Records Across Every Dollar Line, Then a Guide That Implies No Second-Half Growth: Maintaining Outperform

Results vs. Consensus

A revenue and gross-profit blowout, an in-line-to-modest EPS beat, and a very slight miss on the one number the Street models most precisely. The composition matters more than the headline: the beat came from price and mix, not from volume, and not from per-unit economics.

A note on EPS comparability. Carvana completed a 5-for-1 stock split effective May 7, 2026, so this is the first quarter reported on a split-adjusted basis. The $0.42 diluted Class A figure here is not comparable to the $1.69 we reported for Q1 without dividing by five. The prior-year comparatives in the company's own tables ($0.27 basic, $0.26 diluted) are already restated. Note also that Class A diluted shares of 740.0 million understate the true economic count: assuming full conversion of LLC units there would have been 1.127 billion shares outstanding, which is the basis we use for all valuation math below.

Year-over-Year Comparison

Sequential GPU Comparison: the Number the Tape Ignored

The year-over-year GPU headline is unambiguously negative. The sequential picture is unambiguously positive, and it is the one that reflects the reconditioning recovery flowing through.

The same lines measured against the prior quarter, which is the basis that captures the reconditioning recovery flowing into cost of goods sold:

Revenue assessment. The revenue line is the most misread number in this print, in both directions. The bull reading of "52% growth" overstates the business, because roughly a third of the gap between revenue growth and unit growth is an accounting treatment that reverses as a comparison item next quarter. The bear reading of "they only grew units 38%" understates it, because the industry shrank roughly four points year over year in the same period. Share gain of that magnitude, at $30B of annualized revenue, is the entire investment case and it is intact.

Margin assessment. This is where we differ most from the tape's reading. Adjusted EBITDA margin fell 199bps to 10.4%. Total revenue per retail unit rose 10.7%, from $33,780 to $37,380 on a total-revenue-per-retail-unit basis. If revenue per unit had been unchanged year over year, the same $3,897 of adjusted EBITDA per unit would have produced an 11.5% margin. That decomposes the 199bps into roughly 111bps of denominator inflation and 88bps of real per-unit erosion. The denominator half disappears mechanically in Q3 when the gross-revenue-treatment comparison laps. The 88bps that remains is real, and it is the number that deserves scrutiny: it is roughly $300 per unit of lost adjusted EBITDA, which management bridged on the call to a lapped $100 tariff benefit, roughly $75 of fuel cost across the income statement, and higher benchmark rates. That bridge accounts for most of it.

EPS assessment. The cleanest line in the quarter and the least discussed. Net income of $513M was up 67% on 38% more units, net income margin expanded 59bps to 7.0%, and GAAP operating income converted at 88.4% of adjusted EBITDA versus 85.0% a year ago. For a company whose bear case has historically been "adjusted EBITDA is a story number," the fact that GAAP profitability is improving faster than adjusted profitability is the most thesis-relevant disclosure in the release. Retained earnings turned positive for the first time, at $551M against a $9M accumulated deficit at year-end.

The Guidance Question

This section exists because the guide is the entire reason a record quarter produced a 7.4% decline. It deserves arithmetic rather than adjectives.

Carvana guided full-year 2026 adjusted EBITDA to $2.7–3.0B, its first quantified FY26 framework after two consecutive quarters of qualitative "significant growth" language. That is 20.5% to 33.9% growth on FY25's $2.24B. On its own it is a perfectly good number. The problem is what it implies for the back half, given the first half is already in the bank.

The midpoint implies a second half 2.2% smaller than the first, and a quarterly average of $705M against the $769M just printed, in a company that has simultaneously guided Q3 retail units up sequentially. Selling more cars for less total profit is not what the last ten quarters have looked like, and the Street reacted to that inconsistency rather than to the absolute number.

Our take. Three things about the guide are worth separating. First, its style is genuinely unchanged: management has anchored a full-year adjusted EBITDA range at the mid-year print in each of the last two years, and framed this one identically. Second, its level is below where the sell-side was carrying numbers, with published estimates clustered around $3.0–3.2B before the print. Third, its shape is the only genuinely new information, and management refused to discuss it. We read the level as conservative and the shape as an artifact of that conservatism rather than as a signal, but we hold that view with less than full confidence precisely because the refusal to explain it was so consistent. Four different analysts probed the second-half math from four different angles and got the same answer each time.

Operating and Unit Economics

Segment Performance

Retail Vehicle

Retail revenue of $5.507B grew 61.7% on 37.7% more units, with revenue per retail vehicle up 17.4% to $27,908. Retail gross profit of $700M grew 34.4%, producing retail GPU of $3,547, down $89 year over year but up $382 sequentially. Against the Street's $3,282 estimate this was the single largest positive variance in the print.

The sequential recovery is the reconditioning fix arriving in the P&L, since costs are recognized when a car sells rather than when it is produced. The year-over-year decline is a lapped tariff benefit plus fuel. The beat versus estimates, however, came from somewhere else entirely: a regulatory change in how competitors advertise price.

Assessment: Retail GPU is now sitting above the Street's model on a mix of durable operational gain and a windfall of uncertain duration. Management says most dealers have already adjusted to the new pricing rules, which means the algorithmic lag that generated the outsized Q2 capture is largely spent. We would not extrapolate the $382 sequential gain.

Wholesale

Wholesale revenue of $1.343B grew 31.2% on 44.4% more wholesale units, with revenue per wholesale vehicle down 1.1%. Wholesale gross profit of $158M grew 19.7%. Expressed per retail unit, the metric the company reports, wholesale GPU fell $120 to $801, against a $952 estimate. Wholesale gross profit per wholesale unit fell 6.2% to $1,019.

This is the segment that quietly reversed. At Q1 the wholesale franchise was the standout, with wholesale gross profit per wholesale unit near a record and management calling its digital auction platform best in class. This quarter it was the second-largest per-unit disappointment relative to estimates, behind Other GPU, and it received no mention at all in the prepared remarks or in any analyst question.

Assessment: One quarter does not undo the structural argument, and volume growth of 44% is not the profile of a franchise in trouble. But we flagged the wholesale platform as a new discrete bull pillar at Q1 on the strength of a single quarter's per-unit result, and this quarter argues that result was closer to a peak than a new baseline. We move that pillar back to neutral pending another data point.

Other (Finance and Ancillary)

Other revenue of $526M grew 28.0%, the slowest of the three lines, and Other GPU fell $203 to $2,666, the largest per-unit decline in the print and a $188 miss against estimates. The cause is deliberate: Carvana passed back more than 100 basis points of interest rate to customers over the trailing year while benchmark rates rose, and chose to hold customer-facing rates steady in the quarter rather than reprice.

Management's arithmetic on this line was the most useful disclosure on the call, and we reproduce it below because it reframes what looks like deterioration as reinvestment.

“Given all the rates we passed back, all else constant, you probably would have expected something closer to a $500 reduction in Other GPU, but we actually saw was $200. That’s because there’s give or take $300 of fundamental gains in there.”
— Ernie Garcia, CEO

Assessment: If the $300 figure is right, the finance platform generated per-unit gains this quarter roughly equal to the entire year-over-year decline in adjusted EBITDA per unit, and management chose to hand them to customers. That is a defensible long-run choice and an entirely undefended near-term one, because it depresses reported economics precisely when the market is questioning whether per-unit economics can inflect. The pillar is confirmed on substance and is the direct cause of a bear point below.

Key Topics & Management Commentary

Overall Management Tone: Confident on the business and unusually rigid on the outlook. The prepared remarks were the most data-forward in several quarters, leaning on a new regional disclosure to argue that growth is an execution variable rather than a demand variable. In Q&A that confidence held everywhere except the second-half guide, where management drew a hard line and repeated a near-identical refusal to four separate lines of questioning. Posture was less accommodating than at Q1, when every question drew a substantive answer.

1. The Regional Inventory-to-Sales Disclosure

The most important new artifact of the quarter is not a financial metric. Carvana disclosed regional inventory growth alongside regional sales growth, which converts a decade-old conceptual argument about positive feedback into an observable correlation.

“The two regions where we added the most production capacity, the Midwest and the Northeast, grew inventory by 57%. In those markets, sales grew in the second quarter by 54%. In the two regions where we added the least incremental production capacity over the last year, the West and the Southeast, we grew inventory by 17%. In those regions, sales grew by 30% in the second quarter.”
— Ernie Garcia, CEO

Management extended the point later in the call to note that profitability does not vary across those regions, which is the harder half of the claim: growth follows selection without a margin penalty in the faster-growing markets.

Assessment: This is the strongest evidence yet that Carvana's growth ceiling is set by its own production capacity rather than by demand, and it materially raises our confidence in the long-run unit trajectory. It also cuts the other way in the near term, because it means the company is explicitly telling investors that the current quarter's growth rate was capped by a self-inflicted supply constraint. Both readings are true. The market took the second.

2. Inventory Undergrew Sales: the Binding Constraint Migrates

At Q4 the constraint was reconditioning cost. That is fixed. The fix consumed throughput, and the constraint moved.

“Inventory has undergrown sales over the last several months, and that certainly creates a headwind to just the overall business.”
— Ernie Garcia, CEO

Management described a three-step plan: drive reconditioning costs down, return to inventory growth, then normalize the vehicle mix back toward its traditional age and price profile. Step two started around mid-quarter and step three is underway now. The mechanical consequence was stated plainly: “When your inventory is less than you wish, that’s going to show up as lower conversion, and lower conversion means either lower sales or lower economics, all else constant.”

Assessment: This is the honest explanation for both the 38% unit growth and part of the per-unit compression, and management deserves credit for stating it rather than attributing the quarter to macro. It is also a real negative that did not exist in the Q1 picture. The constraint is being actively worked and the tooling is not yet fully deployed, so we treat it as a two-to-three-quarter drag rather than a structural cap. It is the primary reason we are watching Q3 inventory growth more closely than Q3 GPU.

3. Retail GPU and the FTC Advertised-Pricing Windfall

The largest single driver of the retail GPU beat was a regulatory change affecting competitors. New Federal Trade Commission guidance required dealers to include mandatory fees and required add-on products in advertised prices. Carvana charges neither, so competitor advertised prices rose while Carvana's did not.

“As a result, over the subsequent several months, basically from April through June, many dealers were adjusting to that change, that was changing the underlying data that we were seeing that powers our pricing algorithms.”
— Ernie Garcia, CEO

Management characterized the effect as the pricing shift having “led to some of the outsized gains” in retail GPU, and separately framed the longer-term effect as a durable relative-price advantage: “certainly for most dealers, let’s say that their prices on their website just went up $500 or $600 and ours didn’t move, that should be a tailwind in the grand scheme of things once all that settles out.”

Assessment: Two distinct effects are being conflated here and investors should separate them. The transient effect is algorithmic capture during the adjustment window, and management says most dealers have now adjusted, so it is largely spent. The durable effect is a structural widening of Carvana's advertised-price advantage, which shows up as conversion rather than as GPU and is genuinely valuable. We assume none of the transient benefit repeats and we do not yet model the durable one.

4. Other GPU: Roughly $300 of Fundamental Gains, Handed Back

Covered in the segment section above, but the strategic framing belongs here. Asked directly why capital is being returned to customers through rates while the business is production-constrained, management gave the most candid answer of the call.

“It is likely the case that giving back to customers in the form of rates over the last year is not something that we’ve been fully paid back for in growth. The fact that our inventory is tight sort of makes it clear that we didn’t get fully paid back for it in growth, because if the inventory gets tight, your conversion rates go down.”
— Ernie Garcia, CEO

Assessment: An unusual admission and a credible one. Management is saying it made a customer-value investment whose return it could not collect because the supply side was not ready to convert the demand. That is a timing error, not a strategy error, and the framing that it “puts pressure on the machine” to build faster is consistent with how this company has historically forced its own capacity decisions. Investors modelling a per-unit inflection should nonetheless assume the giveback continues, because management has now told them three quarters running that it intends to.

5. The Adjusted EBITDA Margin Optic

Management attributed the margin decline to the revenue denominator rather than to economics.

“Adjusted EBITDA margin was 10.4%, a decrease from 12.4%, primarily driven by increased retail revenue per unit resulting from the traditional gross revenue treatment mentioned previously.”
— Mark Jenkins, CFO

Our own decomposition puts roughly 111bps of the 199bps decline on the denominator and 88bps on per-unit economics, so "primarily" is defensible but not the whole story. The relevant forward point is Jenkins's other disclosure: “The gross revenue treatment change will no longer affect year-over-year comparisons beginning in Q3. We expect revenue growth to be more in line with retail unit growth in Q3.”

Assessment: Q3 will produce a headline revenue growth rate far below Q2's 52%, likely close to the unit growth rate, and the adjusted EBITDA margin optic will mechanically improve at the same time. Both moves are accounting artifacts. Investors who anchor on the revenue deceleration in October will be making the mirror image of the mistake the market made this week.

6. SG&A Leverage and the Q3 Advertising Step-Up

Non-GAAP SG&A per retail unit fell $157, with $272 of overhead leverage more than offsetting $88 of operations expense (largely fuel) and $27 of advertising. In absolute dollars, compensation and benefits fell sequentially from $245M to $237M while units grew 5%, which is the clearest evidence the fixed-cost base is now genuinely levering.

That leverage gets partly spent next quarter by choice. Management guided to higher advertising dollars in Q3 and was explicit about why: with inventory light, marketing is the substitute lever for conversion. “A good way to fill in that gap is marketing dollars.”

Assessment: Buying units with advertising when you cannot buy them with selection is the right short-run trade and a poor long-run one, and management appears to understand the distinction. It is worth noting this is a second-order consequence of the same inventory constraint, and it is one of the mechanical reasons the H2 guide can be soft without anything having gone wrong.

7. Balance Sheet: 1.0x, and a Return-on-Capital Frame

Net debt to trailing twelve-month adjusted EBITDA reached 1.0x, computed on $2.606B of net debt against $2.589B of trailing adjusted EBITDA. Total liquidity rose to $7.024B, interest expense fell 29% year over year, and retained earnings turned positive. Asked about capital returns at this leverage level, management reframed the question around reinvestment economics.

“We generated that $2.2 billion of operating income on only about $7.5 billion of net operating assets. We’re talking about on the order of 30% operating return on net operating assets, which is a pretty great business and a business that you really want to invest in.”
— Mark Jenkins, CFO

Assessment: A 30% operating return on net operating assets is the number that should govern the capital-allocation debate, and it settles it: buybacks at this return profile would be value-destructive relative to building capacity. Management introduced the return-on-capital framing at Q4 2025 as a "greater than 20%" figure; this is the first time it has been quantified and decomposed, and it is a materially better answer than the one this question received in prior quarters.

8. Mix Shift to Newer, More Expensive Cars

Average selling price rose sharply, with retail revenue per unit up 17.4%. Two causes, both temporary in management's telling: rate giveback skewed disproportionately to prime customers who buy more expensive cars, and step two of the inventory plan deliberately leaned into newer vehicles that recondition faster. Asked whether the shift is structural, the answer was direct: “I would say definitely not structural.” Management added that “over time, we expect all of our metrics to move toward the average car buyer.”

A revealing data point accompanied this: growth among customers with over $100,000 of income ran a little more than 60% year over year, well above the 38% company average, with management characterizing weaker growth in lower income bands as a supply outcome rather than a demand one.

Assessment: Consistent with the inventory story and mildly reassuring, because it says the growth deceleration is a composition effect of what Carvana chose to build rather than evidence of demand exhaustion at the low end. It also flags a modeling trap: if ASP normalizes downward as management intends, revenue growth decelerates faster than unit growth for a second consecutive reason.

9. AI and the Customer-Care Cost Curve

Asked about conversion benefits from the company's AI assistant, management redirected to the cost side with a four-year series that is more concrete than anything it has previously offered on this topic.

“three years ago, customer care costs went down 40% year-over-year. Two years ago, it went down an additional 30% year-over-year. One year ago, it went down 20% year-over-year. This year, we went down an additional 10% year-over-year.”
— Ernie Garcia, CEO

Assessment: Compounding, but decelerating in exactly the pattern you would expect as the easy automation is exhausted. The honest read is that this cost line has already delivered most of its available savings and is not a source of future margin surprise. Management framing it as evidence of ongoing AI leverage is the weakest argument on the call.

10. New-Car Dealerships: the First Substantive Disclosure

After declining to discuss the franchise-dealership strategy in each of the prior three quarters, management gave its first real color, still hedged.

“It’s pretty clear that new car NPS is very high and in a discontinuous sort of way.”
— Ernie Garcia, CEO

Pressed further, management added that “New cars are profitable for us today, beyond that, there’s not a ton of detail that we’re ready to provide.” and noted the operational problem is simpler than used, since manufacturing is outsourced. Asked twice whether new-car dealerships are helping the used inventory constraint, management declined both times.

Assessment: Two genuinely new facts after three quarters of silence: the customer experience is discontinuously better, and the unit economics are already positive. Neither is quantified and neither belongs in a model yet, but this has moved from an undisclosed option to a disclosed one. We add it to the thesis as emerging optionality with no value ascribed.

11. Capacity: 19 Sites, the First Full Buildout, and the Distance to 3 Million

Three more ADESA locations gained integrated retail production in the quarter, bringing the total to 19, and construction began on the first full ADESA buildout, expected to produce vehicles in early 2027. Current footprint supports roughly 1.5 million retail units when fully built out, with real estate to support 3 million. Against a Q2 annualized run rate near 789,000 units, the built-out capacity is nearly double current volume.

“When we announced this goal with our Q1 2025 results, we needed to grow to about six times our scale in order to achieve it. Now, five quarters later, we need to grow to under four times our current scale in order to achieve it.”
— Ernie Garcia, CEO

Assessment: The scale-gap arithmetic checks: the Q2 annualized run rate puts the company 3.8x from three million units. Compounding closed a third of the gap in five quarters. The capacity picture is the strongest counter to the growth-deceleration bear case, because the physical assets to support double the current volume are already owned. What is missing is throughput, which is precisely this quarter's problem and precisely what management says it is fixing.

Guidance & Outlook

Implied second-half ramp: $1,259–1,559M of adjusted EBITDA, or $630–780M per quarter against the $769M just delivered. Only the top of the range implies sequential growth from here.

Street versus the guide: Published sell-side FY26 adjusted EBITDA estimates clustered around $3.0–3.2B before the print, so the midpoint landed roughly 8% light and the low end roughly 13% light. Post-print, price targets were cut across most of the desks that publish on the name while ratings were left unchanged, which is the signature of a numbers reset rather than a thesis change.

Guidance style: Consistently conservative and consistently late. Carvana has never given a quantified full-year unit target, gives its adjusted EBITDA range only at mid-year, and refuses line-item forward color. Last year that same mid-year range was beaten. The difference this year is that the range implies a shape investors have not previously had to underwrite, and management declined to defend the shape.

Analyst Q&A Highlights

Q&A was dominated by two related questions asked in five different ways: when do per-unit economics inflect, and why does the second half look flat. Management gave ground on the first and none on the second.

When does adjusted EBITDA per unit return to growth?

The opening substantive challenge of the call, and the one management engaged with most fully. The questioner framed three consecutive quarters of declining per-unit EBITDA as a pattern rather than a blip and asked for a timeline. Management declined to give one but did supply a year-over-year bridge, which is the most useful decomposition available for modeling the line.

Q: “Ernie, in the past you have hinted that EBITDA per unit is an important metric that you care about and you know investors do as well. I mean, the used car industry will continue to have these pricing fluctuations, like the two-year rate fluctuations. We’ve had three quarters of lower EBITDA per unit in a row. I’m curious if you can give us a sense of when investors should expect that to return to growth. Do you think it can happen later this year? Is it a 2027 story?”
— Rajat Gupta, JPMorgan

A: “I think when you look year-over-year, I think this quarter we were down about $300 in EBITDA dollars per unit. Quarter-over-quarter, I think we were up about $300. ... Last year, we had tariff benefits that was around $100. Gas prices moved up pretty dramatically throughout the quarter. That probably cost us something on the order of $75 across the entirety of the income statement.”
— Ernie Garcia, CEO

Assessment: The bridge is credible and accounts for roughly 60% of the year-over-year decline through items that are either non-repeating or macro. What was conspicuously not offered was a date. The refusal to say "later this year" when directly invited to is, in our reading, the single most informative non-answer of the call, and it is consistent with a second-half guide that does not assume per-unit inflection.

Why does the full-year guide imply a smaller second half?

The dominant topic of the call, raised by two different analysts from different starting points. One built the case from the prior year's seasonal pattern; the other observed that back-half profitability has been stronger in each of the last four years. Both were told the company would not discuss the composition of its own guidance.

Q: “In the second half of 2025, you actually grew EBITDA versus the first half, despite a pretty challenging fourth quarter. The guidance would imply a step down this time, despite the fact that you’re actually now catching up to production better in the second quarter. ... Just curious, what’s driving that conservatism or is there anything that we’re missing in terms of the dynamics between first half and second half relative to last year, when it looks like things are actually getting better from an execution standpoint?”
— Multiple analysts incl. Rajat Gupta, JPMorgan; Joseph Spak, UBS

A: “Yeah. I think we got to stick with guidance. I think once we start giving guidance on guidance, I think it gets complicated.”
— Ernie Garcia, CEO

Assessment: The question explicitly offered management the word "conservatism" as an exit and management did not take it. That is the strongest single argument for the bear case, and we weight it accordingly. The counter is that this management team has historically refused to negotiate its guidance in public and then beaten it, which is exactly what happened with the equivalent range last year.

Why introduce a full-year framework now, and does it signal a change?

A well-constructed question that separated the act of guiding from the content of the guide, and got the most complete answer of the call because it let management defend process rather than numbers.

Q: “Now are you providing at least a range of EBITDA guidance that’s new? The question I’m going to ask is why issue the guidance now? Probably more importantly, as you think about that guidance, the parameters around that guidance, or maybe the thought process behind that guidance, are you signaling any type of change in the business from what we’ve seen here in the first half of the year, either from a sales or from a, more importantly, a profit standpoint?”
— Brian Nagel, Oppenheimer

A: “I think as we think about the guidance, it’s actually the same style of guidance that we’ve applied the last three years. ... When we reach mid-year here, we actually give more specific guidance on adjusted EBITDA. The goal there is just to let people know some guardrails around what we’re expecting in the second half.”
— Mark Jenkins, CFO

Assessment: "Guardrails" is the operative word and it supports the conservative reading: a guardrail is a floor you do not intend to touch, not a forecast. Management also explicitly denied signaling any change in the business. Taken with the prior year's beat of the equivalent range, this is the strongest evidence available that the guide is a floor rather than a forecast.

Disaggregating retail GPU: operational gain versus pricing environment

The first question of the call and the most analytically precise, asking management to separate the reconditioning recovery from a favorable pricing backdrop. The answer named a third factor neither side of the question had anticipated.

Q: “First question is on, what’s the progress update on reconditioning operations, specifically? That was something a lot of people were speaking about in the first half of the year. Can you disaggregate how much of the recent Retail GPU dynamics is coming from gains there versus a supportive or favorable used car pricing environment?”
— Daniela Haigian, Morgan Stanley

A: “I think the retail strength, the simplest and probably most dramatic part of that is during the quarter, the FTC put out guidance to all dealers that they were required to update their pricing to include doc fees and also any products that were required to be purchased with a car. For us, obviously, that has no impact. We haven’t had dealer fees, and we don’t have products that are required to be purchased with a car.”
— Ernie Garcia, CEO

Assessment: Management attributed the "most dramatic" part of retail GPU strength to a regulatory change rather than to its own operations, which is unusually candid and materially lowers the quality of the retail GPU beat. Anyone extrapolating the $382 sequential gain into Q3 is extrapolating a competitor-repricing window that management says has largely closed.

Reinvesting in customer rates while production-constrained

The sharpest strategic challenge of the call: why give economics back to customers at the exact moment the company cannot supply the demand it already has. Management acknowledged the apparent contradiction before answering it.

Q: “You guys made a decision earlier this year to reinvest a point back into financing costs. I guess just from a higher level, why is that the right decision versus investing into labor? Or some other choke point that you guys have to drive more production?”
— Andrew Boone, Citizens

A: “Why would you give money back to customers in the face of those constraints? The answer is that we expect to relieve those constraints, and we’re trying to make sure that we build the business in the best way that we can over the long term.”
— Ernie Garcia, CEO

Assessment: Management also stated plainly that growing the machine is not capital-constrained, calling the financial returns "extreme" and the limitation purely executional. That is an important admission: the company cannot spend its way past this constraint, which means the inventory drag resolves on an operational timeline rather than a capital one, and is therefore less controllable than a spending problem would be.

Capital returns at one turn of leverage

With leverage at a career-best 1.0x and cash compounding, the buyback question arrived for the first time with real force. The answer reframed it as a returns question and effectively closed it.

Q: “You guys are now at one times kind of leverage. What should we be thinking about or considering before you guys start to think about capital returns or some other form of return of cash to shareholders?”
— Andrew Boone, Citizens

A: “I think the most important thing that I think of when I think about what to do with our cash, it really comes down to investing in the business and generating returns and building the machine for the long term. ... We’re talking about on the order of 30% operating return on net operating assets, which is a pretty great business and a business that you really want to invest in.”
— Mark Jenkins, CFO

Assessment: The right answer, and the first time the company has attached a specific number to the return metric it began grounding capital allocation in two quarters ago. At a 30% incremental return on operating assets, capital returns would be a mistake, and investors hoping for a buyback should stop. The disclosure also gives the market a new and more defensible frame for valuing the business than the adjusted EBITDA margin it has been anchoring on.

The line-item wall: no forward GPU color

Three separate attempts to get directional GPU guidance for Q3 met the same response. Notable because the Q1 call had supplied exactly this kind of color, quantifying a tariff lap and a wholesale-to-retail spread headwind.

Q: “Given the degradation we’ve seen over the past year, which I think has run between $200-$150 a quarter, do you expect that to narrow as we get into the third quarter? It seems like you might have still some Other GPU compression, but perhaps the retail side is now getting better.”
— Multiple analysts incl. Sharon Zackfia, William Blair; Jeff Lick, Stephens; Tom Babcock, Barclays

A: “I think at a high level, we’re going to stick with our guidance on this. ... I think we’re going to stick with our guidance and try to stay away from giving too much detailed line item by line item color.”
— Ernie Garcia, CEO

Assessment: A clear reduction in disclosure versus the prior quarter, and the most defensible reading is that the full-year range is now the disclosure and management does not want to be held to a decomposition of it. That is internally consistent. It also removes the mechanism by which the market previously got comfortable with GPU noise, which is part of why the reaction was as sharp as it was.

Durability of the FTC pricing benefit

The right follow-up to the retail GPU disclosure, and the one that establishes how much of the beat is repeatable.

Q: “If I could do one quick follow-up, the FTC dynamic that you mentioned, is that still ongoing benefit to you, or do you feel like all dealers have pretty much adjusted their advertised pricing now?”
— Marvin Fong, BTIG

A: “We believe most dealers have adjusted their pricing. ... I think that it looks to us like most of those have probably been passed through. Some of the latter may still be lagging in. We certainly expect over time for that to be a positive for us.”
— Ernie Garcia, CEO

Assessment: The transient capture is over and the structural advantage remains. That is the correct answer and it confirms that the retail GPU beat should not be run forward. It also means the durable benefit, a permanently wider advertised-price gap against the entire franchise-dealer base, will show up in conversion and share rather than in the GPU line where the market is currently looking for it.

What They're NOT Saying

1. Any explanation of the second-half shape. Four separate lines of questioning, four refusals. Management would not say whether the implied sequential decline reflects conservatism, cost, mix, or comp. Given how readily the same team explains adverse per-unit moves, the silence on this specific point is the most notable omission of the call.
2. A quantified full-year retail unit target. Now four consecutive quarters without one, even as the company has quantified adjusted EBITDA. Units are the metric investors use to judge whether the share-gain engine is intact, and it remains the one management will not commit to.
3. Any Q3 adjusted EBITDA framing. The Q1 call guided Q2 explicitly to records on both units and adjusted EBITDA. This call guided Q3 only to a sequential increase in units. Dropping the profit half of that formulation, in the same quarter the full-year number disappointed, is a meaningful reduction in forward commitment.
4. Any per-line GPU direction for Q3. Explicitly declined three times, after supplying quantified GPU headwind guidance at Q1.
5. The wholesale franchise. Zero mentions in prepared remarks and zero analyst questions, one quarter after management called its digital auction platform best in class and we elevated it to a discrete thesis pillar. Wholesale GPU per retail unit was simultaneously the second-largest miss versus estimates. Silence on a line you promoted last quarter is worth noting.
6. Whether new-car dealerships are helping the inventory constraint. Asked twice in direct succession and declined both times, in a call that was otherwise willing to disclose that new cars are already profitable. The specific refusal to link the two suggests the answer is currently no.
7. Any capex figure for FY26. First-half capex of $102M is up 76% year over year and construction has started on the first full ADESA buildout, but no annual number has been provided in either of the last two quarters.
8. Cohort-level economics for mature markets. The fifth consecutive quarter without this disclosure. It remains the single cleanest way to validate the 13.5% margin target, and it remains unprovided.

Market Reaction

• Pre-print setup: CVNA closed at $61.43 on July 30 after entering the print at $66.32. Going in, the stock was -21.4% year to date against the S&P 500's +6.9%, -1.5% over trailing twelve months, and +4.1% over the trailing 30 days. The 52-week closing range was $56.26 to $95.69, so the print arrived with the stock roughly 31% below its high and only 18% above its low. This was a de-rated setup, not a crowded one.
• After-hours move: Shares fell to roughly $61.92 in extended trading, about -6.6% from the close, as the market read the full-year guide.
• Next-day session (July 30): Opened at $58.95, an -11.1% gap. Traded a range of $56.16 to $61.43, or -15.3% to -7.4% versus the pre-print close. Closed at $61.43, down 7.4%, at the high of the day.
• Volume: 31.6 million shares against a 30-day average of 10.7 million, a 3.0x multiple.
• Relative: The S&P 500 rose 1.7% on the same session, so the relative move was roughly nine points.

The intraday shape is the most informative part of the reaction. The stock gapped down 11%, extended to -15.3%, and then closed at the high of the session having recovered about a third of the gap. That is the signature of a guidance-driven flush met by buyers who worked through the arithmetic during the day rather than a repricing of the franchise. The morning traded the headline number; the afternoon traded the first half already banked and the record quarter underneath it.

The attribution is unusually clean and consistent: the full-year adjusted EBITDA range landed below where estimates were carried, and the implied second half was the specific offender. Nothing in the reaction was attributable to the quarter itself, which beat the Street on revenue and set company records on retail units, revenue, gross profit, SG&A per retail unit, GAAP operating income and adjusted EBITDA. The stock fell on the twelve words in the outlook section rather than on the sixteen pages preceding it.

Sell-side behavior confirms the reading. Price targets were reduced across most publishing desks, in a band running from roughly $80 to $93, while ratings were left unchanged and no downgrades were issued. The aggregate published target after the print sits near $117 against a $61.43 close. A universal numbers cut with no rating change is a re-based forecast, not a broken thesis, and it is consistent with our own conclusion.

Street Perspective

Debate: Is the full-year guide a forecast or a floor?

Bull view: Carvana gave a mid-year range of $2.0–2.2B last year and delivered $2.24B, above the top, through a quarter in which reconditioning costs stumbled badly. Management described this guide with the same language, called it "guardrails," and explicitly denied signaling any change in the business. Applying last year's conservatism factor points to roughly $3.04B and a normally-shaped second half.

Bear view: A company that expects to beat its guide usually says something when handed four consecutive opportunities to hint at it. Management instead repeated a rehearsed refusal. Meanwhile the observable drivers all point the same direction: inventory is short, advertising dollars are going up, the retail GPU windfall has passed, and per-unit EBITDA has declined year over year for three straight quarters. The guide may simply be right.

Our take: The bull side, but with less conviction than the precedent alone would justify. The precedent is strong and the "guardrails" framing is the language of a floor. What tempers it is that the observable drivers genuinely do point down for one or two quarters, so the second half could be soft without the guide having been conservative at all. We resolve this by treating the low end as a genuine tail rather than a formality, and by making Q3 inventory growth the variable we underwrite, since it is upstream of everything else in the debate.

Debate: Is per-unit economic compression cyclical or structural?

Bull view: The 199bps of margin compression decomposes to roughly 111bps of accounting denominator that laps out next quarter and 88bps of real erosion, and the real erosion bridges almost entirely to a lapped tariff benefit, fuel, and benchmark rates. Underneath it, the finance platform generated roughly $300 per unit of fundamental gains that management voluntarily handed to customers, and SG&A leverage delivered $272 per unit of overhead savings. Nothing here is competitive.

Bear view: Three consecutive quarters of declining per-unit EBITDA is a trend, and the company's own long-term target requires this metric to rise substantially. Every quarter management discovers gains and gives them away, the 13.5% margin date moves further out. At some point "we chose to reinvest" and "we cannot expand margins" become indistinguishable to a shareholder.

Our take: The bull decomposition is arithmetically correct and we verified it independently. But the bear point about indistinguishability is the better long-run argument, and it is why we treat the customer-reinvestment pillar as the most active risk in the thesis. The resolution is not analytical, it is empirical: Q3 laps the denominator distortion and brings the first quarter of fully-normalized reconditioning costs. If per-unit EBITDA does not rise year over year in Q3, the structural reading gets materially stronger.

Debate: Does the growth deceleration to 38% matter?

Bull view: The industry shrank roughly four points year over year while Carvana grew 38%, and the company disclosed regional data showing sales growth tracks production capacity almost one for one. Growth is capped by throughput, not demand, and the physical capacity to support double current volume is already owned. The two regions where capacity grew most grew sales 54%.

Bear view: Unit growth has now stepped down from 43% to 40% to 38%, retail units came in slightly below the Street's estimate, and the deceleration coincides with the company's own admission that it cannot supply the demand it has. Blaming your own execution is more comfortable than blaming demand but is not obviously better, because it means the bottleneck is recurring: reconditioning cost in Q4, reconditioning throughput now.

Our take: Firmly bull on the substance and alert to the pattern. The regional disclosure is the most persuasive evidence this company has ever provided that demand is not the constraint, and a 38% growth rate against a shrinking industry is extraordinary. The legitimate concern is not the growth rate but the repetition: this is the second consecutive constraint originating in the same part of the business, which suggests the ADESA production ramp is structurally harder to run at 40% growth than the capacity math implies. That is a reason to expect more bumps, not a reason to doubt the destination.

Model Update Needed

Valuation impact. At the $61.43 close and 1.127 billion fully-converted shares, market capitalization is roughly $69.2B and enterprise value roughly $71.8B against $2.606B of net debt. That is 25.2x EV to the FY26 guide midpoint, 23.9x to the top of the guide, and 23.4x the annualized Q2 run rate. On our $3.00B FY26 estimate the multiple is 23.9x. For a business compounding units at 38% into a shrinking industry, generating a 30% operating return on net operating assets, delevering to 1.0x, and holding roughly half its built-out capacity idle, we regard that as an undemanding multiple. It is roughly six turns below where the stock traded on this metric at the start of the year, with the operating record since then having been delivered.

Thesis Scorecard Post-Earnings

Overall: Thesis intact but no longer strengthening. Five of seven bull pillars confirmed, one downgraded to neutral on a single quarter's reversal, one added as unvalued optionality. On the bear side, valuation continues to recede as a concern while two risks moved against us: the growth-sustainability point escalated on a third consecutive step down in unit growth, and a new pillar is warranted around the per-unit inflection that the 13.5% target requires and that has not begun. The franchise is performing. What deteriorated this quarter is the visibility on when operating leverage converts into per-unit economics rather than into customer price.

Action: Maintaining Outperform. The upgrade thesis from Q1 rested on the reconditioning recovery and the balance-sheet inflection, and both delivered: reconditioning costs are fixed and leverage reached a career-best 1.0x. The quarter set records on every dollar line and improved GAAP profitability faster than adjusted profitability, which is the opposite of the pattern a deteriorating business shows. The guide is the bear case's strongest card and we do not dismiss it, but management's equivalent mid-year range last year was beaten, the "guardrails" framing is the language of a floor, and roughly half the headline margin compression is a denominator that mechanically disappears next quarter. At 25.2x enterprise value to the guide midpoint, after a 27% year-to-date de-rate, with half the built-out capacity idle and a 30% operating return on the capital already deployed, the risk/reward remains favorable against the S&P 500 over twelve months.

What would change our mind. We pre-committed upgrade triggers at Q4 and they worked, so we state the downgrade triggers with the same specificity. We move to Hold on any of the following: (a) Q3 inventory growth again fails to at least match retail unit growth, confirming the throughput constraint as structural rather than transitional; (b) Q3 adjusted EBITDA per unit declines year over year for a fourth consecutive quarter after the reconditioning normalization and the denominator lap have both landed, which would break the argument that the compression is transitory; or (c) full-year adjusted EBITDA is tracking to the lower half of the guided range at the Q3 print, which would mean the guide was a forecast rather than a floor and that our reading of management's disclosure style is wrong.

Net: A record quarter sold off on a number management would not explain. The business did what we underwrote at the upgrade; the outlook did not do what the market wanted. We are paid to distinguish those, and we stay Outperform.