Jay Singh — Weekly SSR: the Nebius (NBIS) NeoCloud deep-dive, the RWTS 10% Redwood baby bond, a merger-arb sweep, the VISN customer-concentration update, the historic momentum crash, and the open-source-AI / memory-peak debate

Nebius rents out AI computing power — it buys huge clusters of Nvidia chips and lets AI companies pay to use them (a "NeoCloud"). It was carved out of Yandex, the Russian internet company, and is now run from Amsterdam, which it pitches as a selling point: European governments and firms increasingly want their AI kept on European soil under European law. The demand is real because the big clouds (AWS, Azure) are sold out, so AI startups rent from nimbler providers like this.

Singh's edge is the balance sheet and the hidden assets. Unlike rival CoreWeave — which is drowning in ~$40B of debt — Nebius has essentially no net debt, so it can keep building without constantly selling new shares. And it owns pieces of four fast-growing private companies (a database firm ClickHouse, self-driving AV-Ride, data-labeler Toloka, and ed-tech TripleTen) that he tallies at roughly $8 billion — value he says Wall Street analysts basically ignore. Wall Street's average price target is $244-265 vs a $177 stock, but it's a high-risk name (it swings twice as hard as the market), so size it carefully.

PayPal jumped 20% because Stripe (the big private payments company) teamed up with private-equity firm Advent to bid about $60 a share in cash — a 28% premium. Singh's point is the price is cheap: PayPal throws off about $6 billion of free cash a year, so at $60 the buyers are getting a ~10% cash-flow yield, and the ~8× earnings multiple is below what PayPal has averaged. When a bid is this cheap, other bidders often jump in — so he expects a "topping bid" that pushes the price into the mid-$60s. Because it's all cash, the buyers would take PayPal private (not use it as a backdoor way to list Stripe).

A merger-arbitrage win that already paid off: atai (a mental-health/psychedelics drug developer) jumped 33% when Eli Lilly signed a firm agreement to buy it for $6.75 a share in cash, plus up to another $2.50 a share in a "CVR" — a contingent payment you only collect if the drugs hit certain milestones. It's a done deal (definitive), so this is Singh reporting a special-situation that worked rather than a fresh idea.

This is the income side of the Redwood trade. A "baby bond" is just a bond that trades in small $25 pieces like a stock, so ordinary investors can buy it. Redwood issued this one in late May and it yields about 10% — high for an unsecured bond right now — so Singh keeps adding it to his income book. Practical detail: two ~60-cent dividends are coming (mid-August and mid-November), so buying before the August 14 cut-off nets you about $1.20 a share in payments. (This is separate from Redwood's regular stock, RWT, which he bought earlier at a 42% discount to book.)

A quick update on an existing Shopify long — the software platform that powers online stores. The bullish catalysts he's tracking: fast-growing AI-driven traffic to merchants, higher-margin financial services (payments/lending) scaling up, and winning bigger "enterprise" customers. The market wrongly treats Shopify as an AI loser; he sees it as an AI winner.

The health-insurance giant beat expectations and raised its full-year outlook (guiding to roughly $19.5-20 of earnings per share) — a step up from last quarter's guidance after the stock had sold off. A straightforward beat-and-raise on a name that had been beaten down.

Abbott (medical devices, diagnostics, pharma and nutrition) jumped 11% after finally posting strong results — revenue up 13% to $12.6 billion, ahead of estimates — following several disappointing quarters. (The report calls it "AVT," but the numbers are Abbott's, not Avnet's.)

Micron makes the memory chips AI servers and electronics need, and there's a shortage pushing prices up. After a research shop (Argus) cut peer SanDisk to "hold" and spooked the group, Singh flags Micron as a short-term buy — a trade off the washout, not a forever-hold. His backing: Micron guided next-quarter revenue 15% above expectations and expects the memory shortage to last beyond 2027, having locked in over $100 billion of committed customer orders with floor prices.

SanDisk (flash-storage memory, recently split off from Western Digital) dipped when Argus rated it "hold," which helped trigger the week's AI selloff. Singh brushes that off — "frankly not a big deal" — and calls SanDisk and Micron "both short-term buys." Same idea as Micron: a tactical trade on cheap memory names caught in the panic, not a long-term position.

Gold has been the worst-performing asset lately because "real" interest rates (rates after inflation) rose, which makes non-yielding gold less attractive. The big gold ETF, GLD, has seen $14 billion pulled out since March. Singh reads that the opposite way — as a contrarian buy signal: central banks and regular investors are quietly buying again now that gold is back under $4,000. His line: everyone rushed to buy at $5,600 but nobody wants it under $4,000, "which is what you're supposed to do."

Singh relays Dan Niles' bull case on Intel (which reports Thursday). The idea: the money in AI is safer at the "infrastructure" layer than in the fast-commoditizing business of building AI models. Intel wins three ways — the new wave of "agentic" AI needs lots of server CPUs (a bottleneck Intel supplies), its advanced chip-packaging is pulling in big cloud customers, and its US-government-backed factory business keeps improving. It's Niles' view being presented, not Singh's own deep-dive.

Asked about beaten-down "momentum" names, Singh says he likes AeroVironment — a defense-drone maker — more than the flying-taxi company Joby. The logic is simple: he'd rather own a real defense business than a cash-burning story stock.

On the same defense question, Singh says he "likes the big defense companies" — Lockheed among them — over speculative names like Joby. Established, profitable defense primes over story stocks; a general preference rather than a detailed pitch.

A knock-on beneficiary of the Chinese-AI boom: the hot Chinese AI model "Kimi" runs its computing on Alibaba's cloud for now, which Singh says "should be very good for Alibaba's demand." As cheap Chinese models get more popular, the Chinese cloud that hosts them (Alibaba) sells more compute.

The cheap way to own the SpaceX story into its upcoming share "unlock" (when insiders can finally sell, in August). Singh pairs it: he's long EchoStar (SATS, which he says now trades as "ECHO") and short the public SpaceX vehicle. Last week SpaceX fell more than EchoStar, so the gap between them should keep narrowing in his favor.

This was last week's featured special situation (the ex-CommScope shell that's mostly cash plus a growing broadband-equipment business, Aurora). This week Singh brought in an industry expert to pressure-test it, and the update lowered his enthusiasm. The math still shows real upside — roughly 19.5% on the stock, more on the stub once you strip out the cash and dividend, at only ~3-4× earnings versus a fair 6-7× — but the expert found the business leans on a small number of customers ("customer concentration"), so "not a tremendous amount of upside." Still cheap, but the deal also needs government approval. Hence the downgrade from a featured buy to a hold-and-watch.

TSMC, which makes the world's most advanced chips, reported a great quarter (earnings and margins both beat, ~66% of sales now AI-related). But the stock fell anyway because it raised how much it plans to spend building factories (from $60B to $64B). The market's frustration: when growth slows you want a company to hand cash back, not keep plowing it into capacity. A good business, but not a fresh buy here.

ASML makes the machines that etch chips — the only company that makes the most advanced ones. It reported great numbers and raised its sales forecast for the year a second time, yet the stock still fell on general AI fear. Singh's takeaway is more about market psychology than the company: "the market is very skeptical of these names" and refuses to believe it's still a growth business. He's puzzled, but not stepping in.

Netflix fell 12.5% on Friday — its worst day in four years — on slowing growth, even though it's spending more on content and pushing into podcasts, games and short videos. Singh thinks it's "trying to do too much," which is why growth in time spent was only 2%. He's not buying yet, but says it'll "be a buy again" this year — a wait-and-watch.

A merger-arbitrage situation: Warner Bros. is being bought (the Paramount deal) and the stock trades below the deal price. This week 12 state attorneys general sued to block it on antitrust grounds — but the stock actually rose 2% because the lawsuit contained "very few surprises," so investors had feared worse. At $27 the market is pricing about a 75% chance the deal closes. The remaining hurdles to watch are European and UK regulators (the UK is "the most risky").

A scary-looking headline that Singh argues is overblown. New York froze new permits for big data centers, and the local town added its own pause on expansions — which sounds bad for DGXX's ~140 megawatts of power in the state. But its sites are existing operations that are "grandfathered in," so they're carved out of the new rules and can keep running and upgrading to AI use; there's even a hardship-exception path. It's "not a clean all-clear, but not the outright ban some bears claim," and 140MW of power is a valuable asset in a power-starved AI world.

Commentary, not a position. A rumor that the EV maker Lucid was halting toward bankruptcy sent the stock crashing to ~$270-280; the company then denied it and said it has cash into next year, and speculative traders ("degens") bought it back up to ~$736. Singh is just noting the whipsaw.

Asked about Nokia, Singh calls it "the least interesting" AI name and not the one to rally first. But there's a case: about 30% of its revenue (optical and internet-protocol networking gear) is growing 10-12%, and earnings could grow ~14% helped by cost savings from its Infinera acquisition. On a growth-adjusted valuation measure ("PEG") it looks interesting even if it seems expensive on plain earnings — so it's a possible momentum-reversal name "to do more work on," not a conviction buy.

Asked why he prefers Nebius to CoreWeave (the other big AI-cloud rental company), Singh says CoreWeave is "also likely a buy" and looks cheap — but it carries about $40 billion of debt, burns a lot of cash, and keeps borrowing more. Nebius grows 3-4× faster with essentially no net debt, so he'd rather own Nebius; you could hold both if you believe in an AI recovery.

Via Dan Niles: Google is his favorite "consumer AI" bet because it owns the whole AI stack (chips, cloud, models, products), and its cloud growth has been accelerating sharply. It reports Wednesday and will be the first real read on whether cheaper AI tokens help or hurt. The caution: even Google isn't immune if its biggest customers cut their AI bills, and the market will zero in on profit margins. Presented as a key data point, not a table-pounding buy.

Ahead of Meta's earnings, the worry is whether it can make money from its enormous AI spending. The stock is up 20% in two weeks partly because Meta said it will rent out its spare computing power — but if it also announces it's spending even more on data centers, that could spook investors. Across the big cloud companies, the cash they generate has swung from about $300 billion to roughly zero as they pour money into AI.

Mentioned in passing as a position Singh holds that reports on July 22 — a software name he views as an AI enabler rather than a victim. No fresh call this week beyond flagging the earnings date.

Another held position reporting July 22 — one of the wide-moat ratings/data businesses that form the steady "ballast" of his book. He contrasts it with CME (which he doesn't own and whose stock has been hammered). No new view this week.

Deal housekeeping: EA shares rose on positive press around its take-private buyout, including a first-stage antitrust clearance in Europe. A live merger-arb situation moving toward approval.

A takeover-rumor flag: Lionsgate is reportedly exploring a sale of itself, part of the wave of media-company deal-making that also includes the Warner Bros. situation. Something to watch, not a position.

IBM badly missed and the stock fell more than 20% — its worst drop in years. The reason matters for the whole software sector: companies are diverting their "optional" IT budgets to buy servers, storage and memory now, to get ahead of price increases, which starved IBM's core software sales. Singh expects this "discretionary IT spending is worsening" theme to hit many software companies this earnings season, and adds that IBM has simply "been very slow to innovate."

Singh is wary of Oracle and "wouldn't add it." Its specific problem is being too dependent on OpenAI — and with OpenAI "definitely losing market share" to cheaper rivals, that reliance is a risk. A name to avoid, not a short per se.

Asked whether to buy the flying-taxi maker Joby at $7 or below, Singh says no. It burns roughly three-quarters of a billion dollars a year, will run through its spare cash within about three years, and isn't expected to even reach positive operating profit until after 2030. He'd rather own the big defense companies or AeroVironment among the beaten-down growth names.

The short side of a paired trade. With SpaceX's insider share "unlock" coming in August (which can pressure the price), Singh is short the public SpaceX vehicle while being long the cheaper proxy, EchoStar (SATS). It's a relative bet — he expects the gap between the two to narrow — not an outright call that SpaceX collapses.