaper. Profits (EBITDA) are on track to roughly double from about 2 billion pesos in 2024 to 4.3 billion by 2027, yet the price investors pay per unit of profit has fallen 30%, to under 4× — the kind of multiple you pay for a business in decline. More than 80% of those profits turn into actual cash, which funds a dividend yielding 8.6% that has now been paid 26 quarters in a row. Tupperware chipped in 11% of revenue and 16% of profit in its very first month inside the company, and management thinks combined earnings are 36% higher than the standalone business.

Why is it this cheap? Wall Street reflexively discounts direct-selling companies (it fears the distributor network churns), it's an emerging-market stock, and the acquisition only just closed so nobody has confirmed the cost savings yet. Those three excuses are the mispricing.

AppLovin runs an advertising engine that decides which ad to show you inside mobile apps — and it has been expanding beyond games into e-commerce advertising, a much larger market. Singh added to the position this week and sees roughly 50% upside.

His key argument is defensive: with so many new e-commerce platforms needing to buy customers, AppLovin's business "is not really AI-disruptible" — AI doesn't remove the need for someone to match an advertiser to a buyer, and AppLovin's data and placement network are the hard part. The full workings are in the valuation model saved with this call (deck page 43).

Classic merger arbitrage. The snack-food maker Utz agreed to be bought by the German group Intersnack for $14.25 a share in cash, and the stock nearly doubled overnight from about $7.25. The deal is expected to close in the fourth quarter.

Singh's rule here is simply a price: "anything below $13 is buyable." Buying at $13 to receive $14.25 in a few months is roughly a 10% return for taking deal risk — and he judges that risk low because two snack companies combining raises almost no antitrust concern.

Reddit dipped on a rumor it might walk away from the $60 million a year Google pays to train Gemini on Reddit's conversations. Singh reads that as a bargaining move, not a breakup — Reddit is the sixth-biggest website in the world and probably wants $150-200 million a year, arguing Google gets more value from citing Reddit than Reddit gets back in traffic.

The reason to own it is the ad business underneath: revenue is tracking up 55-60% this quarter against a 46% consensus, and over 70% next quarter — i.e. growth is accelerating, not fading. With $2.8 billion of cash, no debt, and a share price of about 17× next year's earnings while earnings are set to grow 42% over two years, his conclusion is blunt: "those estimates are probably too low."

Roper is a collection of niche, high-margin software and instrumentation businesses — the unglamorous kind that quietly compounds. It's a position he already owns, and it jumped 10%, its biggest move since 2020, after beating earnings and raising full-year guidance.

The lesson he draws matters more than the stock: in a market obsessed with AI, Roper was one of the cheaper software names and "nobody was looking at that name" — so a merely-good quarter produced a big move. That's his template for where to hunt while everyone stares at the mega-caps: cheap, boring, un-followed quality.

ServiceNow is the software layer big companies use to run their internal workflows — IT tickets, HR requests, approvals. The fear across software this year is that AI makes such tools easy to replace. This week's evidence pointed the other way: earnings were "absolutely phenomenal," and JPMorgan's own technology chief said the bank still depends on platforms like ServiceNow for fraud, payments and lending, because the hard part isn't writing software any more — it's scaling it, securing it, and getting people to actually use it.

Singh's framing is that ServiceNow is "the AI organization engine": the place where all those AI agents get coordinated and governed. The market faded the good quarter anyway, which he finds inexplicable — "the market produces an endless stream of information but very little insight."

Intel's long-awaited turnaround is showing up in the numbers: it guided next-quarter revenue to $15.8-16.8 billion against a $15.1 billion expectation, with data-center processor sales up more than 59%. The reason is that the new wave of "agentic" AI — software agents that run tasks on their own — needs ordinary server CPUs alongside the AI chips, and Intel makes those.

His stance is measured: "probably a buy on the dips here, but it's run so much that I don't want to be aggressive." In other words, own it, add on weakness, don't chase.

Micron makes memory chips — the DRAM and high-bandwidth memory (HBM) that AI servers consume in enormous quantities. Three things this week say the shortage is still getting worse, not better: Morgan Stanley forecasts 25% price increases, Micron's HBM output is sold out through 2027, and a joint Micron/Meta paper showed that when a workload outgrows available memory and spills onto disk, it runs 38× slower — so buying more memory isn't optional.

There's a nice tell too: Elon Musk paused a Tesla earnings call to thank Micron for selling him chips at a reasonable price. When your customer publicly flatters you, you have the pricing power. Singh trimmed into the July 21 spike and added the position back at the end of the week — this is an actively traded position, not a buy-and-forget.

Western Digital makes storage drives, and it's part of the same memory-shortage basket as Micron and SanDisk. It jumped 8% on July 21 when Morgan Stanley called for 25% memory price increases. Singh trimmed the basket into that spike and bought it back at the end of the week — a trading position within a structurally tight market, with earnings due in this week's tech block.

SanDisk (the flash-memory business spun out of Western Digital) rose 9% in the same July 21 memory rally alongside SK Hynix, Western Digital and Micron. Same handling as the rest of the basket: trimmed into the spike, added back at the end of the week, held as an expression of the memory-price squeeze.

The news that opened the call: Nvidia is in talks to guarantee about $250 billion of financing for an OpenAI data center — a 10-gigawatt site SoftBank is developing in southern Ohio. A "guarantee" means Nvidia would stand behind the loans, so lenders will fund a project whose whole purpose is to buy Nvidia chips. Singh's dry read: "very shocking, because they were creating their own open-source AI — but I guess they need someone to buy their GPUs."

The bigger point is where Nvidia sits in the money flow. The cloud giants are burning their cash on AI; the chipmakers are receiving that cash. That's why his call is buy semiconductors on dips while waiting on the cloud companies — and why he expects that once the market digests everyone raising capex this week, "you'll see another bid for the semiconductor names."

AMD signed a deal with Anthropic over the weekend for tens of billions of dollars of AI servers — a big validation that there's a real second source to Nvidia. It sits in the same "buy the capex recipients on dips" bucket as Nvidia, Micron and Intel.

One note of caution on the hype: AMD's CEO Lisa Su says the AI accelerator market reaches $1.4 trillion by 2030, which Singh calls "absolutely absurd." He likes the order book, not the fairy tale.

A short follow-up to last week's deep dive on Nebius, the European company that rents out AI computing power. It has now switched on its first complete Vera Rubin Ultra installation — Nvidia's newest generation of AI systems — in a finished facility, meaning the capacity is live and can start earning. After the stock's beating, "maybe we see a bounce there."

Dynex is a mortgage REIT: it borrows short-term money, buys government-backed mortgage bonds, and pays out the difference as a very large dividend. That model gets hurt when interest rates rise (the bonds fall in value and borrowing costs go up) and rewarded handsomely when rates stop rising.

That's the whole bet. Singh thinks the 10-year Treasury yield "might have peaked at 4.7%" now that the war looks like it's cooling, and if he's right, "all of them are pretty much adds here." Dynex is his favourite of the group on yield — "a very, very high yield."

Redwood is the mortgage-finance company he's been buying in stages — first around $3, and he's still adding to the ordinary shares below $5. It's the same rate-peak bet as Dynex and Annaly: a high-dividend housing-finance name that re-rates when the bond market stops falling. He calls the earlier purchase "a great add from a three handle," and he isn't finished.

This is the safer, income-only way to own Redwood. A "baby bond" is a bond chopped into $25 pieces so it trades on the stock exchange like a share — you're a lender, not an owner, so you get paid before shareholders do. It yields about 10%, and it was the most recent thing he added in this part of the book. Same company, lower risk, fixed payout.

Annaly is the largest of the agency mortgage REITs — same machine as Dynex, bigger scale. The news is that it raised its quarterly dividend to $0.75, after paying $0.70 for about six quarters and $0.65 for eight quarters before that. That takes the yield to roughly 13.4%.

Why the raise matters more than the yield: these companies cut their dividends when the maths stops working. A raise is management signalling that the spread they earn has stabilised — which is exactly the confirmation you want before buying a rate-sensitive income name. "Annaly after the dividend raise is also an add."

Another agency mortgage REIT — it owns mortgage bonds guaranteed by the US government, so the risk isn't borrowers defaulting, it's interest rates moving. Singh says "AGNC is interesting" alongside Dynex, and repeats the umbrella condition: if you believe rates have peaked, every name in this basket is a buy here.

The Warner Bros. takeover arbitrage is still open, and the gap between the current share price and the agreed deal price is back out to about 15%. That's a large payoff if the deal closes, and after this week's news Singh is adding to it rather than trimming. A deal-completion bet, not a bet on the media business.

SPY, the usual S&P 500 fund, weights companies by size — so a handful of giant tech names drive most of it. RSP holds the same 500 companies in equal amounts, so the giants matter no more than the smallest member.

His closing call is that RSP beats SPY "simply because Max7 is underperforming right now." That's the index version of the entire call: the mega-caps are burning cash on AI data centres and their valuations are compressing (from 32× earnings to 22×), while the other 490-odd companies don't carry that burden. If you want the market without the capex problem, buy it equally weighted.

Alphabet's headline quarter was a record — $112 billion of net income — but 87% of it wasn't from operating the business. It came from marking up the value of its stakes in SpaceX and Anthropic on paper (and it can't even sell $80 billion of the SpaceX stake yet). Strip that out and profits grew about 2% from the prior quarter. That's why the stock fell 6%.

Underneath is the real story: for the first time since it went public over 20 years ago, Alphabet burned cash — free cash flow of minus $5.9 billion — because it is spending $195-205 billion this year on AI infrastructure. And its contracted future commitments (chips, data centres, electricity) jumped to $811 billion, up half a trillion in three months. The cloud business is genuinely booming (+82%, with a $514 billion backlog), but at lower margins.

The uncomfortable part is the advertising teardown Singh relays from a large advertiser: Google quietly abandoned its long-standing "second-price auction" (where you paid a penny more than the next-highest bidder, not your own maximum bid), widened keyword matching so advertisers pay for irrelevant clicks they can't switch off, and started blowing through daily budget caps by 2×. That's how you grow revenue while search volume shrinks to AI chatbots — and it's borrowing from the future. His verdict: "at the altar of AI capex, Google is sacrificing its golden goose of search." Not a short, but wait.

Amazon reports this week and Singh's advice is to wait for the reaction, not front-run it: like Google, Amazon will probably announce it is spending more on AI data centres, and the market has just shown it punishes that. He expects a possible "sell the news" drop — buy after, not before.

The specific worries: AWS has a high bar to clear now that Google's cloud grew 82%; expensive oil raises delivery and logistics costs and squeezes shoppers; and moving the four-day Prime event from July into late June pulled sales forward, flattering last quarter at this quarter's expense. (Grimly, Amazon is also a casualty of the war — Iran's IRGC claims it destroyed Amazon's main data centre in Bahrain.)

Same "wait for the print" stance. Azure has to clear a high bar after Google's cloud accelerated to 82% growth. The positive: as companies try to control AI spending, Microsoft's 365 Copilot is well placed because it lives inside the apps where office work already happens, so it doesn't require a new habit.

The risk is the 27% stake in OpenAI. Dan Niles is negative on OpenAI because it's squeezed from both sides — Google in consumer AI and Anthropic in enterprise AI — and Microsoft owns a quarter of it. Microsoft is also quietly adopting cheaper open-source Chinese models, which tells you where costs are heading.

Meta should have had a strong quarter of user engagement thanks to the World Cup. Two catches: last year's third quarter grew 27% versus 22% the quarter before, so the comparison it must beat is unusually tough, and if Meta launches a public cloud/AI API it will likely announce higher spending to support it — which the market is currently punishing.

The cushion is valuation: Meta trades in the mid-teens times earnings against about 22× for the S&P 500, so "downside could be limited." Wait for the report, don't chase it.

Not a position — evidence. Tesla fell 14% last week after missing on profit and signalling much heavier capital spending, the same punishment being handed to every company that says it will spend more on AI. The telling moment: Musk stopped his own earnings call to thank Micron for selling him memory chips at a decent price. Singh's reading: "he knows he's burning cash now, so he has to play nice" — a striking illustration of how the power has shifted from the companies spending on AI to the companies supplying the parts.

This closes out a trade that ran for months. Singh had been long EchoStar (the cheap way to own a slice of SpaceX, since EchoStar's value was tied to it) and short the listed SpaceX vehicle at the same time — a "pair trade," where you profit if the gap between two related things narrows, regardless of the market direction.

SpaceX has fallen further than EchoStar, so the gap has closed and the trade has done its work. His instruction: take the profit on the short side and sell EchoStar — "you can just get out of the whole position." This is a completed trade, not a negative view on the business.

The short leg of that same pair trade, now being closed. Because SpaceX fell more than EchoStar, the short made money — "take your short-covering profits on SpaceX" and exit both sides. Worth noting SpaceX's other appearance this week: Alphabet's $94 billion stake in it (of which $80 billion can't be sold yet) generated most of Google's paper profit. After covering, Singh has no active position either way.

Apple has an enviable position in one respect: it isn't spending tens of billions building its own AI infrastructure — it lets Google do that and partners with it — so it keeps the cash its rivals are burning.

But the case against it is straightforward. The estimates for this quarter and next assume revenue and profit margins that Dan Niles thinks are too high, because the components Apple buys — especially memory chips — are getting sharply more expensive in the AI shortage, and Apple can't pass all of that on. And at 37× earnings the stock is priced for good news. "He's not a buyer of Apple here." The longer-term bull case (the foldable iPhone upgrade cycle) is a story for later.

Asked directly about IBM after its big drop, Singh's answer was a clean no. He expects growth to slow further — corporate technology budgets are being redirected to core AI infrastructure, which starves IBM's traditional services and software — so "next quarter could miss as well. I would wait for another quarter or two." Its quantum-computing work sounds impressive but contributes almost nothing to earnings.

The blunt version: "I just don't see a reason to really own it. It's also not super cheap." He expects three or four quarters of subdued spending before any recovery — and would rather put the money into cheap AI names for a trade, or the overlooked cheap software companies like Roper.

The European chipmaker fell 17% — its worst day in a year — after missing earnings and guiding lower, and it was one of the reasons the whole market wobbled last week. STMicro serves cars and industrial equipment rather than AI data centres, which is exactly the split running through the semiconductor world right now: anything feeding the AI build-out is sold out, anything feeding factories and cars is struggling. Singh's only comment on it: "luckily, we didn't have any shares here."

OpenAI is the subject of the week's biggest headline — Nvidia may guarantee $250 billion of financing for a giant 10-gigawatt data centre being built for it in southern Ohio. Note what that actually says: the company needs someone else's balance sheet to fund its compute.

The competitive read is the negative one. Dan Niles remains bearish on OpenAI because it is "caught between Google in consumer AI and Anthropic in enterprise" — squeezed on both flanks by rivals with either a bigger consumer distribution or a stronger corporate foothold. That matters to public-market investors mainly through Microsoft, which owns 27% of it.