Newsletter · · Ashutosh Agarwal

Kaspi Micron and the AI Memory Boom Plus Short Calls on Oracle and Caterpillar - Weekly Podcast Idea Digest - Week of July 13, 2026

Cross-sector best-ideas digest for the week of July 13, 2026, drawn from episodes aired July 6 to 13. A high-conviction Kaspi long, a bull-versus-bear Micron memory-cycle debate, short calls on Robinhood, Oracle, and Caterpillar, and single-name ideas across biotech, energy, and small caps.

Weekly Podcast Idea Digest

Week of July 13, 2026: Kaspi Micron and the AI Memory Boom Plus Short Calls on Oracle and Caterpillar

This was a loud week, not a quiet one. Across the podcast library, at least fifteen episodes contained a real, specific stock idea, a named company plus an actual argument for why to own it or bet against it. No formal idea-conference pitches (Sohn, the Robin Hood conference, Delivering Alpha, MOI, ValueX, the Pershing Square Challenge) turned up in the seven-day window; everything below came from fund-manager and analyst interviews, plus a couple of value- and biotech-focused shows. A running theme tied many of them together: is the AI/memory boom being priced for perfection, and where do you hide (or short) if it isn't?

Below are the ideas worth your time, organized by conviction and type. I've kept the numbers and the reasoning intact so you can judge each one yourself.

The high-conviction long: Kaspi ($KSPI)

Two episodes in the We Study Billionaires network, The Investor's Podcast ("TIP829," July 9) and The Intrinsic Value Podcast ("TIVP082," July 12), both with Daniel Mahncke and Shawn O'Malley, did a genuinely deep two-part dive on Kaspi, the "super app" that dominates Kazakhstan. This was the most thorough single-company work of the week.

The pitch, in plain terms: Kaspi is one company that plays the role of Visa, Amazon, and a consumer bank all at once, in a country of 20 million people where more than 70% of the population uses it actively, an astonishing 77 times a month per user (about 2.5 times a day, versus roughly 25 times a month that the average American taps a debit card). Three businesses:

  • Payments, only 16% of revenue but about 40% of net income, at a net margin above 65% (higher, they note, than Visa's). The reason it beats Visa: Kaspi owns the whole payment rail, so when you scan a QR code the money goes straight from your Kaspi account to the merchant's Kaspi account, with no Visa/Mastercard "middleman" taking an interchange fee. Every extra transaction is almost pure profit. A new palm-scan product, "Alakran," signed up half a million people in its first 90 days.
  • Fintech/lending, about 38% of revenue, 33% of net income, on roughly $24 billion of loans. Because it's a super app, Kaspi knows your income (your salary lands in the account), your spending, and your repayment history, so it auto-approves 99.9% of loan applications in under six seconds. Bad loans (non-performing loans, or "NPLs") sit at just 6%, the same level as Nubank and about 10 points below MercadoLibre, even while the loan book grows 20%+ a year. Funded cheaply by ~$14 billion of deposits from 6 million customers.
  • Marketplace, take rate of 12% (16% with delivery and ads). The high-margin advertising piece is growing 70%+ a year with only ~7% of merchants using it yet, so there's a long runway.

In total: ~18 million transactions a day, 14.5 million payment users, 750,000 merchants. Mohnish Pabrai is a holder ("heads I win, tails I don't lose much," where Kazakhstan is the "tails").

Valuation: at roughly a $14 billion market cap, their discounted-cash-flow model assumes only ~11% revenue growth (well below the ~30% historical average), flat margins, and no multiple expansion, and still gets to a 20%+ expected annual return over five years, including an ~8% dividend yield. As Shawn put it, "to get this company with the kind of quality and growth prospects that they have at this valuation does almost feel too good to be true."

Honest caveats they raised, and why this is a watch-list name rather than a table-pounder: the whole business runs on the Kazakhstani currency, and emerging-market currencies "not uncommonly get cut in half in USD terms over a decade" (it's already happened once). There's political risk. And the recent $1.1 billion cash purchase of a 65% stake in Turkey's Hepsiburada opens a debate, is Turkey a cheap "call option," or a sign the home market is maturing? Turkey's main rival, Trendyol, is backed by Alibaba and will force aggressive spending. Both hosts said they'd keep Kaspi "high on the watch list," possibly take a starter position, and want to talk to management first, a measured, not breathless, endorsement.

The short calls

Robinhood ($HOOD), "a phenomenal business at a dangerous price"

Deep Values (July 9) ran Robinhood through a strict deep-value screen at its $109 price (~$98 billion market cap). It fails almost every traditional test: price-to-earnings of 51.88 (value investors want under 15), a PEG ratio of 3.46 (want under 1), price-to-book of 10.14 (want ~1.5). The business itself is excellent, 94% gross margins, a sticky ecosystem (gold subscription, credit card, 5% cash yield, IRA match) that "weaponizes your own convenience against you," like Apple's.

The heart of the bear case is a discounted-cash-flow reality check on ~$1.9 billion of trailing owner earnings:

  • No growth forever implies worth ~$21 a share
  • 5% growth implies ~$30
  • Even a very aggressive 15% growth for five years implies $50.54

To justify $109, the market needs Robinhood to grow free cash flow 25%+ for a decade with flawless execution on Europe, prediction markets, and its credit card, "zero room for error." A telling detail: in March 2026 the board authorized a $1.5 billion buyback right after buying back $1.2 billion at an average of $46 (and the 2022 low was $8.14, below tangible book). Management is "managing for scale, not value." And founders Vlad Tenev and Baiju Bhatt hold ~60% of voting control despite a minority economic stake, so activists can't force change if it breaks. The takeaway: great company, no margin of safety. (Worth noting: Deep Values is an AI-narrated deep-value screen show, not a named fund manager, so treat it as a framework, not a PM's book.)

Oracle ($ORCL), the AI-capex hangover

On DHUnplugged (#809, July 8), money manager Andrew Horowitz laid out a short on Oracle. The stock fell 19% in a single week, its steepest drop since August 2001, the depths of the dot-com bust. His case: capital expenditures surged 162% in the latest fiscal year, the company has $24 billion in negative free cash flow and $130 billion in debt, and it's down about 55% from its high. Oracle "built stuff and gathered all this in anticipation of others coming to the rescue by funding them," a reference to the OpenAI-linked deal that once sent the stock "haywire," while leaning on lower-margin offerings. "Its knickers are in a bind." He added Oracle as a short (alongside a Tesla short he admitted was "thin, momentum"). Same episode flagged a report that NVIDIA's next architecture ("Kyber"/Rubin) could slip 12 months into 2028; NVIDIA pushed back and Mizuho called it "noise."

Caterpillar ($CAT), Michael Burry's target

Smart Investing with Brent & Chase Wilsey (July 11) relayed that Michael Burry (of "The Big Short") has put on a bearish options position against Caterpillar, notable because "he doesn't short things." The reason: valuation. At $952.41, CAT trades at a forward P/E over 30x and a trailing P/E of 47.7 versus an industry 23.4; price-to-sales 6.4 (industry 0.8), price-to-book 23.6 (3.5). The stock is up 66% year-to-date. The business is genuinely strong, 13.3% net margin, 50.5% return on equity, but earnings per share grew only 2.2% last year (industry +8.7%). The hosts' own math: Dec-2027 estimated EPS of $30.41 implies a fair sell price of $504.81. "You're getting a tech multiple on a company that generally does not have a tech multiple."

The big debates: Meta, Apple, and SpaceX

Meta ($META), "In Zuck we trust?"

Pitch The PM (Ep. 39, July 9) featured a sharp bull-versus-bear back-and-forth with Avory & Co. CIO Sean Emory. His bull case: Meta is the second-fastest-growing and highest-margin company in the Mag 7, yet "somehow doesn't get the love." It owns both its 3.6 billion-user distribution and its infrastructure, creating a flywheel; ad revenue is growing 33%; the Meta AI app now ranks in the top 10, above Gmail. He trimmed the position from ~20% of his portfolio to ~5% around $800, but it's "getting more and more interesting."

The tension: the stock now has "90% correlation to 2027 free-cash-flow revisions," which have gone to roughly zero (or slightly negative) because Zuckerberg is spending $60–80 billion a year on AI, "bigger than the metaverse." The host (a former hedge-fund PM) pushed hard: why own the low-multiple infrastructure layer instead of divesting it via a sale-leaseback? Emory: you need it now while compute is scarce, and you can spin it off later. He models capex plateauing around $165 billion in 1.5–2 years, and his contrarian call is that 2027 is the peak for hyperscaler capex. Alternative-data (Oxford) shows Meta's next-quarter growth exiting at 33–34% monthly versus consensus of 27%, "the fundamentals are good and the stock's on its ass. You've got a good setup."

Apple ($AAPL), the "toll booth" thesis

The Compound and Friends (July 7) made an extended bull case. The latest quarter: EPS $2.01 (vs $1.95 expected), iPhone revenue $57 billion (the iPhone 17 was a hit), Services at a record $31 billion, and a Greater China rebound to $20.5 billion (helped by an Alibaba deal). Guidance was raised to 14–17% growth versus a Street estimate of 9%, pushing the stock above $300.

The core bull argument, echoing Dan Ives' $400 target (a sum-of-the-parts call): Apple doesn't need to build the best AI model, it just needs to be "the toll booth." With a "bring-your-own-LLM" approach and iOS 27 letting users pick a default AI model, Apple monetizes everyone else's models across 2.5 billion devices, a $15 billion annual AI-services opportunity that Ives thinks adds $75–100 per share. The near-term risk: gross margin, because memory (DRAM/NAND) prices jumped 98% in a single quarter, and Tim Cook called it a "100-year flood" and said price increases are "unavoidable" (the iPhone 18 Pro could be $200 more). September brings two catalysts: new CEO John Ternus takes over Sept 1, and a foldable "iPhone Fold" launches at a rumored $2,000–2,500. The pushback from a co-host: it trades at ~33x forward earnings, and "only an AI contract moves it, and I'm skeptical Siri works." Growing 15% revenue / 17% EPS "at this size, no one but NVIDIA is doing that."

SpaceX (private), Morgan Stanley initiates

On Squawk on the Street (July 8), Morgan Stanley's Adam Jonas discussed his new overweight rating on SpaceX with a $300 base case and $600 bull case (the stock is around $160, down ~7% on the week). He calls it "fairly simple to model": two proven businesses, launch (cost of ~$1,000 per kilogram, about 20x cheaper than the $20,000 industry average, with a Starship goal of $100–200) and Starlink (a "50% margin cash-flow machine" modeled on subscribers × revenue-per-user), plus optionality in orbital/AI compute (a "revenue-per-watt" neo-cloud model). The risks are real: SpaceX may need roughly $700 billion of external funding over time ($84 billion a year), won't be free-cash-flow positive for a decade, carries heavy "key-man" (Elon Musk) risk, and faces launch anomalies, "there will very likely be an explosion and the stock will be down significantly." A $300 valuation would put it in an "NVIDIA/Broadcom bucket." It's private, so not directly investable for most, but the framework is useful.

Separately, on Thoughtful Money (July 7), investor Eric Jackson said SpaceX at ~$150 is "optimistic but not ultra-bullish," with maybe $175–200 upside over two years, "that doesn't get my juices flowing," and he'd "never short a Musk company."

Two more single-name longs

Canada Goose ($GOOS), a turnaround inflecting

Tom Hayes (Hedge Fund Tips, Ep. 351, July 9) updated his long thesis on Canada Goose (a small-cap; the company reports in Canadian dollars). Fiscal 2026 (ended ~March): Q4 revenue up 17.9% year-on-year to $453 million; full-year revenue $1.53 billion, up 13% and over $1.5 billion for the first time ever. Direct-to-consumer comparable sales accelerated to +10% (a fifth straight positive quarter).

The thesis has two legs. First, the shift to direct sales: DTC is now 76% of revenue (up from 29% in FY2017), heading to a target of 80% by FY28, at mid-70s gross margins versus mid-to-high-40s for wholesale, and full-year gross margin hit 69.7%. Second, the margin recovery: adjusted operating (EBIT) margin peaked at 24.9% in FY2019, bottomed at 9.7% in FY2026 (dragged down by $67 million of one-offs: a $43 million arbitration payment, a $15 million bad-debt provision, an $8.4 million store impairment). Management guides FY2027 to 11–12%, and Hayes sees a path back toward the mid-teens over several years, which "would unlock significant earnings power and a meaningful re-rating." The company has also evolved from a winter-parka business (down outerwear was 95% of revenue 15 years ago, now about half) into a "365-day lifestyle brand." His stance on a skeptical market: "We are happy to take the other side."

ACM Research ($ACMR), a China semiconductor play, US-listed

On Barron's Live (July 6), asset manager Tiffany Hsiao offered ACM Research as her favorite way to play China's forced push into homegrown chip equipment. It's a US-listed, US-origin company with a Chinese founder that started in semiconductor cleaning equipment, where demand rises structurally as chips get more advanced (cleaning intensity goes up), and has expanded into higher-value manufacturing steps. Crucially, it runs a separate subsidiary serving Chinese customers, so US investors get real exposure to China's semiconductor boom (driven by Western export bans) that they can't get from a typical US equipment maker. The China portion is "growing exceptionally well" while it keeps diversified customers outside China. Her caveat: China moves at "warp speed," so single-name China positions are more tactical than buy-and-hold.

The memory-cycle debate: Micron ($MU)

This was the week's cleanest bull-versus-bear pairing.

Bull: Adam Parker (MacroVoices #540, July 9) ran "10,000 simulations" on the memory cycle and concluded Micron trades at roughly 4–5x peak earnings and 10x normalized earnings, "a little too cheap," with "a lot of upside to base and peak earnings over the next six months." He argues investors are wrong to underweight the most volatile AI cluster. (He also likes healthcare, where the market implies a ~0% chance it's the best sector over five years and he says 30–40%, plus energy and utilities as AI-adjacent diversifiers.)

Bear: Tom Hayes (same Hedge Fund Tips episode, July 9) warned that people are "getting faked out" by cheap-looking memory multiples on peak earnings. His analog is the 2000 semiconductor peak: earnings and estimates almost doubled from 2000 to 2001, yet the stocks rolled over because the market looked two years ahead. "When they normalize, you go from a 6 multiple to a 600 multiple in 12 months." Smart Investing (July 11) made the same point about SK Hynix accelerating its expansion "by more than a decade," classic late-cycle capacity building.

Two credible investors, opposite conclusions on the same stock, a good one to form your own view on.

Biotech corner

Two specialists were active as the sector runs hot (three $10 billion-plus buyouts in a single month: Nuvalent, Apogee, and Crinetics).

Value Hive ("Seedy19," July 10), a biotech trader who frames every idea around a built-in M&A exit (typical takeouts at 2–3x sales, 3–12 months after approval):

  • Vera Therapeutics ($VERA), an IgA-nephropathy (kidney) drug that just won FDA approval into a large market alongside Otsuka's successful drug; >$1 billion peak-sales potential; he expects an acquisition within 6–9 months. Trading in the $40s.
  • Erasca ($ERAS), a RAS/PanRAS cancer drug similar to Revolution Medicine's (RevMed is worth ~$40 billion with zero revenue). Erasca trades at roughly a quarter of that; the drug was licensed from a Chinese company and RevMed is suing for patent infringement, but he says lawyers view the litigation risk as "overblown." He sees "50% to 100% left in the valuation."

Investing Experts (Jonathan Faison of ROTY, July 8), focused on late-stage/commercial names with "high strategic value to acquirers":

  • Journey Medical ($JRNY), ~$200 million enterprise value, with rosacea drug Imrosi (patent-protected to 2039 and showing head-to-head superiority). Conservative $200 million peak sales would put it at ~1x EV, his sweet spot.
  • Syndax ($SNDX), bought around $9; already has two approved drugs (Revuforge in leukemia, Niktimvo in graft-versus-host disease), with a 50/50 US split on Niktimvo with Incyte that could make Syndax a buyout target.

Small-cap and contrarian picks

Eric Jackson (Thoughtful Money, July 7), who prefers beaten-down small caps over mega-caps:

  • Dave Inc., a cash-advance fintech he bought at $165 in April, now around $400. Average loan is ~$200 for eight days; it takes essentially no credit risk (a bank partner holds the loans) and has a "microscopic" default rate. He thinks it's "just scratching the surface" of its market.
  • Nextdoor, the verified-neighbor social app, trading at ~2x price-to-revenue with a cash-rich balance sheet. His AI angle: its bot-free, verified local data could power a local-services matching business; if it re-rates toward Reddit's ~16x revenue multiple as founder Nirav Tolia turns it around, the upside is large.

Energy, Adam Rozencwajg (Goehring & Rozencwajg, The Trevor Rose Podcast, July 9): US natural gas is "the cheapest molecule of energy on planet Earth," trading at one-sixth to one-seventh of global prices, "gas can go from 2 to 20 on a spike." His largest position is Range Resources, plus Tourmaline and EQT, waiting for US gas supply growth to crack (the aging Permian's "gas kick" is masking it) just as AI data centers and LNG exports surge. On oil, he owns Canadian Natural Resources ($CNQ) (bought around $40) and recently added Suncor, long-life, depletion-resistant oil-sands assets in a friendly jurisdiction, with optionality to the $100–120 oil he thinks is needed mid-cycle. Size it to your volatility tolerance, he cautions, "it's volatile as hell."

Quick hits

  • Money Life lightning round (Tom Martin, GlobalT, July 9), buys across the AI-infrastructure build-out and quality compounders: Quanta Services ($PWR) (data-center electrical contractor), Palantir ($PLTR) ("ontology" plus forward-deployed engineers), Visa ($V) ("a moat not readily replicable," on sale), GE Aerospace ($GE) (engine backlog into the next decade), Welltower ($WELL) (senior housing riding an aging population), and TKO Group ($TKO) (combat-sports media rights).
  • Money of Mine (Matt Griffin, July 10), Australian small-cap resources (ASX-listed, prices in Australian dollars): Catalyst Metals (expanding from one gold mine to five toward ~200,000 ounces, "the cheapest producer out there"), Bellevue Gold (hedge book repaid in ~6 months, a proven re-rating catalyst), Leverup (lithium; the Sayona/Piedmont merger, now cash-flow positive at ~half the multiple of PLS), and GR Engineering (a contractor with $1.1 billion of awards in three months and a possible 7–8% FY27 dividend yield).
  • Palisades Gold Radio (Dr. Nomi Prins, July 9), silver miners AYA Gold & Silver and First Majestic, uranium name UEC, and Colombian oil producer Ecopetrol.
  • Money Rehab (July 6), a Hightower strategist's cautious take on NVIDIA: at only ~14x forward earnings it's "done nothing in six months" while AMD, Broadcom, and Marvell outperformed, and it faces custom-silicon competition from Amazon and Alphabet, buyable for the long term, but "temper your expectations."