CNBC — research hub

Stock & name index

Aggregated across episodes and interviews; each ticker links up to the consolidated per-security page. Because this is a network-level archive, the "Current thesis" names who held the view — the Halftime committee member, or the interviewed guest. Newest mention first.

▲ Positive

► Neutral / referenced

▼ Negative

Overall thesis

• When the crowded post-Fed bet fails, buy the de-rated leaders and the unloved chips — but know the rally is narrow. On 2026-sep-21 oil and yields cracked and the S&P gapped up 1% (Nasdaq +2%). Terranova: speculators short Treasuries into the Fed "didn't get that" spike — "2 words risk on." Link reads AAII bears at a year high against GDPNow 5.1% and rising estimates and calls the analysts' pile-on "absolutely the wrong decision"; she bought NVIDIA near the lows at 17× (vs 35×) and holds Amazon at 20× (vs 44×). Lebenthal doubles Qualcomm on momentum plus a data-center story, likes Meta for the same reason, and final-trades Micron; Talkington ("oil and yields") owns QQQ and RSP and will add on the midterm drawdown, sells NVIDIA calls under ~232, and trades Uber 70–78. Lilly CEO Ricks reports Foundayo at ~1 in 3 oral starts and ~7 in 10 new Medicare GLP-1 patients (Terranova owns Lilly; Lebenthal prefers AbbVie/AstraZeneca/Vertex; Novo falls). Balance: Santoli calls it a narrow, defensive mega-cap rally, not "genuine risk," and Terranova warns the Xi–Trump summit may be priced. Other finals: Natera (Link), Datadog (Terranova).
• With the 10-year at 5%, the argument moves from direction to size — trim at the valuation band, not on the story. On 2026-sep-18 (10-year at exactly 5%, oil +30% in a month, S&P flat on its June high) the desk agreed on the facts and split on the arithmetic. Raskin: AI is "a financially unsustainable arms race" — ~$200B of cumulative AI revenue against ~$1T of capex this year. Sechan: the hike was "the best outcome," inflation is transitory, "we have been buying bonds like maniacs," and "we will overbuild… that is not today's problem." The Apple fight is the portable part: Sechan says "you can't be underweight Apple" because it owns the customer; Bill Baruch cuts it 14% → 8% at "2 standard deviations above its median valuation" and buys NVIDIA and Amazon; Lebenthal trimmed half but keeps a $15-basis core. Financials get split by model (capital-markets "haves" JPM/MS/GS vs deposit banks), On stays "a no touch" until estimates rise, and the final trades lean on AI networking (Broadcom, Arista, Cisco) plus a 9× Brazilian bank (Itaú).
• The day after a priced hike, tighten the fringes and follow the cost the Fed can't fix. On 2026-sep-17 the S&P bounced back to 7,636 and the committee split on posture, not direction. Simpson: "I'm not going to let one day's tape change my thesis… the bar for risk has moved higher" — more liquidity for covered calls, tighter stops, "with a 10 year at 5%, good enough probably isn't good enough anymore." Terranova separates the reflex (momentum ETF +2%, banks "totally in the momentum funds") from the durable rate trade (insurers: Allstate, Travelers, Chubb, Progressive), time-stops his own Cboe position after 90 flat days, and holds Valero and the refiners because diesel is "a supply shock" a hike can't touch — while telling non-holders to wait on J.B. Hunt. Ethridge locates the stress in AI financing: "$75 billion worth of issuances of hyperscaler debt just since July that it can't find takers for," Oracle borrowing against its own equity, and the OpenAI/Anthropic IPO pacing. Brookfield's Bruce Flatt dissents on rates altogether ("Did rates go up?… it just doesn't matter" for long-duration real assets) and names power as AI's binding constraint, with Westinghouse "the heart of the industry." Final trades: UnitedHealth, Goldman Sachs, Teradyne.
• When a theme is threatened, split it into who pays and who saves — and for the payers, ask who funds the backlog. The Amodei-essay selloff (2026-sep-14) is the cleanest case in the archive. Santoli: "maybe the spenders are going to be OK, even if the vendors have to do with less in the future" — the tape agreed (SOX −5.3%, Microsoft and Alphabet green). The furthest-out vendors took the worst of it because their value is years 3–5 of an order book, and Terranova named the channel that turns a slowdown headline into lost orders: "if you tell me we're slowing the rate of spending and if you tell me that yields are going to remain high, that's a disincentive to see more supply in the debt market… that is absolutely going to affect these industrial names." A day later (2026-sep-15) Brown ran the complementary screen — own what every contestant must buy: "I don't care what model you're rooting for… there is no AI without the appropriate amount of cybersecurity… tell me one government or one board of directors… even considering lowering the amount that they're spending on cybersecurity." And Rieder added the cross-asset test that caps it all: with a 7%+ three-year single-A portfolio available, "all of a sudden the alternative to stocks is real."
• When you fear a macro variable, name the corporate action it actually stops — and prefer channels that end in a financing decision. The archive's most portable macro instruction (2026-sep-08). Everyone on the desk can list the September risks; only Terranova ranks them and finishes the sentence: "are you more troubled by oil rising or yields? I think it's actually yields… if yields begin to rise, that's when you get the earnings degradation. And in addition to that, that's when you get the technology and the hyperscalers begin to pull back on issuing the debt… there's a price sensitivity. There's a point at which you're not going to issue debt because the yield becomes unattractive to the buyer, and those hyperscalers are going to step back." That converts an unfalsifiable worry about rates into a dated, reported event you can watch — an issuance calendar, not a multiple. The same channel appears one layer down at a single company, where Lebenthal's Oracle checklist puts the credit market ahead of the P&L: "I would like to see them not raise their CapEx expenditures… I think it's at 70 billion… the five year credit default swaps are over 200 basis points, don't want to see that going higher… ultimately what you look for… is the top line growth… 33% year over year. If they can hit that, then we can worry about the margins." Note the shape of the list: one item is a non-event (do not raise capex), which is the opposite of how a guidance raise is normally read.
• When a whole group is weak, stop explaining the members one at a time — and go looking for the beneficiary. The cleanest methodological disagreement in the archive (2026-sep-08), and neither side wins it on air. Terranova puts a cross-section against a set of anecdotes: "all these quick serve restaurant names the last several years, they have not been trading well. I don't hear anyone present the reasoning that potentially these GLP-1s are contributing to quick serve not seeing the type of demand that it saw 5, 6, 7 years ago… McDonald's, Shack, Wendy's, it's universal across the board… you can even throw up Domino's." Belski answers per-company (Domino's has "problems… on the operating side", Chipotle "headline risks… and issues from the leadership") — which is exactly what a shared cause is meant to displace — then tries a reductio that backfires: "then you can throw in Mondelez… Coca-Cola… General Mills… because at the end of the day, people are not eating those names," and Lebenthal simply agrees ("consumer staples has been nauseous"). Wapner supplies the missing half of any demand-shift thesis — the mirror chart: "I'm guessing that the Lilly chart against those looks like it's in the opposite direction." The transferable rule: an individual explanation for every member of a group that moves together is usually five explanations too many, and the test that settles it is the beneficiary's chart, not the victims'. Belski's closing concession ("I've never thought about the GLP-1 side of things") is the honest version of an untested hypothesis.
• Rotate into a sector with a numeric screen attached — and let it delete good companies. Terranova's healthcare discipline (2026-sep-08) is stated with a count, which is what makes it checkable: "we own 15 names in the JOET ETF in healthcare. The common denominator is you have to have the revenue growth… there's only three of the names that we own that don't have double digit revenue growth." The exclusion then follows mechanically and expensively — "medical devices… are trading awful. They don't have the revenue growth. It's Stryker, it's your Baxter, it's Intuitive Surgical" — three franchises nobody disputes the quality of, deleted by one number. His instruction to viewers separates the sector decision from the security decision: "I want you to be there because of what the characteristics are, what it could offer in a portfolio, but make sure you have the growth." The corollary worth carrying: the names sitting on the boundary are the ones to watch (Amgen "right there at around 10%"), and the thing that should change is the screen when the market's paid-for factor changes — not the screen when a name you like fails it.
• Price the input that reaches the index, not the one on the screen. Lebenthal (2026-sep-08) refuses the headline commodity: "it's not oil, it's diesel, which is at a record and shows no sign of abating. There's no inventories to fall back on. There's no China cutting demand in diesel. Diesel factors into everything. It's truck shipping, it's farms, it's crop prices." The three-part test embedded there is reusable on any commodity scare — inventories to draw on, a marginal buyer who can cut, spare capacity — and missing all three converts a spike into a level. Note that this coexists with Terranova's opposite crude call in the same hour ("by Election Day in November, you're going to be looking more at a 7 handle on the price of oil"): the refined product and the barrel are separate trades, which is also why the refiners printed 52-week highs on the day. The dated resolver is the same for both: CPI later that week.
• Date your systematic buy against the sell-side note that follows it. A small but clean discipline (2026-sep-08). Airbnb is upgraded to Outperform at a $200 target — and Terranova's rules-based JOET already owned it from "strong momentum on August 7th," before Josh Brown's best-stocks list and a month before the upgrade. Every fundamental he cites afterwards pre-dates both ("2nd time this year they increased their annual revenue forecast… price moves after earnings that they have not seen since 2020… a traveller that has not stepped back… the World Cup"). The value is in keeping the ledger: the gap in days between the systematic add and the street's call, tracked across names, is the only honest measure of whether a rules-based sleeve leads or lags — and it only means anything if the misses are recorded too (cf. the JOET's Robinhood sale on 2026-sep-01, "you struck out three times… It's a bad job").
• Refuse the basket — a group chart that turns is only a thesis once you can name the constituent doing the work. The archive's cleanest transferable instruction (2026-sep-04), and it is delivered as a compliment. Wapner opens on Santoli's six-month Mag-7-versus-S&P chart; both responders instinctively convert it to one name, and Simpson names the move itself: "what I liked about the opening from both of you is you didn't talk about the MAG 7 as an acronym. You both talked about NVIDIA. So what we're doing is we're isolating the winners within the MAG 7." Lebenthal's mechanism supplies the test that distinguishes a theme from index beta: NVIDIA's print was "the clearing event" whose read-through was delayed a week and then escaped its own sector — "it's not just chips. In fact it's things like Microsoft, it's things like Meta." And his evidence that the capex is real comes from the buyers' numbers, not the seller's: a sub-three-year data-centre payback (Andy Jassy), Alphabet's cloud +80%, Azure +40%. The durability test is a ratio, not a chart — earnings and guidance rising with price, so that "the stocks are moving higher, but in essence, they're getting a little bit cheaper" — which is the difference between a re-rating you hold and a rally you time. Sechan's dissent rides along inside the same fact: the move is "concentration and weakening breadth," so he would be a buyer into weakness rather than into this. Note against the archive's own record: the leadership returned while OpCo reports the Mag 7 turning negatively correlated with momentum for the first time ever — the factor the desk told you to wait on two sessions earlier was not the thing that had to prove itself.
• Enter with the invalidator attached — and prefer the one someone credible has just argued in front of you. Two members model it in the same hour (2026-sep-04). Lebenthal makes a genuinely non-consensus call and labels it before arguing it: "I do think the Fed is going to raise rates. Now I have to point out this is not the Cerity Partners house view… that's what I personally think… I don't think this market is priced for it" — sourced from a Kevin Warsh speech, timed by the political calendar ("if they're going to go, they need to go in September, because if they do it just before the midterms, the roof is going to blow off"), and resolved by a dated release: "it'll come down to CPI next week." Simpson then buys more Freeport and Agnico Eagle and volunteers his own kill-switch, which is Lebenthal's call: "here's where I could go wrong. If Jim's right about rate hikes… then forget the gold trade. I'm completely off base with this." The pattern to copy is the pairing — a position, and a specific dated observation that ends it — and it recurs across the hour: Baruch sells Amer Sports because the driver he bought (Chinese consumption stimulus) was "eaten away from tariffs," not because the brands got worse; Simpson trims TJX after a double beat because the core division's comp fell from +6% to +1%; and Lebenthal, whose Adobe CEO-catalyst trade arrived and failed, says so first ("the market wanted an outsider. It's just that simple") and then restates what the position now rests on and what would change his mind.
• Momentum update: the factor broke below the situational-awareness lows — and "washed out" stopped meaning "buyable." The sharpest change of stance in the archive so far (2026-sep-02). The tape: Mizuho reports the momentum baskets "have now completely round tripped and broken below their July lows, taking out the levels seen during the situational awareness unwind," with the move "increasingly self fulfilling as systematic strategies de-risk"; Goldman's TMT-and-momentum pair had its worst month on record in a series dating to April 2011. Terranova keeps his bottom call but detaches it from the entry: "I think it's washed out… but the other side of that is that I don't think you step in here until that factor proves itself, because in the wake of situational awareness you had that false start… you have to have a degree of apathy towards it" — a week after naming Micron as his tell. He also reframes the whole exercise as judgement rather than method ("it is not a science, it is an art. It is tacit knowledge", evidenced by David Tepper exiting SanDisk in Q2 in the best quarter the company ever had) and gives the destination if AI stays dead: the factor "is going to continue to rotate more and more into energy, healthcare and financials… the same thing it did in 2022." The live test he set — and refused to trade — was Broadcom: 21× forward against 32× in June and an 18× ten-year average, "sometimes there's no trade… but if they deliver 200% earnings growth and 85% revenue growth and that's not enough, then that AI momentum trade is going to continue to sit silent."
• Positioning beats fundamentals on the day — and the cyber complex is the case study. Palo Alto posts record earnings, "a quarter they have never had before," and falls 10½% (its worst day in ~2½ years, still +82% ytd), taking CrowdStrike, Okta, Fortinet and Datadog down with it (2026-sep-02) — exactly the reaction Terranova had warned about the day before. His diagnosis is the transferable one: "if you're relying purely on fundamentals, then this quarter would tell you that all the cybersecurity names should move higher. But there's more to it… you have to look at positioning, you have to look at sentiment… positioning is full and the expectations could not be exceeded no matter how well they actually performed" — and it takes weeks, not a session: "I don't think it's a one day event either… I'm going to sit long and painfully watch it." Weiss adds the structural leg that positioning alone doesn't cover: the AI labs themselves (OpenAI, Alphabet, Anthropic) are shipping cyber tools, "so I think the competitive landscape's changed here as well," on valuations he says can't be justified. Note Datadog is observability, not security, and fell anyway — factor selling follows baskets, not business models.
• The data-center risk hardened: unpriced, and it peaks in 2028 rather than at the midterms. The one place the desk's view moved materially against itself (2026-sep-02). Sam Altman supplies the supply-side worry ("unsustainable silliness of random new neoclouds… that don't have the revenue to support" their announcements) and the Treasury Secretary the political one, at the G20: the AI companies "have done a horrendous job of explaining themselves to the American people… we need a big reset on this" — which Wapner reads as "an absolute direct shot" from an administration that had been close to big tech. Liz Thomas prices it as a tail: "the government gets so involved and actually oversteps that it starts to affect earnings… a risk that is not priced in right now," peaking "into the 2028 presidential election" — a materially longer horizon than Terranova's "midnight on November 4th" from the prior session. Weiss points at the tape (GE Vernova down, Vertiv going nowhere) and names the retail channel that makes it political ("that is driving my electricity prices through the roof") — then supplies the counter-evidence from a private board seat: data-centre leasing prices "going through the roof." Lebenthal's counter-model is fracking plus fiscal self-interest: after December's newly elected officials take office, "they're going to promote data centers because that's where the money is that they can tax." Two useful discipline notes fall out: price the incentive behind any executive's warning about their own industry (Altman "doesn't want more going up because he's already locked in what he needs"), and find an outside price signal the speaker doesn't control.
• The network is not only the desk — a single-guest interview can settle a question the panel only argues about. The first non-Halftime item in this archive (2026-sep-02): Greg Abel on Squawk Box from Tokyo, giving the mechanics of the Alphabet block first-hand — Buffett initiated ~15 months earlier, then a Sunday-morning call in late May where "really, no terms or amount were set" and Berkshire named both ("10 billion and more… I'd recommended a 6.5% discount") — plus an operator's correction to the desk's own AI-power debate: energy is the constraint, but "we can produce the energy", the binding item is "how long it would take to get the sites prepared." With one hard number the panel never has (Iowa ~8% of utility load from data centres last year) and one scoreboard ("we have not had any specific site rejected to date" against rising community pushback). Treat guest disclosure as evidence and panel stances as positioning.
• Before acting on a base-rate statistic, ask which side of it you are on. The cleanest transferable test of the week (2026-sep-01). Brown: "it's very nice for somebody to say, in 61% of all Septembers the market the first two weeks goes down. OK, how do you know when you're in the 61 versus the 39? If you're not going to tell me that, then I'm not whipping positions around and generating taxable gains for my clients… a lot of it delves into this world where it's astrology for middle-aged men." The two supporting moves: extend the window (Link — September averages −2.7% over 50 years, but October is +2.8% and November +3.9%) and find a conditioning variable that is observable (Snipe — five of the ten best Septembers since 1950 came when the market entered September strong, and in midterm years). A statistic with no discriminator is context, not a signal — and its after-tax version is usually negative.
• Check whether the feared thing already happened: date the peak, measure the damage, find where the money went. Brown (2026-sep-01): "momentum as a factor actually peaked relative to the S&P on June 22nd. We are now 51 days past the peak. Momentum is down 13.7% since then, with the S&P up 2.4% — that is 16.1 percentage points of relative underperformance. So this is not something that's about to happen." The corollary is that the unwind has been constructive, because the index held its highs while the flow relocated into the names his colleagues own ("look at the healthcare rally… I got energy stocks") — corroborated by JP Morgan ("the momentum unwind is largely complete"). Terranova's version is the retreat-venue test: "while momentum has been down double digits, quality has actually been up 1½% — so the market has somewhere to retreat to", with one falsifiable tell named out loud ("Micron… that's going to be my indicator"). When one factor bleeds and another rises, it is rotation; when all of them fall together, it is de-risking.
• Own the cost of a rules-based process, and use it to override your own biases. The two halves ran ten minutes apart (2026-sep-01). Debit: the JOET July-31 rebalance sold Robinhood just before a bitcoin spike lifted it, and Morgan Stanley upgraded it today to $150 — Terranova refuses the escape hatch Wapner offers him: "you struck out three times. You can't wash that away. It's a bad job. You're going to make mistakes. That's this business." Credit: Brown on Airbnb, a name his own judgment rejects — "I would never say Airbnb should be considered one of the best stocks in the market. But pull the chart back… that's where a rules based strategy can help you. It forces you to reconsider a stock that you have a bias against." The confirmations he requires are the generalisable part: a break of the range that has held since the IPO, visible accumulation, and no ground given back on days the market falls.
• Track a spending boom in dollars, not percentages — and separate the noise's end date from the spend's. Link's invalidator for the whole AI-infrastructure complex (2026-sep-01): "if they start slowing spend dollar wise, not percentage wise… but if you're going to see 1.1 trillion this year, 1.6 trillion the following year, 2 trillion the following year, that is very supportive," with order backlogs supplying the visibility "that is the thing that I want to pay now for." Against 11 states' moratoriums she argues the friction lengthens the cycle by preventing a boom-bust. Terranova doesn't dispute the dollars but prices the noise, and does it against his own Vertiv and GE Vernova: intense headlines for 60 days, ending "at midnight on November 4th… then it comes down to actual legislation." Wapner and Morgan Stanley both refuse the tidy end date — the midterms are "an important signal, not necessarily the inflection point," material federal risk is "more likely after the 2028 elections," and the base case is "a conditional build out rather than broad CapEx destruction" — which leaves Wapner's valuation question standing: can you pay the same price today once the goal posts have moved out in time?
• When two names express one theme, buy the one with the fewest extra drivers — and say out loud that you reached. Terranova's Nutrien buy (2026-sep-01) is a complete worked example. The theme is priced first ("corn, wheat, soybeans, cotton, all up double digits and more in the month of August," plus a forecast El Niño event hitting global food supply) and confirmed in a pure instrument (Krinsky's DBA up 13% on the month, breaking out to multi-year highs). The sub-exposure is chosen on cost position — "potash is the cheapest of the nutrients… Nutrien is the largest retail producer of potash itself" — and the rejection carries the method: CF Industries is a "great ag name… but there is a strong correlation to where price goes for CF Industries" and whether the Strait of Hormuz is open, "so that's why I didn't go there." Then the entry discipline most managers omit: "the stock… is extended. So I might have reached for it here slightly. I'll buy some more in a pullback because thematically I want the exposure."
• Separate "will the quarter be good" from "will the stock go up" — and trim on size, not on view. Palo Alto into its print (2026-sep-01): Link forecasts a great quarter (RPO ~33% growth against a 20–22% whisper, total revenue +32%, margin expansion) and has still trimmed — for the honest reason, "it got to be a 6% position" after +145% off the April low — with the re-entry stated ("I would buy it back if it's down a lot"). Terranova supplies the reaction data that the beat forecast cannot: 11 straight quarterly revenue beats and a stock that "tends to fall post earnings" — while holding the sector view separately ("thematically, there's still fundamental strength surrounding the cybersecurity names"). The companion discipline from Brown: set the invalidation before the event — Dell's "$400 on a closing basis is your line in the sand… the buyers need to step in there and if they don't, it tells you the sentiment has changed" — and re-base every level after a corporate action (Amphenol's 145 stop halves on its 2-for-1).
• The catalyst-failure test: good news + no rally = a positioning problem, not a fundamental one. The cleanest use of the desk's own method (2026-aug-31). Write down in advance what counts as unambiguously good; score the event on that; then look at the price. NVIDIA cleared its own bar (~70% guided revenue growth, a rare 2028 view against a street at 44%) and the tape went red anyway — Talkington: "NVIDIA should have rallied but didn't… you have to ask the hard questions, why aren't these stocks rallying?" Terranova's read is that the marginal seller is mechanical, which is what makes the momentum unwind self-reinforcing: everyone running the test gets the same answer. Harrington extends it forward — "if NVIDIA couldn't be a catalyst, why would Broadcom be?" The invalidation is the mirror image: the same names rising on no news.
• Measure the factor, not the story — and then find where the money went. When unrelated names fall together, price the factor directly rather than reaching for a theme (2026-aug-31). The momentum index is −9%+ since July 1 against the S&P's +2.8%, the worst quarterly underperformance in 25 years, with July the second-worst momentum month in ~40 years, beaten only by April '09 — a historically extreme flow event with a fundamental cover story. Factor money relocates rather than leaves: Terranova "the software names are benefiting because capital is going away from the momentum names," with IGV near its 118 highs (Wolfe), Palantir "the signature of that move," and Marvell's rally at "a screeching halt."
• Run the fundability test on the destination before you rotate. Terranova audits his own trade (2026-aug-31): pull four-quarter-forward consensus earnings for each destination sector and ask whether it has historically met elevated expectations — "if you rely on history, you're going to be disappointed. It's only specifically been in the mega caps, in the technology names." Then eliminate destinations that are secretly the same trade ("I'm not sure we want to go to industrials, because that's part of that AI universe"), leaving only financials and healthcare with real weight — plus intra-sector rotation as the alternative. Saccocia adds the arithmetic limit: the better-performing sectors "don't make up enough of the index to really move the needle."
• Value software with the replicability test, and stress-test hyper-growers on cash conversion. Harrington's experiment (2026-aug-31): do the job yourself with a general model. It replaced Adobe for a design task, so Adobe and Salesforce become "impossible to put a valuation on"; it cannot recreate DocuSign — networks, security and integration pipelines — so a 13× multiple with a 10% free cash flow yield is "totally de-risked… buy it now." The balance-sheet twin from Talkington: 60% of NVIDIA's last-quarter revenue sat in accounts receivable — "why can't these people just pay their bills?" — a flag to re-check next quarter, not a thesis.
• Check a demand narrative against the physical pipeline, and trade the gap between priced and expected policy. Against the data-center backlash, Talkington goes to what has actually started: 60% of data centers approved for 2027 haven't broken ground (bullish for the length of the build-out, bearish for this year's delivery dates), with CenterPoint and PG&E both funding ratepayer rebates out of data-center load against a 200,000-account bot swarm pushing the opposite. On rates, Saccocia's version: three hikes are priced in, which is why financials and small caps are pressured ("the Russell's down more than everything else") — "but if you don't think that's going to happen, which we don't, the catalyst could be a rotation into those names." Resolution date: the mid-September Fed meeting.
• In a deal, price the seller — and time the M&A wave off the political calendar. Terminal example (2026-aug-31): Terranova owns Aon and still says "the better trade and opportunity here is the seller. It's KKR… making about $3.3 billion" on the $17B USI sale — "a big bite for Aon." Read the day's deals as a set (SLB into data-center thermal management, ONEOK into the Permian/LNG, Lilly $20B diversifying away from weight loss), express the wave through the alternative managers (Saccocia's final trade), and note the deadline Harrington supplies: "if we go purple… it means gridlock. I'll bet a lot of people are going to try and get deals in before that."
• Rank a scheduled catalyst by the width of its outcome distribution, not by the size of the company. Santoli's tiebreak into the biggest week of the quarter (2026-aug-21): Warsh over NVIDIA, because "the market seems to know how to prepare for and metabolize NVIDIA at this phase — we kind of price the likely move, we think we know what the raise of guidance is going to be" whereas with Warsh "there's a wider range of probabilities in what policy might be… that explains a lot of what's happening in yields, just that what-if factor you have to price in." Sechan's version from the top down: there is no meeting next week, but "they can indicate — kind of tip their hand," and the Fed's only levers (hike, or effectively QE while tightening the balance sheet) are both unpalatable, which is why he keeps returning to the one genuine relief valve — "the thing that would save everything is opening the Strait, because it relieves the inflationary [pressure]."
• When an event is fully priced for the reporter, trade the read-through set. Brown concedes Harrington's pattern — NVIDIA runs into the print and then yawns — and relocates the catalyst (2026-aug-21): "Jensen Huang is the AI Fed chair. He is to the earnings growth of the S&P what Kevin Warsh is to interest rates… there are two to 300 large cap stocks where what NVIDIA says about 2027 and 2028 will absolutely have an impact." The surface area is enumerated, not asserted: Dell $300B, HPE $74B, Arista $235B, Marvell $200B, Broadcom $1.7T, GE Vernova $270B, Eaton $165B, Vertiv $100B. Harrington's own mirror of it — "it might matter more to CoreWeave or Microsoft or OpenAI" — puts the uncertainty with the buyers of compute, not the seller. Wapner's operative test: the market needs confirmatory, not revelatory, and Brown's evidence for why that is enough is that multiples contracted this year while prices rose — nobody is extrapolating.
• Invert the discount-rate trade — and diagnose why rates are high first. Harrington's construct (2026-aug-21), the most transferable idea of the week: "if we take Rob's construct that if interest rates come down the momentum trade takes off because long-dated cash flows are worth more with lower interest rates — the opposite's true here." The cause is supply, not the Fed: "corporate debt supply driven by the AI trade, huge government debt supply — and the supply is keeping interest rates up. That favors short-term, here-and-now, cash-flow-oriented companies… because cash today is more valuable than cash in the future." She runs the identical screen twice in one show — healthcare (PFE +17% ytd, BMY +25%, TMO +6% on the week) and midstream energy (ET/EPD/KMI/MPLX). Raskin holds the invalidator: long rates stay "unless we get a material slowdown in growth" — and if they fall, "probably the reason for that is going to be something that scares people."
• Test a hot sector with internals, then decompose it into commodity vs business. Against Krinsky's "energy is running on fumes" (2026-aug-21), Brown measures instead of arguing: the E&P-only IEO is at all-time record highs, +54% vs XLE's +45%, and — the stat that settles it — "30% of XLE names are at 52-week highs this week. That's not an overbought sector." (He removes Exxon's ~30% index weight before reading the XLE at all.) Harrington then splits the same sector by what caused each stock to move: the majors are a commodity bet in an equity wrapper — "Chevron started the year at like 165, is trading at 200 now, and that is purely driven by the fact that oil went from $58 to $86" — and she doubts $86 holds given the Hormuz workarounds, so she rotates within energy into fee-based midstream rather than leaving. Sechan won't trim at all (energy overweight all year as a rates/inflation hedge: Suncor, CNQ); both sides land on Devon as the price-exposed name rising for its own reasons.
• Own the layer that doesn't require picking the winner — and put a hurdle on it. Harrington's Cisco case (2026-aug-21): "it doesn't matter who wins. It doesn't matter if Claude wins… ChatGPT… Gemini… Microsoft, Meta, or Apple. As the AI trade makes the technology world bigger and more complex, Cisco wins. They're the infrastructure behind all of that" — with a 5% free cash flow yield "whereas a lot of the other guys have given up their free cash flow." That same 5% hurdle is what makes her refuse Coca-Cola at 27× for 7% growth and the Mag 7 alike — and when a sector can't clear it, she changes the instrument rather than the rule: her final trade is the Microchip preferred at a 5¾% yield, "a nice way to get a little bit of tech exposure and still get some income." Her own flag on Cisco: after +48% ytd it is "a little premium to the market where it used to trade at a discount."
• Deletions from a rules-based list are the sell signal; the cannibaliser is the long side. Cboe, CME and Intercontinental have all fallen off Brown's best-stocks-in-the-market list (2026-aug-21), and the shared cause is structural: perpetual futures (24/7, high leverage, never settle) plus prediction markets attacking "some of the last most profitable commissionable activity at all of the brokerages and all of the exchanges" — a fear he calls legitimate, with political fights still ahead. The forced response defines the trade: "the brokerages have embraced this. They almost have to — they sort of have to cannibalize themselves," and on Robinhood's last report "this is far and away the best thing that they're doing right now in terms of margins — better than even crypto." Renick's counter-evidence: the exchanges "have fully recovered and are now rallying" because regulation may start "the join-them phase rather than the fight-them phase."
• Yield-curve control with an expiry date — and the flow map that follows. The Treasury doubled long-end buybacks the day after the 30-year printed above 5.30%, which Liz Thomas reads as the level at which it becomes uncomfortable (2026-aug-19). Terranova puts a calendar on it: "going into a midterm election, do you really want to see mortgage rates rising if you're the party in power?… It'll be 60 days, early September right up to the midterm election" — and the same incentive covers oil ("do I think in the last two weeks of October oil is approaching $90 to $100? No, I absolutely don't"). The actionable output is the flow, not the yield: capital out of the momentum factor, out of memory, out of semiconductors, and into healthcare and energy. Two limits to hold alongside it — Simpson's ("no matter what the Treasury does or what the Fed does, they can't control the long end") and Santoli's (the measure "has already undone a lot of what happened… we have to see what the half-life is").
• The one contrarian macro risk on the desk: a bear flattener out of Jackson Hole. Liz Thomas (2026-aug-19) is alone in pricing a tug of war between the Treasury and the Fed: the buyback flattens the curve, "which classically is bad for financials — however financials have done so well this entire year with a flattening curve that I don't think that matters right now." But "if Kevin Warsh comes out at Jackson Hole and says anything hawkish, the yield curve is going to get confused" — yields rising at both ends with the short end faster, a bear flattener that breaks the Operation-Twist-2011 floor analogy she is otherwise leaning on.
• Healthcare is the rotation that finally sticks — and it just got its catalyst. Moderna more than doubled (2026-aug-19) on the first-ever phase-3 of the Merck-partnered personalized cancer vaccine, which met its endpoint in melanoma in combination with Keytruda; Simpson's Merck angle is the patent cliff ("Keytruda, which I think is coming off patent in 2028, could be extended now for these other uses"), while Terranova refuses to chase it — the momentum score was fine, "the revenue growth was the challenge… nearly 55 times. That's a hefty price to pay. That's for 3% growth." Liz Thomas's durability test is the transferable part: last year's post-Mag-7 rotation into healthcare didn't stick, "this time it's sticking because now we have market movement upward both in technology and in healthcare… a sector you get paid to wait for." Her AI overlay carries a timing constraint — healthcare is a big AI beneficiary, "but the cost of AI has to come down first." Lilly (+371% in five years off the 2023 GLP-1 inflection) is the calibration for what the payoff phase actually looks like.
• Energy is a refining record, and Terranova is crowding-testing his own call. XLE hit a record on its eighth straight up day (2026-aug-19), but the driver is the 3-2-1 crack spread at an all-time high — refined products, with Ukrainian strikes on Russian refineries removing capacity — not crude. Terranova still says "you need to own the refiners, especially going into the fall," then applies his own positioning discipline to VLO/PSX: they "almost are beginning to look like the Micron type of memory trade at the end of June… everyone's there already… it is extremely bullish at this point." The answer is rotation inside the theme (DVN, FANG), not exit. Talkington's structural leg: E&Ps "really aren't drilling that much since this war started," the SPR is at an all-time low and has to be refilled, and XLE was flat in 2025 so "you had to be there already" — "stay long, because we've got to continue to drill."
• Read the contingency, not the headline (and the second-order competitive risk). Marvell +10% on Alphabet's option to buy a $12.2B stake in a custom-chip deal (2026-aug-19) — but of ~59 million shares, 57 million are contingent on the relationship and on Alphabet actually delivering future purchases: a very large option on execution, plus the credibility that "Marvell needed this… it is restarting the momentum." Broadcom's decline, per Terranova, is not current TPU share loss but the later risk that Alphabet turns to Marvell for silicon purchases once embedded — "so these relationships I think ultimately are going to continue." Simpson takes the customer side and buys it: Alphabet now runs Broadcom, NVIDIA and Marvell, and the spend has become "more disciplined and targeted."
• Slow-walk the exit, and say so. Simpson on IBM (2026-aug-19): selling began July 14 after earnings "so disappointing I couldn't believe it as a shareholder" and a ~25% one-day drop — but he didn't dump into the gap, he sold into the recovery. Asked whether more is coming: "absolutely, because we have 1% left of IBM… I do everything at a glacier pace." The proceeds funded Alphabet, the residual bull case is named and specific (quantum, "probably the long-term holy grail"), and the reason to go now is opportunity cost plus a software overhang he says he worries about less than most.
• Tom Lee's core list: two adds unopposed, one avoid overruled three ways. Adds JPM ("best of breed financial services… the awakening of IPOs is not over") and Arista (Terranova core holding, "participating in AI networking along with Cisco and Ciena… revenue growth is accelerating" — the piece of the AI complex he is not trimming). But his Robinhood avoid draws rebuttals from opposite premises (2026-aug-19): Simpson (long) says it has outgrown the bitcoin-proxy label — revenue +32% to a record $1.13B, EPS +48%, a record $22B of net assets, an RIA acquisition and a custodial platform — "the market correlates it with Bitcoin… absolutely, and incorrectly"; Talkington (called away at 110) agrees with the correlation and is bullish anyway because "crypto's finally waking up"; only Terranova is out on process — a momentum breakdown — and prefers IBKR.
• Options tape as the macro read-through: gold, bonds and bitcoin off one announcement. Renick (2026-aug-19): GLD calls outpacing puts ~3:1 on 2½× average volume (most-bought the 420 call expiring Oct 16, needing +5%); TLT call buying double puts; and IBIT with almost 5× as many calls as puts on 6× average volume on favourable regulatory news — "arguably the most bullish in a long time." Wells Fargo reiterated $4,900 gold for 2026; Liz Thomas re-underwrote the metal on renewed central-bank reserve demand (the spring selloff was banks raising cash against high oil) and sees 5,000 in this cycle; Simpson owns the miner instead — AEM, up ~50% in a month yet still −9% over six. And into Walmart's print, two-thirds of ~$60M of premium sits on the put side against a 4.6% implied move — which Terranova ignores on the record (revenue beats every quarter since the pandemic, plus tariff refunds this quarter and next).
• The bear case, priced: 5–10% and a re-entry problem. Terranova's decisive framing (2026-aug-11) — grant the bears a hot CPI and a hike priced at 75–80%, and the payoff is "maybe 5 to 10% down, and then you have to be fast enough to get back in, because the dip buyers will return." Real bearishness needs a setup where the market falls 20%+ and stays there, "and I don't see any fundamental evidence for that." Sechan's supporting data: revenue growth the fastest since 2021, a record 88% beat rate, the P/E down, and institutional positioning in the 37th percentile (retail somewhat more extended) — light positioning is fuel, not risk.
• Earnings quality survives the "bubble" test. Brown's numbers (2026-aug-06 and -11): 28.3% growth ex-tech; net margin revised up mid-season to 15.6%; sales +15.2%, 300bp above the two-month-old expectation; 10 of 11 sectors growing profits; and the statistic that does the work — analysts have raised first-month-of-quarter estimates by +0.1% versus a −1.9% average across 80 quarters. "You can fake earnings, you can't fake revenue." On valuation he refuses the historical anchor: "is 20 times cheap? No. But why would the multiple on this crop of companies be 16 times because it was in 1994?"
• The counterweights worth tracking. Re-concentration — Terranova's own invalidator: the cap-weighted S&P beat the equal-weight by 4% in days, and the narrative dies if momentum piles back into NVDA/MU/MRVL/GLW (he now watches the equal weight and the dollar; a new equal-weight high on Aug 7 validated the bull case). Pull-forward — Ethridge: the four hyperscalers pre-funded next year's purchases into this year to lock in prices, so third-order beneficiaries (CAT, GLW) may air-pocket in 2027; "I think that probably is through the end of this year." The 2000 rhyme — Krinsky: the largest four-day SOX rally since COVID and an S&P +5% in four sessions into a 52-week high, only three precedents in 30 years including March 21, 2000; plus "musical chairs" as money sloshes momentum→value→momentum. Instant re-risking — Santoli: the five-day put/call is rock bottom, though institutions still have room.
• The NVIDIA $500B consortium is the trade of the period. Confirmed 2026-aug-11 with MOUs from Goldman, Apollo, Blackstone, Brookfield, KKR and BlackRock to raise third-party capital so NVIDIA's customers can finance compute. Brown: the world is "structurally short compute… for at least the next five years," so socialize the risk and build "a bridge between what Wall Street is looking for — more stuff to invest in with a yield — and what Silicon Valley needs," instead of "having Amazon and Google do a debt offering every month." Sechan: NVIDIA retains only 25%, which de-risks the name and declares compute an investable infrastructure asset. Terranova: GPU pricing risk is off the table, the compute is the collateral, and within five years compute becomes one of the world's largest futures markets alongside oil (Wapner's caveat: a futures market implies commoditization). It also answers the Cisco-1999 circularity charge Brown had already dismissed with the Jenga test — "if you believe in SK Hynix and Samsung, you have to believe in NVIDIA financing its customers."
• NVIDIA gets bought twice in one show, on two different arguments (2026-aug-07). Simpson at 218 — "this is the OG… not a trade, an investment" — deliberately ahead of the Aug 26 print because NVIDIA reports last and the read-throughs are already public. Link for the first time ever, purely on valuation: 18×, cheapest since 2019, "cheaper than Colgate or Hershey" against 85% revenue growth, mid-70s gross margins and free cash flow doubling. She frames it as a laggard-catch-up pair with Broadcom ($100B AI pipeline, $20/share earnings power each) and adds Micron for its "$100 billion in bookings… take-or-pay contracts."
• SpaceX finds its price the hard way. First earnings report (2026-aug-05): revenue +92% to $7.8B, AI revenue +247%, but $18.4B of capex and the stock −9%, ~50% off the highs — Simpson splits "enthusiasm for the business" from "enthusiasm for the stock" and won't add through the lock-up. Lock-up day (2026-aug-06): available shares +140% to 1.5bn, float 5%→12%, 50% free float by January, against a September Nasdaq-100 float adjustment that forces index buying; plus Terrafab with Tesla ($16.8B, "in excess of 1 terawatt of compute"). Baruch owns and would "put it away"; Brown calls it the too-hard pile ($86B raised, a fifth spent in a quarter); Ethridge wants it under 100. Then Link buys more (2026-aug-07) on the AI payback being "within a year" and $14B of cloud-service agreements in the quarter; Sechan (locked up, from the private side) wants "Starlink growth, higher connectivity margins and disciplined CapEx commentary."
• Alphabet's brain drain, and Apple's first six sell ratings since 2020. Alphabet fell ~5% on the DeepMind reorg and Jeff Dean's exit after 27 years (2026-aug-05) and the desk was unanimous it was an overreaction — 82% cloud growth, an internal successor, chips/search/YouTube/Waymo — then sat "in the penalty box" on capital raising (2026-aug-11) while Belsky calls it "one of the five stocks in the US you should absolutely, positively own." Apple: Jefferies' underperform, then six sell ratings and only 56% buys — and Belsky ("largest position… you absolutely cannot bet against Apple"), Terranova (sideways-to-lower near term, long toward 400) and Sechan (trimmed the oversized weight, "still very long") all stay long.
• Best-stocks work and the sell discipline. Brown's spotlights: Visa (+3,250% since IPO, golden cross, "the path to 400 is clear and wide open") and Expedia — "a narrative violation: people threw it in with the SaaS-pocalypse names, but it's not SaaS, it's travel," with a 20th straight double-digit B2B quarter, raised guidance and $900M of buybacks; traders' pivot 290, investors 260. The exits are as instructive: Block flagged below 80 ("nobody really trusts Jack Dorsey for long"), Spotify sold on a new structural cost line (policing AI slop) at 34× versus Netflix below the market, Fiserv exited at 7× on lost management credibility ("one, maybe two chances to guide down"), Qualcomm swapped one-for-one into Truist for dividend growth, and Terranova's ETF dropping Expedia (500% debt-to-equity), Royal Caribbean and Las Vegas Sands on rules — a rebalance he then audits on air.
• Where else the money is going. Financials on an eleventh straight weekly gain, the longest since at least 1989 — Brown likes everything except payments (PayPal's consolidation limbo): brokers (HOOD, IBKR), regionals (CFG, FITB/Comerica), money-centers (JPM, MS), insurers (BRK.B, TRV, CB, ALL, MET as AI adopters widening margins), and Schwab at 14.5×. Healthcare as the quiet rotation: Amgen doubled on Maritide optionality, Bristol-Myers at 10× with $11–15B of free cash flow to "buy its way to growth," Lilly, Natera, and Harrington's whole neglected-cash-flow book. Metals: "all roads lead to copper" — copper at a record, XME's best week since March 2022, Link's copper/aluminium deficit call (Antofagasta, Alcoa), Harrington's Freeport at 18× with a 6% FCF yield, Talkington's INFL basket. Defense: a $1.5T budget, RTX (a $70M stock-replacement options roll) and General Dynamics. Avoid: apparel/retail entirely — ONON's worst day ever, Nike "too big," Under Armour "a failed company," Lululemon rebooting, with off-price strength the tell that branded apparel is being discounted.
• Targets race higher on earnings, cleaner positioning and a retail bid. JPMorgan to 8000, Evercore's 9000 bull case, Pascarello's de-grossed/lower-gross-leverage note; Terranova: "8000 is kind of a foregone conclusion… 10,000 by December of 2027, why not?" Firestone's structural leg: retail is now 25% of the market (5% twenty-five years ago), breadth is genuine (equal-weight +15% vs 13% ytd; healthcare +15% in three months). Amoroso's guard-rail: earnings are not a bubble in aggregate (10–15% growth even off the 2027 peak), and a real-time-adjusted core CPI near 1.6% means the Fed — the only plausible spoiler — probably doesn't spoil it (2026-aug-10).
• The alpha hunt runs on momentum direction, not conviction. Terranova's operating frame: in a good fundamental tape everyone (fundamental and quant) is chasing the same scarce new alpha, so rank names by where momentum is building vs waning — building in NVIDIA (breakout, tracing Apple's earlier pattern) and in software's laggards (Palantir +30%); waning in Apple (holders paring back) and Alphabet (AI brain drain — the only big tech name red in a huge week). When a holding loses momentum, ask the hedge question explicitly and answer it: Alphabet no, Apple buy-the-decline-into-the-moving-averages (2026-aug-10).
• Energy re-add: buy the sector positioning has abandoned. Terranova took JOET to a 10% energy weight (~3× the sector's 3.2% index weight) — LNG, COP, EPD, XOM, KMI, OKE, TRGP, TPL, WMB — because repeated "we've got a deal" headlines pushed longs to the sidelines while no deal arrived; oil back above $80 and energy +3.6% (best sector, heaviest ETF volume) paid it off. Lebenthal's overlay: energy is "analyzable" — own the refining bottleneck (buy a refiner into hurricane season: VLO/PSX/MPC) and the export/pipeline links whose economics survive $70 oil; Exxon's 27%/yr five-year return vs the S&P's 13% makes it "an investment." Amoroso: drawn inventories plus US/Brazil/Guyana supply mean the midstream works even if the Strait reopens; Firestone's dissent — energy is only 3.2% of the index, "it's a good trade" (2026-aug-10).
• Software: the pattern is right, the timing is contested. IGV +8.5% in a week (Atlassian +40%, Palantir +30%, ServiceNow +9.5%, Snowflake +7.5%), Krinsky "reports of software's death have been greatly exaggerated," Goldman "the beginning of a multi-quarter fundamental inflection," and an options tape whose most popular directional trade is selling puts. Terranova won't buy it — imminent Anthropic / Databricks / OpenAI IPOs mean "significant disintermediation," short of the software apocalypse but enough to stay in semis; his exceptions are cyber (CRWD, PANW) and Zoom (the Anthropic relationship). Amoroso takes the other side and is "slowly pivoting to software" (earnings +22%) as the beneficiary when the memory bottleneck resolves — the same logic that makes bottleneck relief bullish for Apple (2026-aug-10).
• Momentum reset ≠ top (the semis whipsaw is structure). Brown: $200B of leveraged 2× single-stock ETFs (≈ $500B notional) amplifies every move regardless of view — so the ±7% days are noise. The SMH RSI cooled to 49 and the average semi's 10-day rate-of-change bottomed at −15%, a healthy reset that hands sidelined buyers an entry. The play: buy the memory/semi weakness on unchanged fundamentals and sell calls into the elevated premium (Lebenthal doubles Micron at $60, ~15–20% of the whole S&P's 2026 earnings growth; Link initiates NVIDIA at 18× — cheapest since 2019, FCF doubling to $206B; Bryn sells calls on both).
• Where does the CapEx accrue? — own the customer, not the spender. The Mag-7's premium over the 493 is a decade-low ~10% because they sink >100% of cash flow into ~$700B ($900B–$1T next year) of CapEx, and the market never premium-pays spenders (building, not harvesting). Brown's answer is Apple (2.5B-device gateway to consumer AI, not playing the CapEx game, ~1% from an ATH, $100B buyback, Sept-1 Ternus keynote + foldable + BYO-LLM Siri); Alphabet has the "double benefit" (spender + Gemini). Lebenthal: the spenders are 3-month-plus buys below the market multiple; Bryn's the skeptic on execution (Microsoft "no output," "Zoom shouldn't exist").
• Own the buildout / the supplier, not the hyperscalers. Microsoft/Meta are penalized for footing the AI-capex bill (Talkington: "own the Qs, not the mega caps" — QQQ +15% YTD on the memory/semi weight). Play AI through the recipients of the spend — semis & equipment (MU, SNDK, ONTO, WDC, AMAT, TER), data-center/power/grid industrials (CAT, PWR, GEV, VRT, VST) — and the second-derivative energy winners (Harrington: CVX powering Microsoft data centers, ENB, the whole oil-and-gas value chain transformed by AI).
• Memory: paradigm shift or hot cyclical? Micron's blowout print is the centerpiece — Terranova "irreplaceable resource" with pricing contracted to 2029 (DRAM ASPs +60%, NAND +80%); Lebenthal cheap-cyclical at ~7× forward + buybacks; Brown's caveat — it's all on price, not volume, and Apple is the casualty (telco-subsidy / Starlink risk too).
• Quality financials over private credit. Capital rotates to the money-center banks (GS — Weiss's largest — JPM, BofA, both raising dividends after the stress test; regional breakout KRE/FITB/CFG/TFC, loan growth +8%) and away from the alt-credit managers (Apollo, Ares, Morgan Stanley) hit by redemption headlines — though the BDCs (OTF, ARCC, ~13% yield) stay "money-good."
• Where else the money goes: value/income — Fiserv (16% FCF yield), Stanley Black & Decker, the capital-starved REIT "halo trade" (MRP), defensive UnitedHealth. Caterpillar is the call of the day (Baird PT $1,200) but split — Weiss trims at 36× ("closer to the 'isn't'").
• Macro: oil back above $80 with the Strait still shut and "no deal" repricing energy higher (Terranova: the market can live with $90 as long as earnings grow; his own downside is 65 if it opens); the 30-year at 5.22% into a Wednesday CPI, Jackson Hole and a September meeting the market still gives some odds of a hike; a record-breaking week for call-option buying and Santoli's constraint — "yields up, oil up means the market is not going to be very broad" (Nasdaq 100 beat the equal-weight S&P by 7 points in seven sessions). Earlier: oil ~$69 and a 100% EU digital-tax tariff threat; gold exited (Weiss — "didn't do what it was supposed to"); bitcoin sub-60k and MSTR breaking down (Jun 25).

Transcripts

One dated page per episode or interview — each has its talking points and the saved transcript. Newest first.

To process — backlog

CNBC episodes and interviews not yet processed, newest first. None queued yet.

For personal study — not investment advice. Source material © CNBC.