Jay Singh — Weekly SSR: a soft-data trifecta takes September hike odds under 40%, Unity accumulated at 35-40 for a 55 base case, Riot initiated at $19.85 on the Anthropic compute deal, half of Babcock & Wilcox trimmed into +40%, and WYFI's beat carried through Bit Digital

The new special-situations screener — thousands of ISINs and CUSIPs, built with Claude, live late September — 00:47

• The motivation is a stated weakness in his own process: "one thing that I want to get better at is not to miss names, especially small cap names." The screener is meant "to go through every single special sits out of the thousands of ISINs and CUSIPs, and to identify, using predictive analytics, what names are the most relevant for our trading."
• Scope, per the deck's page-2 preview: filings tied to M&A, spin-offs, strategic reviews, buybacks, rights issues, restructurings, liquidations, delistings and litigation, with a description of each event, a quick valuation of the company, identified catalysts, watchlist additions, and progress monitoring as each situation unfolds.
• Sequencing and delivery: "we have a team of a couple developers now with Claude. It's become a lot easier to do this type of thing… in late September we want to launch the screener. Once the screener is set up, we want to include some predictive analytics. And once the predictive analytics are set up, we want to add some additional functionality." It goes on the website for subscribers to run themselves: "not only to use on the website, but to effectively be able to use the screener to find your own ideas."

The risk plan, announced ten weeks early: smoother into October, then cut — 06:10

• "Until midterm elections or until we see some sort of an escalation in the war with Iran, I do think that the market volatility will be a little bit lower… Now, after November, I plan to take some risk lower. And the reason why I'm telling you guys that early…"
• The causal read on the de-escalation is tactical, not structural: "the President is only backing off on the Iran war because of a lack of munitions and because he doesn't want the war to escalate before the midterm elections. I think after that, geopolitical risks could escalate quite aggressively."
• The election-outcome risk is named explicitly: "especially if he loses the House, I think that there could be some retaliation as a result of that and some frustration… I don't really feel comfortable with the current setup after November, but between now and November, we are going to have relatively smoother sailing. So that gives us the second half of August, September and October."

The economic calendar — a soft set into a strong tape — 15:08

• Empire manufacturing 15.6 → 10.5 ("not a good indicator"); housing starts 1.427M → 1.35M, "while we think mortgage rates are peaking, they're still at the highest levels of the year, which is very bad for the housing industry."
• The leading index at 0.1% is the one he watches: "we're looking at like 2% GDP growth for this year… mostly driven by AI growth, not the consumer. The consumer has been strong, but we just don't see the consumer getting stronger from here."
• PMIs roughly flat: manufacturing 53.9 vs 53.9, services softer at 53.9 vs 54.6, composite 54.5 → 53.7.

Retail week — the earnings that test the consumer — 16:46

• "This is the biggest week for retail earnings. So after the retail sales miss last week, we're going to look at retail earnings and see if any of these retailers are going to discuss weakness in the consumer." Home Depot and Baidu Tuesday; Target, TJX and Lowe's Wednesday pre-market; Walmart Thursday, "the most important reporter on the day"; Alibaba, John Deere, Ross Stores, Futu also Thursday; BJ's Friday.
• The China leg comes with a catalyst: "we think the Chinese tech stocks are undervalued and there will be a stimulus announced later this year."

Caesars — the proxy, the $34 Icahn bid, and a break fee that just doubled — 19:32

• "We had hoped that the Caesars bid would have been higher than what we see here, but it looks like the deal is going to close."
• What the proxy revealed: the Icahn Group bid $34/share in cash on July 10, but faced "significant questions around financing, managing change of control provisions with existing debt and the amount of leverage… and whether the company had enough free cash flow to service that leverage."
• The deal-breaker was the founding family: "Icahn's condition for the Carano family was to roll over their equity into the new deal, something that they had not been willing to entertain." The extended go-shop expired midnight Monday, "meaning the termination fee jumps from 100 million to 200 million."

Warner Bros. Discovery — two legs retired, 20% → 15% — 19:32

• "We have a position in Warner Brothers… it did rally from like 26 to 28," because former California AG Javier Becerra — now a Democratic candidate for governor — said a settlement would be the best outcome in the state's suit to block the $110B Paramount Skydance / WBD merger, without taking a side on the merger itself.
• The theatre constituency flipped: Paramount offered AMC and Cineworld's Regal three-year contracts for 30 films a year, exclusive to theatres for at least 45 days and off streaming for 90 days. "This is giving theaters more power, and is going to help the anti-trust approval and the eventual settlement."
• Net: "that spread has tightened from about 20% down to 15%."

The rest of the arb calendar — Varex/Teledyne, Roku/Fox, PAYO, ATAI — 19:32

• New deal: "Varex Imaging was up 50%… Teledyne, ticker T-D-Y, will acquire all outstanding shares of Varex for [$18.90] in an all-cash deal," unanimously board-approved, closing 1Q27 (~$1.1B including equity awards and net debt).
• Milestones to track: PAYO votes Sept 14, ATAI Sept 8, ROKU/FOXA's HSR waiting period expires midnight Sept 8 after Fox withdrew and re-filed; CRNX cleared Austria, KVUE cleared Mexico, and the EC opened its review of the GTBG merger.
• He singles out the Roku S-4's background section as teaching material: 11 counterparties canvassed from March 2026, only three in real diligence, only Fox bidding, $154 → $160/share (a ~34% premium), Roku holders left with ~27% of the combined company. "It tends to be quite educational for people who are trying to understand how these deals come together."

The consumer paradox — record-low sentiment, resilient spending — 26:58

• The structural reason: "10% of US consumers now do 60% of spending, that's up from 30% 20 years ago," so aggregate spending tracks the wealthy, not the median. "That's one of the reasons why we want to monitor Walmart earnings next week."
• Why the surveys stopped working: the University of Michigan data show Americans "have rarely been more pessimistic" on personal finances and the labour market — because of AI, inflation and cost of living up 25% over five years — yet "the trend of the past five years has been for consumers to say one thing and to do something else entirely. Today, we are most interested in consumer behavior, not consumer attitudes."
• The wealth effect is doing the work: a falling savings rate in a strong economy is "a sign of high consumer confidence, even though the surveys don't say that," driven by retirees who own their homes, hold financial wealth and are insensitive to inflation or the labour market. On the income side, real wages are again losing to inflation, though the one big beautiful bill and tighter immigration enforcement lifted after-tax incomes for low-wage workers for the first time since the post-pandemic labour shortage (Bank of America data, deck pages 7-8).

Housing is the source of the sour mood — 32:33

• Sentiment has been poor all decade, "even during years like 2023 and 2024 when real income growth was actually strong. Something else is behind the sour mood… Something bad enough to make consumers angry, but not bad enough to stop them from spending."
• The mechanism: the jump from refinanced ultra-low mortgages to today's rates has produced historically low home sales, and — unlike the mid-2000s — higher rates have not produced forced sellers, so existing homes for sale keep declining. "Moving brings with it the prospect of living in a smaller home with a much higher mortgage rate."
• The consequence he cares about: "housing mobility is closely associated with job mobility, and job switching is historically linked with faster wage gains" — JOLTS separations and hiring both subdued. But the effect on spending is perverse: locked-out renters "may feel that they have more cash to spend since they aren't worried about saving for an impending down payment."

The divergence that ends the cycle — nominal GDP vs aggregate payrolls — 36:10

• "Consumers are unhappy, but they continue to spend… So what's the problem? Ultimately, the problem is how long this is actually sustainable for savings rates to go lower and lower and lower. I personally think that we have another couple of years left."
• The chart he flags: nominal GDP growth (a proxy for profit growth, and "so much of this GDP growth is coming from AI") is racing ahead of total weekly payroll growth, which is now declining. "One is growing much faster than the other, and that's unusual… Eventually one of these two things will happen": payrolls accelerate (profits transfer to workers) or nominal GDP decelerates via lower inflation, narrowing margins or slower real growth.
• The ordering of the break: "a slowdown in AI spending could mean a big decline in economic growth, whether it's a 2027 or a 2028… We think that the first thing to break will probably be AI growth, and then second will be the consumer, because these savings rates are not sustainable."

The July retail-sales miss, decomposed — 38:53

• Headline −0.6% m/m to $763.6B against +0.1% expected — "the largest monthly drop in over a year." The GDP control group −0.4%; ex-autos-and-gas −0.2%.
• Category drivers: non-store online retail −2.2% (the biggest single contributor, "primarily caused by calendar distortion and pull forward" — Amazon's Prime Day moved to June), motor vehicles −1.8% as June incentives rolled off, gas stations ~−1% on nominal price effects, electronics and appliances −0.5% "because of higher memory costs, higher prices for laptops, TVs." Strength: restaurants +0.5%, modest gains in clothing, furniture and garden supplies.
• Four causes he ranks: the tax-refund tailwind fading (April/May spending was refund-funded, exhausted by July), cumulative inflation plus credit-card balances and borrowing costs squeezing "discretionary spend for the poor, not the rich," the Prime Day pull-forward, and labour-market cooling — "this number four will likely see that in the August numbers."

The pref book against plain-vanilla fixed income — 41:56

• Year-to-date plain vanilla: bank loans +2%, EM bonds +1.8%, local +1.8%, EMD +1.7%, high yield +1.5%, munis +0.4%, MBS −0.5%, investment grade −1%, treasuries −1% — "absolutely abysmal."
• "At the same time, our prefs have been up mid-high single digits on the year… our pref book has outperformed almost every single fixed income class in the IG space, and in the high yield space." He included the Fidelity US fixed-income piece in the Q&A tab specifically "because it shows how much prefs have done better than fixed income."

Optical over memory; the Iran/oil update — 42:35

• Goldman on the divergence: both optical and memory fell 30% in July, but "optical is now outperforming memory 20 points to start August… because the market sees it as a bigger supply issue."
• Oil is desensitizing: September crude +3.5% on August 10 on no peace, then Trump denied Iran compensation and asked for US compensation instead, "which is quite bizarre. But the market seems to be caring less and less about the war over time." Renewed Israel-Hezbollah fighting in Lebanon and continued Hormuz tanker attacks left Brent near $89 / WTI ~$83 in the deck and $88.70 / $81.53 live on the call, with crude +40% year to date and the 10-year at 4.69%.
• The geopolitical oddity: the USS George Washington leaves the Pacific for the Middle East, temporarily leaving the Western Pacific without a US carrier, and Trump ordered joint exercises with South Korea substantially reduced — an 18,000-troop, 10-day drill was due to start Monday — over Seoul's refusal to join the Iran campaign. "I don't know why he's threatening now South Korea."

Babcock & Wilcox — trim half into a 40% gap — 46:11

• "It was up 40% in the pre-market where we trimmed half of the position. It's in the alerts tab and the educational discord." Shares finished +41.64%.
• The numbers behind the gap: revenue $319.7M (+130% y/y), net income $14M vs −$60M, FY26 adjusted EBITDA target raised to $80-105M, a $50M buyback, bookings +38% to $151M, and backlog +533% to $2.6B with a global pipeline above $14B — "which is massive."
• The demand source is the same AI-power thread that runs through the whole call: CEO Kenneth Young cited AI, utility and industrial customers accelerating power-generation investment, and the company secured another 1 GW of Siemens Energy steam turbines for anticipated data-centre projects.

Riot Platforms and Anthropic — the December-2027 start is why the entry existed — 47:21

• "We were able to buy that stock at around $19 at the end of the day. 19.85." The deal: 109 MW from Rockdale, Texas to Anthropic, $9.1B through June 2048, extendable to 286 MW and $16.1B — a 20-year compute contract.
• The valuation walk: $2.4M per MW of annual rent, 80-90% NOI margin, ~$11M/MW of capex, and at a 15× NOI multiple the contract is "worth about $15 a share… which means that Riot should be worth something like 35 to 40 bucks a share once this starts."
• Why the market gave it back: it spiked to 23 and sold to a 19 handle "specifically because this deal doesn't really start till December of 2027 with a full deployment in June of 2028… the market wants to see it happen." Sized as speculative — "as all of the miners are that have now shifted their business model towards AI" — but "it really validates their business model… a game changer for the company."

Dan Niles' tech update — the pain trade is higher, but hedge the gains — 48:50

• The bull case, five points: generative-AI adoption has "at least a year more to run," with token production up ~7.5× since end-January more than offsetting a ~50% token-cost reduction; combined OpenAI + Anthropic annualized revenue $29B → ~$100B with Anthropic profitable in Q2 (and, per friends of his, "looking to actually IPO in October"); hyperscaler capex accelerating +84% → +92% → ~+100% forecast supported by big-three cloud revenue 23% → 35% → 43% and operating margins 34% → 37% → 39%; the $500B financing deal with up to $125B from Nvidia adding low-cost money for non-hyperscalers; and the Situational Awareness plus retail liquidation clearing the froth.
• The negatives: government bond yields near the highs with a weak 30-year auction at 5.3%, the highest since 2007; Iran likely to "hold the Strait of Hormuz hostage past the US midterm elections" — an explicit analogy to the 1981 hostage release after 444 days, "just hours after President Reagan was sworn in"; and the exhaustion of the AI technical rebound.
• Where value accrues: "the infrastructure layer, which includes public cloud vendors such as Amazon, Microsoft, Google, and the semiconductor companies like Intel, which is his favorite semi company" (+0.8% on the week despite a $20B offering / 5% dilution, and +178% YTD). "The pain trade he thinks is likely higher."

Midterm-year seasonality — the arithmetic behind the October plan — 48:50

• Since 1990, from end-July through November 9 (which "covers the reaction to all midterm results"): in midterm years the median S&P gain is +0.9%, positive 56% of the time, with a median peak loss of ~6.2%; in non-midterm years, +2.7%, positive 59% of the time, with a median peak loss of only 3.5%.
• Niles' conclusion is bullish-with-hedges — "during midterm years such as 1990, the S&P 500 has gained about 4.2% to its peak before giving some of that back closer to the election… he would add some hedges back on further market gains and get more selective." Singh's own overlay is the same: "This is what I said at the beginning of the call before even going through this. We think that going into the midterm elections, October, you should cut risk."
• The political tail risk is priced explicitly: "with the momentum seen by the socialists, which are not business friendly, I see more risk than the normal, which means he thinks that the socialists could win the house."

Jensen Huang's $500B answer to the circularity charge — 54:02

• The structure: financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize "over 500 billion or half a trillion of third-party capital," moving the industry "from an era in which companies bought chips and built data centers project by project to one in which AI factories can be financed as productive infrastructure." Singh's translation: "by creating this $500 billion financing, he's effectively going to guarantee GPU purchases. So it makes sense for him to do this."
• The anti-circularity claim, in Huang's words: "the capital providers independently underwrite each project, including its customer, demand, utilization, cash flow and residual value. NVIDIA provides the platform, the investors make independent financial decisions" — with Nvidia's own exposure limited to residual-value support for up to 25% of an opportunity, "designed to complement, not replace, independent underwriting."
• The durability evidence Singh finds persuasive attacks the bears' depreciation thesis directly: a 2020-vintage A100 still in commercial AI-training use six years later with rising lease rates; one-year H100 rentals $1.70 → $2.35/GPU-hour (Oct-25 → Mar-26); cross-provider on-demand medians $2 → $2.70; B200s at $5.30-7/GPU-hour. "Their value is not fixed at installation… the installed base remains productive well beyond its initial depreciation period, which is one of the bear's main theses." Verdict: "I think in the short term, he is right" — but on ROI, "the return is in the usefulness of AI, which I would say needs to become revenue."

The red flag — 33,575 unsold private-equity companies — 1:00:19

• "Private equity is stuck with 33,575 businesses, which they have to freaking sell. And this number has been going up every single year" — 32,451 at the end of last year and only 15,923 a decade ago (PitchBook). "And 80% of private equity deals are just PE firms selling their companies… from one private equity firm to the other. And I think less than 10% are IPOs. So it's really just a game of hot potato, or a Ponzi scheme where these guys are just selling companies to themselves."
• The returns: "from July of 2022 to March of 2026, US private equity firms generally generated annual returns of only 6.4%… far below the 15% of the S&P 500 and the 19% of the NASDAQ." Exits: "since 2022, only 70 private equity backed companies have actually gone public over the last four years… from 2017 to '21, there were 424."
• Software is the epicentre — PE "gorged on software companies since 2021 during COVID, before the advent of ChatGPT when software companies were trading at all-time highs," and now "the value of many software firms has declined as investors worry that AI will cut into future earnings." The case studies: Thoma Bravo's $12B Proofpoint ("extended the terms of its loan by two years because it almost went bankrupt"), Blackstone's ancestry.com (bought 2020, maturities extended six years later), and Vista's Solera Holdings ($6.5B in 2016, IPO filed 2024, never completed). "And imagine the pensions stuck with this crap."

S&P Q2 — 15% revenue growth, 50.4% EPS growth, 32% once two names come out — 1:06:04

• Revenue: blended growth ~15%, the highest since 4Q21 (16%) and "more impressive because in 2021, we're coming off the COVID cycle." The estimate path was 9.5% (Mar 31) → 12.2% (Jun 30) → 15%. All 11 sectors positive; IT +35.9% (semis +77%, hardware +31%, electronic equipment +21%, comms equipment +20%, software +18%, IT services +4%). Strip IT and energy and index revenue growth falls 15% → 9.7%.
• Earnings: 88% of the index has reported, an 86% beat rate (vs 78% five-year / 76% ten-year), and aggregate earnings 29.2% above estimates — which would be the highest since FactSet began tracking in 2008, against a prior record of 23.2% in 2Q20.
• The asterisk: Alphabet at $9.11 vs $2.88 (a $98B other-income gain on unrealized Anthropic marks) and Amazon at $5.75 vs $1.82 (a $53.4B gain) — "combined 150 billion of gains." Excluding both, the surprise falls 29.2% → 11% and blended growth falls 50.4% → 32% (still the second straight quarter above 25% and the seventh straight of double digits).

Unity — the model behind 55 and 71 — 1:12:02

• The business in two halves: Create solutions (about half the company) is the real-time 3D engine publishing to 20+ platforms across iOS, Android, PC, PlayStation, Xbox, Meta Quest and Apple Vision Pro, sold as tiered SaaS and extending into digital twins, automotive visualization, medical simulation and AR/XR training; Grow solutions is the ad-tech stack that "generates a majority of the company's total revenue now" and "competes with APP" — the AI-driven ad network and user acquisition, the LevelPlay mediation auction, and offer walls/on-device placement via Tapjoy and Aura.
• The three thesis legs: Vector, the AI/ML ad platform, +23% q/q and past a $1B annual run rate two quarters early, helped by day-28 ROAS optimization replacing seven-day metrics, plus a deliberate mix shift out of the low-margin legacy ironSource network and Supersonic; a management reset under CEO Matt Bromberg (ex-Zynga) after the pricing-policy debacle, several hundred layoffs and EBITDA margins expanded to 27%; and the engine moat — a duopoly with Unreal underpinning 60% of top mobile titles and 70% of the top 1,000 mobile games, a captive audience for the monetization layer.
• The model: 2Q revenue $546.5M vs $514M, EPS $0.28 vs $0.22, contribution margin 82-83% giving "flexibility to reinvest and improve the margin trajectory," and against a $40 model price a $55 base case and $71 upside — "over 25% upside and over 60% upside in the upside case." Risks named: Axon 2.0, and a post-run-up multiple — "which is why we think you should buy more on a pullback." Watch item: "the linearity of performance in the quarters ahead, given the inherent unpredictability of underlying model improvements."

AI and jobs — Apollo's 7% displacement estimate — 1:19:26

• The framing he endorses: Goldman has AI raising labour productivity 15% over ten years and adding $7T to GDP (deck page 31), but Apollo's piece — the one he circulated — argues "the impact on job losses will take longer."
• The evidence: employers cut almost half a million jobs this year and cited AI in only 23% of cases, and "we estimate that over time, 7% of jobs are likely to be displaced, which is a lot less than was projected before," with the majority of jobs complemented rather than replaced.
• The split: "roles with defined tasks and predictable environments are more likely to be substituted while industries involving data, problem solving and human elements are more likely to be augmented. Here, technology, financials, health care and service sectors stand out to us as long-term AI beneficiaries" — and new AI job openings are skyrocketing.

The week's tape — CoreWeave, Nebius, WYFI, Gemini, SanDisk — 1:21:56

• CRWV, owned into the print: revenue beat at $2.6B, EBITDA $1.5B vs $1.43B at a 59% margin, backlog $104B (+46%), active power tripled to 1.5 GW with the target lifted above 1.85 GW, FY26 revenue guided to $12.4-13.2B and a year-end run rate of $18.5-19.5B; shares +17% pre-market.
• NBIS: revenue $582M vs $573M (+454% y/y), adjusted EBITDA $236M vs $175M, cloud revenue +514%, $8B of cash — +34% on the day, and Michael Burry's short "got completely smoked." WYFI +15% pre-market on revenue of $28.8M vs $19-24M plus $380M of new cloud contracts, carrying BTBT from the low 140s to 157.
• Elsewhere: Gemini crossed 1 billion MAUs from 400M in May 2025 ("because they're basically forcing people to use it"); SanDisk +14% on investor-day targets and the HBF claim that one HBF GPU can replace eight HBM GPUs (an 8× capex efficiency); RDDT's S&P 500 inclusion produced "a huge rally for our holdings"; SMCI missed on revenue ("too risky to get involved in"); HIMS beat but is being sued; Chinese makers shipped 97% of global humanoid robots (19,000 units, Agibot 44% share) with Unitree's $9B Shanghai IPO oversubscribed ~8,000%; and Trump is weighing a capital-gains tax cut with inflation indexing plus exemptions for home sales under $2M — "there has not been a major US capital gains tax change since 2003."

Workday's LBO as a software floor; Micron's $400B — 1:33:14

• "The Workday software LBO… could be one of the biggest software LBOs of all time. Silver Lake, which is a premier private equity firm with over 100 billion in assets, focused on tech is potentially looking to take over the software company, likely putting a floor on the software sector." Before the 18% pop, WDAY was −20% on the year against IGV −6% and the S&P +20%.
• The same unwind lifted a holding: "Team also, which we own, absolutely crushed it and rallied 35%, as software has rallied after the situational awareness blow up, given all of the shorts they had in software versus long semiconductors."
• Memory: Morgan Stanley models ~$400B of combined operating income for Micron across 2026 and 2027 on HBM, with gross margins 80-89% and operating margins 81-87%; BofA models $236 of EPS by 2030 with $100B of revenue locked in by long-term contracts. The corroborating datapoint from Monday: "customers remain largely insensitive to memory pricing in data centers."

Q&A — BDCs, and the "Bessent put" under preferreds — 1:36:12

• On whether the PE exit bottleneck hurts BDC recoveries: "it actually is bad for BDCs because BDCs own a lot of these PE companies that will need to refi and will not be able to service interest, which is going to mean more PIK interest" — i.e. interest paid in more debt rather than cash.
• On preferreds replacing corporate bonds if the long end breaks (10-year above 5%, 30-year near 5.5%): "there's a short-term risk there, but… after that short-term sell off, which we're trying to avoid, [Bessent] would effectively force Trump to pull back. So I do think there could be something short-term, but if the 30-year hits 5.5%, I don't expect a 10% sell off in prefs. I expect like a 3 to 5% type sell off in the riskiest prefs. And then for the more stable ones like Duke Energy, etc., probably less."

The closing idea list — two shorts and five watches — 1:39:12

• Shorts: GBFH — "a bank with like 50% downside if you're worried about interest rates… a $1.4 billion bank, valued as a gaming payments platform, but really an undiversified economy hotel monoline. This is half of the book." And DFNS — "a shell company run by Manny Shalom, which pivoted from FinTech to defense, and has very little revenue at 3.6 million. So it likely could be a fraud. He's done a squeeze and dilute type of thing with SPACs in the past."
• Asset-value watches: SMHI (Seacor Marine) — offshore vessels "under activist pressure, with Pointillist owning 7.2% and [Sofware] 3.5%. The stock's at about 10 bucks versus broker NAV… above $20 a share." NAVI — Gator Capital's long at 40% of tangible book, where "people used to see it as a melting ice cube because of legacy student loans, but private lending is now growing."
• Negative-EV net-cash watches: BASO — a $40M market cap with $38M of net cash after unwinding a money-losing IT business and selling [NetWolves] for $14M, leaving "30 cents of net cash with 12 cents of biz value." GRVY — a Korean studio at a $430M cap with $400M of net cash, "0.3 times EV to EBITDA once you take that cash out," making Ragnarok, with "high concentration." And OCTV — "an AI-powered SaaS business, which actually could benefit from AI, trading at 11 times EV to EBITDA. Need to do more work on that."