Recurve Capital's Aaron Chan on the scalability of Carvana's $CVNA business model
Summary
- Aaron Chan’s differentiated bet is that Carvana ($CVNA) has entered a third rerating phase in which scalability—not survival or proof of profitability—is the central question. Phase one, “it’s not going bankrupt,” carried the stock roughly from $4 to $40; phase two, proving the model could outperform the industry, took it toward $125. With market share near 1%, sell-side unit-growth expectations in the low-20% range looked too low against early production data tracking closer to 50%, while Chan believes incremental margins can remain positive rather than deteriorate with growth.
- Carvana’s 99% collapse reflected a rare collision of operating leverage, financial leverage, and dependence on freezing capital markets—not a structurally collapsing used-car market. Industry volume only fell from roughly 40 million to 36 million units, but Carvana simultaneously overbought inventory, acquired ADESA with expensive capital, and carried costly third-party logistics and reconditioning. Since 2022 it has internalized more work, rebuilt unit economics, and reduced its liquidity risk; against roughly $1.5 billion of floor-plan capacity, utilization was under $100 million last quarter.
- The moat is a physical and financial system that would be extraordinarily expensive to recreate at Carvana’s current service level. Its DriveTime heritage supplied early facilities, loan-servicing expertise, and subprime underwriting knowledge; ADESA then helped regionalize a production footprint that had inefficiently shipped cars from the middle of the country toward the coasts. A startup buying at auction, outsourcing reconditioning, and delivering locally would pay variable costs everywhere yet struggle to match Carvana’s “almost Prime-like” three-to-four-day delivery.
- Carvana’s reported profitability looks optically incomparable with peers, but Chan argues genuine vertical-integration advantages remain after correcting the accounting definitions. Carvana excludes logistics from gross profit and includes wholesale profit in GPU, while CarMax classifies items differently; apples-to-apples rebucketing is therefore essential. Even then, sourcing roughly 80% of inventory directly from consumers, operating high-volume inspection and reconditioning lines, owning logistics, and originating financing across the credit spectrum may deliver $1,000 to nearly $2,000 of efficiency per unit.
- The short thesis clusters around the Garcia family’s control, Ernie Garcia Sr.’s legal history, related-party transactions, and Carvana’s exposure to subprime borrowers, but Chan finds the alleged distortions too small or too externally testable to carry the bear case. His estimate was that doubling the cost of all related-party arrangements would reduce EBITDA by only about 2.5%. He also rejects the idea that a roughly $400 billion ABS market repeatedly buys obviously defective paper without noticing: Carvana’s prime performance was better than CarMax’s, while Chan said some weaker early-2023 cohorts led Carvana to tighten underwriting.
- The cleanest way to falsify the long thesis would be mature-market share flattening near 5% or 6%, because that would turn a long-duration growth asset into a business approaching a visible wall. Carvana could still be profitable around 3 million units, or roughly 7.5% market share, but its multiple could compress sharply if investors could not underwrite growth beyond that point. Chan worries less about autonomous vehicles, believes EV reconditioning can be cheaper when battery replacements are avoided, and says an extra $200 rural delivery cost is manageable against roughly $7,300 of GPU.
- At approximately a $55 billion enterprise value, the valuation depends on how quickly Carvana fills infrastructure built for far more than its current roughly 400,000 units—and whether its advantages strengthen enough to extend the runway. Chan’s illustrative math uses about $5,000 of incremental EBITDA per unit: 3 million units would imply $15 billion of EBITDA, although the timing and terminal share remain uncertain. His strongest formulation is that “growth is more of a choice for Carvana than it is for any other business I’ve looked at,” with third-party fleet inventory, peer-to-peer transactions, and future marketplace distribution offering additional paths.
Deep dive
1. Carvana turns a hated purchase into a sub-hour transaction
Chan’s opening frame: Carvana is a “fully vertically integrated retailer,” combining financing, logistics, delivery, inspection, and reconditioning inside one system. Founded in 2012 by Ernie Garcia Jr. and spun out of DriveTime, it now holds roughly 1% of a large, stable market containing more than 40,000 independent used-car dealers.
The investment pattern Chan seeks is “disruptive companies in non-disruptive industries”—a differentiated operator attacking a sleepy category where the incumbent experience remains miserable. Walker’s family anecdote captured the stasis: a new-car purchase took five and a half hours, and the finance employee explained, “Nothing has changed. You just do this so infrequently, you forget how bad it is.”
Carvana’s Netflix-like disruption is experiential: searching, purchasing, and receiving a vehicle can require less than one hour of customer labor—roughly 10 to 20 minutes to find and check out, then another 10 to 20 minutes at delivery. The limitation is frequency: “You use Netflix every day, and you buy a car every six years,” so consumers repeatedly forget the pain Carvana removes.
Chan had studied the company for years but waited until the 2021-22 drawdown because he had not yet seen enough evidence that growth was scalable and repeatable. He had no legacy position as the stock fell from approximately $250-$300 into single digits.
2. The thesis has moved from survival to scalable incremental margins
Chan divides the rerating into three phases. Phase one was the distressed wager that Carvana would survive, carrying shares roughly from $4 to $40; phase two began with the debt restructuring and efficiency program, then ended as Carvana surpassed CarMax’s profitability around the first or second quarter, supporting a move from approximately $40 to $125.
Phase three is “the lowest IRR, but probably the biggest, longest phase”: determining how fast the company can grow, how much market share it can capture, and what economics attach to that growth. Even bullish models often assume deteriorating unit economics as volume accelerates; Chan thinks “that’s a mistake” and expects positive incremental margins.
Sell-side expectations called for low-20% unit growth, while early production data from the preceding three or four months suggested something closer to 50%. Chan stressed that the evidence was preliminary, but inventory was expanding and “the machine is kind of ramping up.”
Walker’s pushback—worth keeping—is that investors feared precisely this setup before COVID: Carvana pursued hypergrowth, “got ahead of their skis,” and suffered sharply negative economics. The unresolved trade is whether the repaired machine can accelerate without recreating the behavior behind its near-death experience.
3. The 99% collapse required three kinds of leverage to break at once
Chan’s closest analogy was American Tower around 2002, when the stock fell from approximately $60 to $0.60 before recovering over the long term. His “potion for massive volatility” combines high fixed costs and operating leverage, heavy financial leverage, and capital-market dependence just as growth slows and liquidity tightens.
Used-car demand itself was comparatively stable: annual industry volume is around 40 million units, plus or minus 10%, and the 2022 contraction was roughly from 40 million to 36 million. At sub-1% share, Carvana did not need industry growth; the damage came from consumer retrenchment, sharply higher capital costs, excess inventory, the ADESA acquisition, and impaired channels for recycling loan and inventory capital.
Before the crisis, Carvana filled infrastructure gaps with expensive vendors for paintless dent repair, logistics, middle-mile trucking, drivers, and other functions. It was therefore not yet operating the fully integrated model its architecture implied. The 2022-24 response pulled infrastructure, labor, workflow, and proprietary technology in-house, producing much of the subsequent unit-economics improvement.
Chan does not claim macro immunity: another shock could slow growth and hurt results. His narrower claim is that the previous “negative leverage on every front” is harder to recreate because organic cash flow is stronger and the company barely uses its floor plan—under $100 million drawn against roughly $1.5 billion of capacity last quarter.
4. Replicating the network means losing money before matching the service
Carvana was “born with advantages” unavailable to Vroom, Shift, and other challengers: DriveTime facilities, loan-servicing infrastructure, subprime-origination expertise, and inherited operating processes. Those advantages mattered before Carvana had national scale; its current brand, delivery density, and installed network make a fresh attack still harder.
Chan walked through the entrant’s bind. Buy cars at wholesale auctions and the challenger pays auction and acquisition fees; outsource reconditioning to a provider such as Manheim and it adds variable cost; require pickup and it remains local; offer home delivery and it must build logistics before having density. Vroom and Shift sometimes took around 20 days, versus Carvana’s roughly three-to-four-day average.
Even a deep-pocketed competitor would face physical constraints. Chan said Carvana’s Rocklin facility likely took about seven years to develop because California permitting and zoning are difficult and communities resist these sites. Before ADESA, this produced an “almost agricultural footprint”—several Ohio facilities, little coastal production, and expensive long-haul shipping toward demand.
His verdict was not that Amazon-sized capital could never enter, but that the funding proposition is unattractive after Carvana “crossed the chasm.” Asking a new company to prebuild national infrastructure and compete head-to-head now resembles building fulfillment centers to challenge Amazon: “Good luck.”
5. National branding sits above a deliberately regional inventory system
Walker challenged whether national scale matters in a historically local business with expensive last-mile delivery. Chan conceded the core point: buyers usually want a regional vehicle quickly, not one hauled from Florida to California, and the advantage is “more about what can you do regionally or locally” than exposing every VIN nationwide.
Carvana reconciles national selection with regional economics by charging customers for long-distance shipping, nudging them toward nearby inventory. Every vehicle is a “unique VIN,” unlike replicated fulfillment-center SKUs, but 200 locally available Toyota Camrys may substitute well enough for a nominal national pool of 1,000. ADESA added hubs, parking, and local production capacity to support that regionalization.
National advertising still builds awareness more efficiently than hyperlocal dealer campaigns, while the inventory and delivery algorithms operate locally. Chan’s honest non-answer on how much spending creates durable brand versus reacquires infrequent buyers was simply, “I don’t know.”
His advertising “aha moment” was instead denominator-driven: Carvana’s ads acquire sellers as well as buyers. It sources about 80% of inventory from consumers and buys more cars than it retails, yet expenses are commonly divided only by retail units sold. Counting both “sell to Carvana” and “buy from Carvana” transactions reveals operating leverage obscured by reported per-retail-unit metrics.
6. Accounting definitions explain part of the GPU gap; integration explains the rest
Walker noted that Carvana’s EBITDA per unit appeared to exceed some peers’ gross profit per unit. His comparisons included Lithia at roughly $4,000-$5,000 of gross profit per vehicle and CarMax around $2,300 on a used vehicle plus approximately $650 of wholesale gross profit—raising the question of how Carvana could earn more after SG&A than competitors before it.
Chan’s first answer was definitional: Carvana excludes middle-mile logistics from gross profit whereas CarMax includes it in cost of goods sold. Carvana also adds wholesale revenue and profit to GPU, then divides by retail units sold even though wholesale activity is not generated by that retail sale. Shipping, wholesale, and SG&A therefore require line-by-line rebucketing before comparison.
Genuine advantages remain after adjustment. Carvana acquires directly from consumers rather than paying auction economics, runs inspection and reconditioning facilities processing roughly 40,000-50,000 vehicles annually versus about 5,000 sales per CarMax location, and uses proprietary technology and assembly-line scheduling for high-volume production of nonstandard items. Chan estimates these integration gains exceed $1,000 and may approach $2,000 per unit.
Financing supplies another major delta. CarMax Auto Finance originates roughly 45% of CarMax volume, mainly prime loans, while lower tiers involve third parties and sometimes payments to place difficult credit. Carvana originates across the spectrum, with financing attached to about 80% of transactions; its DriveTime heritage creates comfort in higher-margin subprime lending that competitors may avoid.
7. The short case combines combustible optics with testable economics
Chan understands why Carvana attracts prominent shorts: Garcia family control, Ernie Garcia Sr.’s savings-and-loan-era legal history, super-voting shares, subprime exposure, and related-party dealings with DriveTime affiliates create an unusually potent collection of “buzzwords.” Carvana leases some facilities from DriveTime, Bridgecrest services its loans, and DriveTime pays commissions on vehicle-service contracts worth roughly $400 per unit.
The narrative itself has flipped. During Carvana’s losses, bears argued Garcia Sr. was overcharging it and transferring shareholder money to DriveTime; after profitability surged, they argued he must be undercharging Carvana and selling his stock to fund losses at DriveTime. Walker characterized the reversal as “really 4D chess. How can it be both?”
Chan’s quantitative rebuttal was that doubling every related-party cost would reduce EBITDA by only about 2.5%. Walker’s fair pushback was that Carvana should eliminate the arrangements if they are financially immaterial yet create constant suspicion; Chan agreed he would prefer arm’s-length counterparties, particularly where the market offers no clean external check.
Bridgecrest loan servicing does have a check: it services Carvana-originated loans even after buyers such as Ally acquire them. If servicing were deficient or overpriced, Chan argues those buyers would not keep renewing flow agreements, while ABS investors would reflect the problem in pricing and demand.
8. Subprime is uncomfortable, but loan vintages—not adjectives—decide the argument
Chan accepts that Carvana over-indexes to nonprime and subprime borrowers, aided by Bridgecrest’s servicing capabilities. What others call “unsavory”—loans at 22% or 23% interest with perhaps $4,000 down—can still be a viable product for someone with poor credit who urgently needs transportation; “it doesn’t mean the terms are attractive.”
The workflow broadens access: after a soft credit check that does not affect the customer’s score, buyers can “shop by monthly payment and down payment” across the entire inventory. Traditional dealers reverse that order—select and negotiate a car first, then disclose financing and the resulting monthly cost.
Chan’s empirical test is vintage performance. Carvana’s prime loans performed better than CarMax’s, in his analysis, while subprime results were broadly consistent with comparable paper. Recalling a 2023-N1 cohort, he estimated expected lifetime losses had risen from approximately 17.5% to roughly 22%-22.5%; Carvana tightened underwriting later in 2023, and subsequent curves returned closer to historical trends.
He called it “an arrogant position” to claim a roughly $400 billion, liquid ABS market repeatedly buys and trades Carvana paper without recognizing supposedly obvious defects. On former-employee accusations, Chan’s dozens of ecosystem calls contained both praise and criticism; fired employees and stock-owning alumni each bring incentives, so he tries to “take it all with a grain of salt” and verify the claims independently.
9. Mature-market saturation is the real risk; platform optionality is the upside
Chan’s clearest failure condition is mature-market share stalling around 5%-6%. If Atlanta and other seasoned markets flatten there, Carvana might still reach 3 million units—about 7.5% national share—and remain highly profitable, but the right-tail case would collapse as a growth stock “slams into a wall,” inviting severe multiple compression.
Autonomous transport and private-ownership erosion are lower-conviction wild cards because Chan believes Americans remain attached to owning transportation assets. EVs appear manageable: the Tesla Model 3 was reportedly Carvana’s most popular model the prior year, EV reconditioning can be cheaper, and Carvana avoids vehicles likely to need the costliest repair—a battery replacement. Rural service also works if an extra $200 delivery cost sits against roughly $7,300 of GPU.
The valuation case starts with unused capacity. At approximately 400,000 units, Carvana may utilize only 15% of infrastructure expandable to more than 3 million units with another roughly $1 billion of capital. With $7,300 GPU, $2,300-$2,400 of variable costs, and about $5,000 incremental EBITDA per unit, 3 million units imply $15 billion of EBITDA against the discussed $55 billion enterprise value.
Capacity is not destiny, and Walker pressed why a truly superior standardized platform would eventually stop at a minority share rather than take 50%-70%. Chan’s answer was optimization, not monopoly: some customers demand test drives, distrust reconditioning, dislike shipping or return terms, or accept a dealer’s “knife fight for four hours” for a lower price. Maximizing free cash flow per share need not mean serving 100% of transactions.
The longer-run option is marketplace expansion. Chan estimated peer-to-peer activity at roughly 15 million-16 million of the 40 million annual transactions, while Hertz is already placing fleet inventory through Carvana. Walker characterized the arrangement as EBITDA-per-unit neutral; Chan emphasized that Carvana captures retail, financing, and other services with light capital intensity. Chan likened third-party inventory to Amazon’s transition beyond first-party retail; future OEMs could likewise access national demand without building hundreds or thousands of dealers.
That optionality underpins Chan’s strongest claim: “Growth is more of a choice for Carvana than it is for any other business I’ve looked at.” As density increases, Carvana could shorten delivery, add pickup locations, pay more for supply, price more aggressively, or trade customer surplus across financing, convenience, and vehicle price—though “what should happen and what does happen rarely matches,” even over long horizons.