Summary

Cliff Sosin designed CAS Investment Partners to maximize long-term compounding, even if that made the firm almost impossible to market. He typically owns four to eight businesses, buys or sells roughly one per year, and exits only for a better opportunity or when reality breaks his mental model. Traditional managers optimize for short, legible feedback loops; Sosin instead asks investors to accept that “Cliff thinks really hard” across noisy three-, five-, or 10-year periods.
His preferred businesses combine multiple durable advantages inside a system whose future remains understandable. A “contained” business rests on mechanisms relatively invariant to social and technological change; software tooling can feel like “building a castle on sand,” while fixed cruise capacity or nicotine-driven brand loyalty is easier to model. The knife-edge is finding something simple enough to understand but difficult enough that the market still misses it.
Carvana’s moat is the accumulated scope of acquisition, logistics, reconditioning, software, lending, registration, and trust—not merely selling cars online. Sosin likens it to starting Amazon while also manufacturing every book, building FedEx, replacing Mastercard, and handling title registration for the buyer’s second-largest purchase. “If anything I’ve said sounds easy, it’s because I haven’t described it right”: Carvana took more than 10 years and roughly $10 billion to build what competitors repeatedly failed to reproduce.
The economic model that was conjecture before the crisis now appears substantially stronger than a traditional dealership. Sosin cites EBITDA margins around 10.5% and rising toward management’s stated 13%-14% range, versus roughly 4.5% for the average dealership; his work suggests Carvana sells comparable cars about $500-$600 cheaper, before somewhat higher financing charges. High-frequency data recently showed 45%-50% year-over-year growth, while selection, density, processing efficiency, and fixed-cost leverage should improve with scale.
The 99% collapse came from several shocks arriving together, not one disproven business model. After selling roughly 425,000 cars in 2021 and preparing to double or more in 2022, Carvana faced immature operations, Omicron-induced logistics gridlock, the sharpest independent-dealer downturn of the period, uniquely irrational auto-loan pricing, early-adopter demand pulled forward into 2021, and an all-debt ADESA acquisition. “It turns out they’re all happening now,” Sosin says of operational problems he had expected to arrive separately.
The strangest external shock was an auto-finance market pricing loans as though interest rates had barely moved. Credit unions adjusted slowly—Navy Federal at one point offered loans below a comparable-duration Treasury—driving industry spreads to their tightest in Sosin’s pre-2008 time series even as other consumer spreads widened. Carvana could not subsidize borrowers indefinitely; Capital One also priced to economic reality and saw auto originations fall 50%.
Sosin held, bought at roughly $80 and again in the mid-$20s, then was legally blocked from buying when his data finally turned green. He had planned to purchase half immediately and half after instrumenting the operational recovery, but a poison pill protecting Carvana’s NOLs prevented holders above 5% from adding. The summer 2023 debt exchange reduced debt and interest expense, yet Sosin calls it “the cherry on top,” not what turned the underlying business.
The lasting lesson is incremental humility, not abandonment of concentration. Sosin now weights management more heavily, applies a harsher base rate to loss-making companies, and wants only businesses stout enough to survive unimaginable combinations of shocks; diversification gets “a teaspoon of medicine, not the whole bottle.” AI already gives a small team extraordinary research leverage, but much of a genuine investing edge still resides outside the public corpus—in proprietary data, former employees, accumulated context, and knowing which narrow lane is actually yours.

Deep dive

1. The fund was built against the incentives of asset management

Sosin entered engineering expecting to become an inventor, then discovered that obsessively debugging physical systems was not how he wanted to spend his life. An internship in private equity revealed that game theory and other academic tools could generate insights practitioners were not using; restructuring at Houlihan Lokey, a year at Silver Point, and five years investing UBS’s proprietary capital followed.
UBS operated like a hedge fund with one LP, emphasizing catalysts, event trading, hedging, and steady short-term profits. Sosin kept asking why a stock with a three beta required shorting three dollars of the S&P for every dollar invested. What began as disagreement became his realization that the institution could not change because a large drawdown might cost the team its capital.
The industry’s real objective, in Sosin’s framing, is not maximizing performance but “maximizing marketability.” Managers need low-noise, short-feedback-loop decisions that signal talent to committees and boards; owning a business for three, five, or 10 years produces an awkward report: “We’re down this quarter because Cliff thinks really hard, apparently.”
CAS launched in 2012 with $5.2 million: $2 million from Sosin, $2 million from his mother, $1 million from a friend, and several smaller checks. It now has roughly 100-200 investors across several vehicles and between $1.5 billion and $2 billion, depending on markets. He credits compounding but also sequencing luck: rearranging the same up and down years might have made fundraising impossible.

2. A small portfolio works only when the business is deeply understood

Sosin generally holds four to eight investments—perhaps once as many as 10—and averages about one purchase or sale per year. The default premise is permanent ownership; he sells when a clearly superior opportunity creates an upgrade or when evidence makes him conclude, “You don’t know which way is up anymore.”
Each thesis is a model of how the company competes inside its ecosystem, and that model must make predictions testable against reality. Profitability over time is roughly “market opportunity times its advantage”; attractive businesses deliver substantial consumer value through several interwoven advantages competitors cannot readily copy.
A contained business depends on a relatively narrow set of forces in an area of life unlikely to transform abruptly. Software tooling may be “building a castle on sand” because nobody knows how software will be written in 10 years; nicotine’s habit formation, secondary reinforcement, and distribution economics are much less dynamic.
Yet containment creates its own valuation problem: anything obviously simple is likely already understood. Sosin describes investing as “dancing on the knife’s edge”—the company must be comprehensible enough for him to model and difficult enough for others to miss. Its relentless accountability also makes investing, in his phrase, “the emperor of intellectual pursuits.”

3. Cournot economics explains why constrained capacity can retain profits

Sosin contrasts two oligopoly structures through cookie sellers at a state fair. If both can manufacture unlimited cookies and choose price, one charges $1.99 against the other’s $2, prompting an undercutting spiral until price equals the $1 cost and neither earns an economic profit: the non-cooperative Bertrand equilibrium.
Fix each seller’s morning inventory, however, and one more cookie adds a unit of profit while lowering the price received on every existing cookie. Each seller internalizes the damage to their own tray but not the competitor’s, behaving like a monopolist facing more elastic demand. That Cournot equilibrium preserves some monopoly profit without explicit cooperation.
Cruise ships fit the model because supply cannot be summoned quickly; operators choose capacity first and then manage yield. Competitor count, demand elasticity, and whether current supply sits above or below equilibrium determine the profit pool, but the central advantage—ships take years to build—is relatively invariant to technological change.
Construction-equipment rental, where Sosin first had major success, offered the same constrained local capacity with highly inelastic demand. Nobody rents a manlift because its weekly price fell $100, and nobody cancels a building because it rose $100. Once the business mechanism is established, valuation is almost anticlimactic: “20 pages about the business and one page on the valuation.”

4. Psychology can reveal brand power before the financials do

Secondary reinforcement links a reward to the surrounding context, with stronger associations when the stimulus is powerful and arrives quickly. Rats can learn to prefer the light paired with cocaine and keep seeking it after cocaine disappears; smokers similarly repeat not only a brand but often the same place and time.
Sosin’s rough packaged-goods margin ladder runs from cigarettes and dip through Coca-Cola, coffee, candy, cookies, savory foods, tomato sauce, bread, and water. The ordering broadly tracks stimulus potency and delivery speed: inhaled nicotine reaches the system rapidly, while caffeine and sugar arrive more slowly through digestion and create weaker contextual associations.
That construct helped Sosin underwrite an early nicotine-vaping investment before the historical data proved its brand loyalty. Friends who invested successfully in Philip Morris International were observing Zyn, and Sosin says the same intellectual construct gave them confidence in its potential brand loyalty before that was obvious in the historical data. But Coca-Cola’s quality is no secret: “All of these tricks, they’re not useful until they are,” typically when uncertainty or fear stops the insight from being priced.
Sosin rejects imposing a manager’s private moral framework on client capital: society’s collective boundary is the law, while probable changes in law or norms belong in risk analysis. He does look for “moral panics” around legal businesses, analogous to economic panics, while warning that backlash can produce damaging regulation. Mere ESG exclusions, however, may not create large mispricing unless aversion becomes genuinely broad.

5. Years of context made Carvana look obvious in a few minutes

Sosin encountered Carvana in 2018 through a pre-IPO video and quickly thought it was an extraordinary, underpriced business—“provided everything they just said is true.” The apparent flash of insight actually rested on years studying CarMax, dealerships, auto lending, Amazon, logistics, manufacturing, and software.
Carvana combined scale and skill advantages from every one of those fields. Purchasing, pricing, physical movement, reconditioning, lending, registration, and software reinforce one another through economies of scope; excellence in a single component is insufficient because a failure anywhere can destroy the transaction’s economics and the customer experience.
His prior obstacle was the conventional belief that consumers would not buy cars online. Carvana’s cohort curves showed that customers clearly loved the experience, turning the encounter into “love at first sight.” The remaining analytical task was understanding how each operational layer could compound the advantage rather than merely add complexity.

6. The physical network turns used cars into routable inventory

A seller can photograph a license plate, answer roughly four questions, receive an offer exercisable for seven days, and either request pickup or visit a hub. Behind that simplicity, Carvana maps plate to VIN and equipment, estimates retail value, transport and reconditioning costs, predicts transaction profit, and calculates the offer most likely to maximize expected economics.
Large inspection and reconditioning centers can process up to roughly 40,000 cars annually and hold 6,000-8,000 vehicles. Smaller local hubs connect customers to those IRCs: single-car haulers perform pickup and delivery, while nine-car carriers move vehicles between hubs and the national backbone.
The hub-and-spoke design collapses thin point-to-point demand onto a limited set of dense routes. Rather than waiting for enough traffic between Fairfield, Connecticut, and Mobile, Alabama, trucks can shuttle continuously between IRCs “almost like train tracks,” making national transport both faster and cheaper than the fragmented historical alternative.
Reconditioning becomes a routed production system: inspection determines the required work, then specialized stations handle tires, oil, cleaning, paintless dent repair, painting, or advanced mechanical jobs. Skill matching and repetition reduce cost and time versus one dealership mechanic switching among tasks; outbound sold cars also balance carriers returning with newly acquired inventory.

7. Digital finance, selection, and trust deepen the operating moat

Carvana lets a buyer search every vehicle by an individualized monthly payment, adjusting term and down payment to the penny rather than displaying an estimate. Doing so requires real-time underwriting across every borrower-car combination and a vertically integrated prime and subprime financing stack that, Sosin says, competitors still have not replicated.
Selection is a major scale economy because the space of makes, models, trims, years, mileage, and locations is enormous. More inventory raises conversion, nearby inventory improves delivery speed, denser logistics lowers unit cost, and accumulated loan performance improves underwriting. Large IRCs also spread overhead while routing and process systems lift throughput.
Trust compounds more slowly than infrastructure. Buying a car sight unseen requires years of strong experiences and word of mouth; loan buyers likewise must trust that Carvana originates to specification and that performance will match representations subject to economic conditions. “You can’t buy that. You have to build it up over time.”
The Amazon comparison is apt but understates the difficulty. Imagine launching the online bookstore while manufacturing the books, building FedEx, replacing Mastercard, and completing title registration—all for the consumer’s second-largest purchase. Sosin says Carvana needed more than 10 years and $10 billion to reach this point; nearly every attempted imitator failed or remains far behind.

8. Sosin changed his mind about how much management matters

Five years earlier, Sosin concentrated on business and price because he doubted his edge in judging executives. Subsequent outcomes changed him: had he ranked company management teams at the outset, that ranking would have “perfectly predicted” performance relative to his expectations. Great management cannot rescue a structurally awful business, but it now receives material weight.
Polished CEO meetings provide limited signal because people who become CEOs usually learn to sound impressive. Sosin instead studies “human capital exhaust”: former employees who stayed five or 10 years reveal what the organization selects and develops. Ten long-tenured Capital One alumni, he argues, will “blow your mind,” saying something important about the institution they left.
His assessment of Ernie Garcia is categorical: “Someday people will compare Jeff Bezos to Ernie Garcia, not the other way around.” Sosin credits Garcia with building a version of an exceptionally hard system that succeeded where nobody else in the world had been able to succeed, and notes that, despite opportunities during the crisis, the Garcias did not disadvantage outside shareholders.
Garcia also reasons about second- and third-order cultural effects. When Sosin asked why Carvana did not price more aggressively against Vroom, Garcia argued that defining success through a competitor would teach employees that Vroom losing meant Carvana winning, displacing attention from customers. That organizational judgment mattered as much as the analytical work on price elasticity.

9. The post-crisis economics validate what spreadsheets once only implied

Before 2022, rapid growth was visible but superior unit economics remained debatable. Sosin says Carvana now produces EBITDA margins around 10.5%, with little stock compensation and relatively modest capital expenditure, and sees no reason to doubt management’s stated path toward roughly 13%-14%. A typical dealership earns about 4.5%.
His team tracks every Carvana listing and compares it with similar CarMax vehicles and market indices. Their conclusion is that Carvana prices cars roughly $500-$600 lower, though financing is somewhat more expensive, while offering broader selection and a better experience. Recent high-frequency data suggested 45%-50% year-over-year unit growth.
The pooled national inventory is already larger than all other dealership inventory available across Connecticut, by Sosin’s estimate, yet coverage of the total vehicle possibility space remains limited. Selection, proximity, brand, and process efficiency can therefore keep improving, while fixed-cost leverage still has room to develop and traditional dealerships face organizational and technological barriers to rebuilding their workflows.
That outcome was hardly visible in 2022. Carvana sold on the order of 425,000 vehicles in 2021, when cars disappeared as soon as they reached the site, and prepared to double or more the following year. It hired aggressively and expanded supply—just as demand, financing conditions, and internal execution all reversed.

10. Growth exposed a company whose operational reach exceeded its grasp

Doubling annually, Sosin described the tenure mix as follows: “less than half of your employees, on average, have been with you for less than a year.” Carvana deliberately favored speed over hardened processes because management believed scale would determine the eventual winner while credible competitors still existed. Tribal knowledge and culture could carry local operations until COVID disruption and sheer size pushed “reach” beyond “grasp.”
Receiving carriers at an IRC illustrates the hidden complexity: workers must unload, locate, sequence, reload, and dispatch thousands of cars while handling staffing peaks and vehicles that will not start. Later software specified staffing, parking, protocols, and even where to keep jump equipment, making the best workers better and everyone else nearly as effective—but nobody had previously written that software.
Another failure emerged after Carvana shifted from mostly auction purchases before 2019 to highly successful consumer sourcing. Net vehicle flow reversed, so some nodes began receiving more cars than they released. Finite parking created congestion, and a logistics network designed around the old direction needed new controls while the company was still growing rapidly.
Omicron turned sequential processes into systemwide traffic jams. One sick driver could strand a truck and nine cars hundreds of miles away; enough interruptions overwhelmed recovery capacity, extended promised delivery times, depressed sales, and caused inventory to accumulate. Resolving the gridlock took roughly three to six months—and obscured that underlying demand was simultaneously falling off a cliff.

11. ADESA debt landed just as independent used-car demand broke

Carvana bought ADESA in February 2022 using debt after discussions that had reportedly lasted years. The auction business may decline over time, but the prize was 54 large, centrally located properties suitable for IRCs and storage—exactly the sort of roughly 200-acre, auto-zoned land that is nearly impossible to secure near major cities.
Sosin considers ADESA a “huge home run” in hindsight because closer IRCs improve delivery speed and labor access, while the auction operation adds other benefits. The problem was timing: Carvana levered up before recognizing the demand collapse, just as capital markets were closing and operational problems were consuming liquidity.
Chip shortages had elevated new and used vehicle prices, making the cost of upgrading greater and reducing consumers’ willingness to swap cars. A market normally around 40-42 million annual used transactions fell from roughly 39 million and change in 2021 to about 36 million for 2022, perhaps touching a 34 million annualized rate; Sosin cautions that his exact figures may be slightly off.
Franchise dealers also received an unusual subsidy from lease returns whose buyout values had been fixed before prices rose. Customers returning rather than exercising their options handed dealers inventory far below wholesale value, pressuring independents such as Carvana. CarMax’s comparable-store sales fell about 20% for all of 2022—worse in duration than its brief near-20% Great Recession decline.

12. Auto lenders created a shock absent from every textbook model

Sosin initially expected rising rates to affect the overall market and vehicle prices somewhat, including the depreciation curve. Because Carvana finances a spread off rates, the absolute level—1% or 4%—should not eventually matter much. “That’s totally correct,” he says; the damaging catch was competitors’ delayed repricing during the transition.
An auto-loan pool’s effective duration, including prepayments and defaults, is roughly two years, making the two-year Treasury a sensible risk-free benchmark. Yet as that yield raced upward in late 2021, many credit unions priced from deposits, Fed funds, or effectively “a napkin,” then updated only through quarterly committees, 25-basis-point increments, and implementation delays approaching 60 days.
Navy Federal at one point offered car loans below a comparable-duration Treasury. By late 2022, industry auto spreads were the tightest in Sosin’s series extending to before the 2008 crisis—even though other consumer-credit spreads were wide and used-car collateral was exceptionally expensive. “This is definitely not in the textbook.”
CarMax could originate at uneconomic spreads and recognize the pain later; Carvana had to price closer to reality because it needed the money, leaving its loans materially more expensive. Capital One made the same rational choice and saw auto originations fall 50%. Sosin expected irrational pricing to last weeks, not roughly nine months, making this his emblem for unforeseeable idiosyncratic risk.

13. Carvana’s early adopters amplified both the boom and the bust

Carvana’s adoption curve posed a puzzle: it could offer most of a region’s relevant inventory, enjoy high awareness, and still hold roughly 1% local share rather than stepping immediately toward its theoretical availability. Sosin suspected word of mouth because buyers said they recommended the service to about four people, but he had not measured whether those recommendations actually caused purchases.
A new survey question found roughly 70% of buyers considered a friend or family recommendation somewhat or very important. Only about one-third purchased without that validation, establishing the viral mechanism and focusing attention on the unusually adventurous people willing to act alone.
Those unaided buyers disproportionately had Robinhood accounts, owned Bitcoin, and shopped for groceries online. Sosin’s conclusion was that early adopters need less social proof; their initial transactions seed recommendations, which gradually widen the addressable customer pool and explain why an apparently superior product ramps rather than instantly takes share.
In 2021, the same cohort may have enjoyed gains from SPACs, cryptocurrencies, or similar assets and pulled car purchases forward. Carvana had only about 1% share, so even 0.3% of market demand disproportionately concentrated among its natural customers could feel enormous. A year later those buyers had suffered losses—and many had already bought their cars.

14. Cutting costs initially made the negative flywheel spin faster

By spring 2022, Carvana faced an industry downturn larger than the Great Recession for independents, the tightest auto-credit spreads in Sosin’s record, exhausted early adopters, system congestion, immature processes, and fresh ADESA debt. Each factor worsened over time, so a diagnosis that looked complete would be invalidated six months later.
Scale normally begets selection, efficiency, trust, and more scale; contraction reverses the mechanism. Carvana slashed advertising and inventory as external demand weakened, reducing demand further and forcing another cut—“chasing a ball down a hill.” From March through November, it became only moderately better at efficiency; after November, Sosin says, it became “really amazing.”
By fall, the company had burned substantial liquidity and lagged its May operating plan. If the following year resembled 2022, Sosin estimated cash might run out in 13-14 months. He still believed unit economics worked and irrational credit pricing would normalize, but “reasonable people could disagree” over whether cost reduction would outrun cash burn.

15. Operations turned before the headline debt rescue

The banking system eventually noticed that rates had risen, easing Carvana’s financing disadvantage, while management accelerated expense reduction. Around January or February 2023, Sosin’s data suggested the company was restraining demand rather than chasing it: unit sales stayed steady even as website delivery queues lengthened.
Manufacturing recovered, vehicle prices ground lower, and industry volumes partially improved, though they remained depressed. Carvana’s cost cuts then translated its underlying economics into visible profitability, resolving the foundational question of whether the online system could actually earn superior margins.
Management also exploited a prisoner’s dilemma among bondholders: Carvana might not repay everyone, but creditors accepting less first could receive safer, higher-priority secured debt. The documents permitted the maneuver, and holders exchanged old bonds at a discount for new secured claims in summer 2023.
Popular retellings credit that transaction with saving Carvana. Sosin disagrees: by then, his operating data already looked strong. The exchange materially reduced debt and interest expense, but it was “the cherry on top,” not the event that turned customer demand, costs, or unit economics.

16. Position sizing and live data shaped every decision on the way down

Sosin could see Omicron disrupting first-quarter logistics, so the initial disappointment looked like a temporary operational hiccup after the stock had already fallen to roughly one-third of its value. He did not add because Carvana remained above his position-size ceiling; when delayed sales failed to recover, the deeper demand problem became apparent.
After Carvana issued equity around $80 and presented its operating plan, Sosin treated it as the fix and bought substantially more. The stock then fell toward $20. After visiting management and rebuilding the underwriting from a blank sheet, he concluded the business should survive, buying more mostly in the mid-$20s.
Recognizing deeper operational work than expected, he chose to buy half then and half after observing a turn. His team expanded web scraping, credit-card matching, logistics tracking, and other instrumentation around leading indicators. Instead, nearly every reading worsened; after Thanksgiving, sales dropped seasonally and then simply failed to return.
A year-end poison pill stopped holders above 5% from buying, protecting net operating losses from IRS ownership-change rules while the market capitalization was exceptionally low. Six to eight weeks later, Sosin finally saw “all green shoots” but could not add. His prior promise to himself was essential: being right would make him successful whether or not he captured the second half.

17. A rational thesis did not prevent a brutal psychological experience

Sosin describes two simultaneous selves: the investor who could calmly explain that abnormal credit pricing would end, and the person awake at 1:00 a.m. while an unkind inner voice berated him. It was the first and only period when someone normally able to govern his internal monologue lost that control.
Explaining the position became nearly impossible. He had expected roughly 800,000 sales, while Carvana tracked toward 300,000; he had expected positive EBITDA, while losses approached $2 billion; the stock was down 99%. Saying “I think things are gonna be okay” after conceding all three made him sound as though he had “lost the plot.”
Partners reasonably grilled him through meeting after meeting, sometimes followed by another 8% decline. One investor instead drove a long distance solely to say, “You’re awesome…we’re Team Cliff.” That gesture still shapes how Sosin treats embattled CEOs: ask the necessary questions, but remember that “even if they’re idiots, they’re trying.”
He repeatedly contextualizes the suffering—he was not terminally ill or “fighting the Japanese in the Pacific”—without minimizing its intensity. The episode combined lost sleep, responsibility for partners’ capital, deep uncertainty, and the burden of defending an apparently absurd conclusion 30 times over. “I’d rather not go through it again.”

18. The postmortem calls for better base rates, not a new identity

Management quality is now a core input because it predicted which companies beat or missed Sosin’s expectations. He also has a “new and deeper appreciation” for how much harder going from unprofitable to profitable is in reality than on paper; loss-making companies remain investable, but his willingness to underwrite that transition has fallen.
The crisis strengthened his preference for wide rather than merely adequate advantages. At about $300 in 2021, he called Carvana unusually stout; three once-in-a-generation curveballs, internal failures, and new debt nearly broke it anyway. Had the mature model supported 5% margins rather than 13%-14%, he doubts it would have possessed enough surplus to survive.
Concentration also gets a marginal adjustment, not repudiation. More diversification could protect against “the Navy Federal Credit Union factor,” especially among less robust holdings, but concentration drove the fund’s long-term success. The proper dose is “a teaspoon of medicine, not the whole bottle.”

19. AI expands the research surface but does not yet reproduce the edge

Sosin uses various AI systems daily, especially where public information is abundant. For Medicaid managed care, an AI can scan government reports, think-tank work, RFPs, and awards to tabulate the last 50 state procurements, identify whether incumbents or entrants won, and summarize the qualitative reasons—work that once consumed enormous time.
Carvana exposes the current boundary: much of Sosin’s knowledge came from former employees, data scraping, and prolonged synthesis, none of which sits cleanly online. Agents may eventually gather more, and the corpus will grow, but today extended questioning of an AI about Carvana “pretty quickly” exhausts the useful public material.
If machines can independently make investing decisions, Sosin thinks they will already be far along the capability spectrum. He is sometimes grateful to have succeeded beforehand, while remaining optimistic about productivity: “My great-grandchildren will marvel at my poverty.” Quantum simulation producing synthetic data for AI training is one pathway he finds especially intriguing.

20. Home-field knowledge and opportunity cost define the investable lane

Sosin’s US focus reflects epistemic limits more than a categorical claim that only America has great companies. Ask an investor pitching a British restaurant chain to name three places to buy a power drill in Britain; failure exposes how much tacit local knowledge even a seemingly familiar foreign market requires.
He views the United States as an extraordinary system and speculates that migration selected for independent, entrepreneurial people willing to cross an ocean for a better life. Still, his practical rule is simpler: an interesting Chinese stock goes “on the bottom of my list right after all the American stocks.” The domestic opportunity set is already vast.
For an average investor, his advice is conventional: an S&P 500 or broad-market ETF is “a great way to go.” Microsoft, Amazon, Google, and other giants are excellent businesses, but his concentrated fund cannot own every attractive idea; many belong in the joking “$100 billion portfolio,” not ahead of current holdings.
The real work is opportunity cost. Sosin may study a company deeply, conclude it is a great investment, and still reject it because it is slightly worse than what he owns. “The key isn’t to understand everything”; it is to know a handful of businesses, expect them to do well, watch them closely, and stop worrying about everything outside the chosen lane.

112: Talking Embodied Intelligence with 千寻’s 高阳: How Someone Who Looks Like a Robot Builds Robots That Look Human Market Overview, April 29, 2025