Digging Deeper: Conversations in Fundamental Investing

Summary

• David Rosen’s (Rubric Capital) core setup: with volume flowing to quants, macro funds, and pods that all need momentum, a falling stock has “literally no buyers” — “I can go out and be 50% of the volume, and the stock will still go down 10%.” His pick: Viatris (VTRS) at 7x current / 6x forward earnings with 100% cash conversion, “more than a double the next 12 months” on pipeline readouts in 6-9 months — and the derivative trade, Idorsia (IDIA SW), a potential 10-bagger from CHF 4 to 42 if QUVIVIQ descheduling, the wakefulness label, and the ADHD/autism pediatric data all hit. Rubric bought 10% of the company in three weeks.
• Andrew Ballou’s long is Carvana: biggest and most profitable player in used cars yet only 1% share of a $1T, 40,000-player market — “incredibly rare to find in a single company.” Word-of-mouth referrals on a 7-year purchase cycle mean 2026 revenue was seeded in 2021-23, making 30%+ growth unusually projectable; at ~20x pre-tax next-year free cash flow, “I don’t know if it’s a 10-bagger in 2 years, but it could be in 5 or 10.”
• Eduardo Marques (Pretensul Partners): with ~23% of the S&P now semis and optical, “it’s hard for you not to treat the AI opportunity as a cyclical” — so he’s mainlining “the crack cocaine of Korean value investing”: SK Square as SK Hynix at a 47% discount, and Samsung Life as the Lee family’s stake in Samsung Electronics at 0.48x book and ~4x earnings. Catalysts: foreign inflows reversing a decade of Korean retail buying Mag7, the Value Up program mirroring Japan, and Interactive Brokers opening Korea to US retail just last month.
• John Lennon’s (Pleasant Lake) contrarian long is Reddit, down ~50% and mispriced as an AI loser: engagement still growing (24B→25B posts/comments), Google/OpenAI licensing deals repricing “at like 5 or 6x” in the back half, and the Anthropic lawsuit “either great for Reddit or amazing for Reddit.” At 14x his next-year number (~10x 2028) with Meta’s ~70 RPU versus Reddit’s low 20s, he sees a double-to-triple if it’s rebasketed as an AI winner.
• The panel’s sharpest short thesis, from John Lennon: the best shorts are now high-quality compounders facing “service-level deflation” invisible in near-term earnings — starting with Australian classifieds firm REA, whose push-price-5%-forever model breaks once agentic AI interjects in the purchase. Already, “if you run a search today in Manhattan for a three-bedroom apartment… on OpenAI, the results are good enough” and listings come direct from Compass, not Zillow.
• The inverted AI-winner call: “The true AI winner will be a mediocre business that will just be able to expand a low EBITDA margin by two percentage points.” John’s math — 50% margins going to 52% is +4%, but 50bps going to 2.5% is +400% — is driving Pleasant Lake’s public-to-private deals in “ostensible shitco retailers” at “20 to 50 cent dollars.”
• The moderator’s market-structure observation doubles as an opportunity map: after the Iran-war oil spike, consumer companies beat and guided well and “none of them went up” — everything non-AI is a source of funds, leaving a de-risked universe where significant capital can be deployed.

Deep dive

1. Deep value with no bid: the Viatris-to-Idorsia 10-bagger

• Rosen’s opening frame — “it is a lonely thing to be a deep value guy” — is also his edge: momentum-dependent quants, macro funds, and pods dominate volume, so downside volatility finds “literally no buyers.” “I can buy all the stock all the time.”
• The anchor position is Viatris (VTRS): 35% generics, 65% branded generics and specialty pharma, low-to-mid-single-digit growth, 7x current / 6x forward earnings, 100% converted to cash, half returned via dividends and buybacks, half spent on M&A — including a pipeline bought from liquidity-stressed Idorsia: selatogrel, “literally like an EpiPen for somebody who has a heart attack,” and a lupus drug (heard as “cenegermod,” likely cenerimod). Both prospectively blockbusters, reading out in 6-9 months — why Rubric thinks VTRS more than doubles in 12 months.
• Digging into the seller produced the bigger trade. Idorsia’s QUVIVIQ, a non-drowsy orexin-antagonist insomnia drug, “sells like hotcakes in China and in Europe” but was scheduled at its 2023 US approval because the FDA didn’t know if the novel class was addictive. Scheduling creates two big barriers to blockbuster potential: doctors avoid any abuse-risk drug, and refills require in-person visits. With hundreds of thousands of patients of clean data, Idorsia is now seeking descheduling.
• The re-rating catalyst: Contessa (likely Centessa) sold for $6.3B on the thesis that orexin dysregulation maps to ADHD and autism — and Idorsia has already said its pediatric ADHD/autism trial hit significance, with clean abuse data filed at the FDA. Rosen’s sum: descheduling plus a wakefulness label adds $22 a share, or 26 bucks, and the autism/ADHD/peds data plus SPV assets another $16 — CHF 42 versus 4 today, “a 10-bagger if all these things hit.”

2. Carvana: the rare AI beneficiary

• Ballou’s screen is e-commerce post-“AI onslaught”: “some babies have been thrown out with the bathwater,” and companies with proprietary product, proprietary data, and vast physical infrastructure may be AI beneficiaries, not victims.
• Carvana’s rarity is three traits in one company: biggest in the industry, most profitable, and only 1% market share — possible only because the addressable market is a trillion dollars across 40,000 fragmented players.
• The growth is unusually projectable: consumers refer purchases “to 10 other friends,” but the average American buys a used car every 7 years — so 2026 revenue was seeded in 2021-23, and today’s referrals seed 2029-31.
• The structural loan edge: Carvana sells loans to the same investors year after year, incentivizing conservative underwriting versus a salesman who just needs the car sold; and reconditioning at million-car-a-year scale makes each car worth $500-1,000 more to the lender on repossession. North-of-30% growth for a long time at ~20x pre-tax next-year free cash flow.

3. Korea is the value trade: “we treat it as if it were still in the Asia crisis”

• Eduardo’s dilemma frames the panel: reconciling belief in “this new industrial revolution” with margin-of-safety principles is hard when ~23% of the S&P is semiconductor and optical names — “it’s hard for you not to treat the AI opportunity as a cyclical,” and as a cyclical, valuations look full. The known example: Micron at 10x 2026 earnings versus SK Hynix at six-ish, both “massive cash gushers” for at least two or three years.
• Two ways to “abuse” Korean discounts: non-voting preferreds, and holdcos — SK Square is effectively Hynix at a 47% discount, with the holdco promising to close it via buybacks and dividends. “This doesn’t exist elsewhere in the world.”
• The most convoluted and most exciting: Samsung Life, nominally an insurer, actually the Lee family’s vehicle controlling ~10% of Samsung Electronics — a trillion-dollar DRAM/NAND powerhouse at 7x current earnings with a zero-margin foundry that could produce an Intel-like boom if fixed. Mark the stake to market and you’re buying Life at 0.48x book, ~4x earnings on next year’s dividend flow-through.
• Why now, per Eduardo: favorable trade balance, foreign inflows replacing “Korean crazy retail degenerate traders” who’d been shipping money into Mag7, the government’s Value Up program mirroring Japan’s governance reform, and IBKR opening Korea to US retail last month. John’s trip color compounds it: Hynix redistributes 10% of operating profit as employee bonuses and Samsung is matching — a ~$36-37B tailwind, “mid single digits percent of GDP” — and on consensus next year Samsung is the most profitable company in the world, Hynix second, Nvidia third.

4. Reddit, and everything else in the source-of-funds bin

• John’s behavioral test for longs — “the thing we’d be the scaredest to be short” — puts Reddit “at the very top of my list.” Down almost 50% from the highs in the AI-loser factor basket while posts and comments grew from 24 billion in 4Q to 25 billion this quarter.
• The inflection he expects in the back half: Google and OpenAI licensing agreements reprice — his work says “on an apples-to-apples basis at like 5 or 6x” — while the Anthropic lawsuit’s outcome “will be either great for Reddit or amazing for Reddit. You can only choose one.” Monetization gap: Meta’s RPU ~70 versus Reddit’s low 20s; he models 50%+ revenue and 75%+ earnings growth, 14x his next-year number, closer to 10x on 2028 — a double-to-triple if rebasketed as an AI winner. Short interest: high single to low double digits.
• The moderator widens the aperture: after the Iran war spiked oil, consumer-facing companies reported good numbers and good guidance “and despite that, none of them went up” — because capital only wants AI. His conclusion: a whole universe of de-risked, ignored names where significant capital can be deployed.

5. The best shorts are great businesses; the best AI winner is a mediocre one

• John Lennon’s soapbox, taking the short prompt: “it’s easier to spot the losers than the winners” (a line he credits to a Third Point letter), and the danger is “service-level deflation” that isn’t visible in near-term earnings — which is why great companies are printing good numbers while their stocks fall. The perpetuity-value question is spreading from software, BPO, and call centers “to all of the knowledge economy.”
• His specimen short: REA, Australia’s real-estate classifieds firm and the space’s highest-multiple stock. No AI threat today — but two or three years out, agentic AI interjecting in the buying process breaks the model of pushing price 5% a year forever with no audience growth. Proof of concept already: a Manhattan three-bedroom search on OpenAI returns good-enough results, with listings also coming directly from Compass and others, not just Zillow. These compounders face a chasm to value-stock status — “then they’re probably great cash flow yield stories that we should be long, but there’s a big hole.”
• The moderator’s pushback-by-question: early this year the consensus was that asset-heavy businesses win with AI while asset-light gets disintermediated — yet the market has narrowed to semis and neoclouds. “Why is nobody buying the other stuff now?” John’s answer became the panel’s best line: “The true AI winner will be a mediocre business that will just be able to expand a low EBITDA margin by two percentage points that will make a massive boom.”
• John is acting on exactly that math — 50% margins to 52% is a 4% gain, but 50bps to 2.5% is 400% — via public-to-private transitions pairing better management with AI tools, buying “20 to 50 cent dollars.” Not traditional activism, and he owns the loneliness: “buying ostensible shitco retailers is a lot more contrarian… I’m even lonelier than you.”

6. Career advice: from fist fights to “the spirit of abundance”

• John’s confession to the 25-year-olds in the room: in his 20s he found the industry “extremely distasteful” and almost left — “I would rather have had a fist fight in the street with most of the people in our industry than share my investment ideas.” And the jab that got the crowd: “If you don’t think the industry is full of douchebags, then you should probably look in the mirror.”
• The turn came from two pieces of advice: the East Rock crowd’s “spirit of abundance,” and a line from likely Oz (Ballou’s partner, as heard “Os Duwan”): “wait to see what it feels like to make money with your friends.” Ten years later he’s crediting the panel itself — Eduardo “the best short seller I know,” Rosen “on my Mount Rushmore of living investors” — and closing with the lesson: “you can be maniacal and competitive but also have wonderful virtuous cycle relationships, and I wish I knew that earlier in my career.”