Carvana CVNA

Reference table

1. The June 2026 quarter — records and a margin that went the wrong way

All figures filing-verified from the 8-K filed 2026-07-29 and the 10-Q filed the same day.

The per-unit table is where the quarter actually lives, and it is not in the vendor payload in any form:

Read those two tables together and the quarter resolves cleanly. Carvana sold 38% more cars at prices roughly 10.7% higher, which is why revenue grew 52%. Gross profit per car fell $412 and fixed cost per car fell $278, so contribution per car fell $134. Because the denominator — revenue per car — rose $3,601 at the same time, the margin percentage fell twice as fast as the economics did: 200 basis points of adjusted EBITDA margin, of which roughly a third is genuine contribution loss and two thirds is arithmetic from higher average selling prices.

That distinction matters and we make it explicitly because it cuts both ways. A margin percentage falling because average selling prices rose is not a deterioration; a $298 fall in adjusted EBITDA per unit is. The per-unit measure is the honest one and it is down 7.1%.

Management's explanation, quoted in full because it is the causal claim the whole file rests on: "GAAP and Non-GAAP Total GPU were in line with our expectations in aggregate, with a shift in allocation between (i) Retail GPU, which was higher due to increasing industry retail prices following recent FTC guidance requiring automotive dealers to include mandatory dealer fees in advertised prices (note: Carvana doesn't charge these fees), and (ii) Other GPU, which was lower due to increasing benchmark rates. With both industry retail prices and benchmark rates drifting higher in the quarter, we followed the market on retail pricing but kept customer-facing interest rates stable."

Half-weighted as management framing — and note what it concedes. Carvana chose not to pass rising benchmark rates through to customer interest rates, and absorbed the difference in Other GPU. That is a demand-protecting decision, it is defensible, and it means the Other GPU compression is a policy choice with a volume benefit rather than an involuntary squeeze — which also means it can persist as long as the company keeps choosing volume.

2. Where the profit actually comes from

This is the single most important structural fact about Carvana and it is not derivable from any field in the vendor payload.

Other gross profit is principally the gain on sale of the auto loans Carvana originates, plus ancillary products. In the June quarter it was $526M against $680M of operating income — 77.4%. For FY2025 it was $1,733M against $1,881M of operating income — 92.1%.

The knowledge base names this correctly and it is the most useful thing in the lane. A claim dated 2025-11-25 states that "80% of profit comes from finance/interest, not car buying/selling." The filings put it at 77.4% for the most recent quarter and 92.1% for the most recent full year. That is a knowledge-base claim that is verifiable against a primary document and turns out to be right, which does not happen often enough to pass over.

What it means for risk. Carvana's operating income is, to a first approximation, a spread business layered on a retail business. The spread is set by the gap between the interest rate Carvana charges customers and the rate at which it can sell the resulting loans. Management has told us it is currently choosing to compress that spread to protect volume. A 15% decline in Other GPU from $2,666 to about $2,266, on an annualised 790,000 units, is roughly $316M of annual gross profit and, given the fixed cost base, close to the same amount of operating income — approximately 12% of the guided full-year adjusted EBITDA. That sensitivity, not used-car demand, is the risk that determines this equity.

3. The capacity that is already paid for

The bull case for Carvana is not the growth rate. It is that the growth needs almost no capital.

From the second-quarter shareholder letter: "In the quarter, we integrated retail production capabilities at three additional ADESA locations, bringing our total integrated sites to 19. We also began construction of our first full buildout at an ADESA location, which we expect will begin producing vehicles in early 2027. Our current footprint gives us fully built out annual capacity for approximately 1.5 million retail units, with real estate to support annual retail production of 3 million retail units."

Against a June-quarter run rate of 197,325 units — approximately 789,300 annualised — that is roughly 53% utilisation of built-out capacity and 26% of the land.

The capital numbers corroborate it. FY2025 purchases of property and equipment were $147 million on $20,322 million of revenue — 0.72% — against depreciation and amortisation of $275 million. Property and equipment, net moved from $2,814M at 2025-12-31 to $2,859M at 2026-06-30, a $45M increase, in six months during which retail units grew 38%. First-half 2026 capital expenditure was $102M.

And the operating leverage is already visible in the one line where it should appear first: SG&A per retail unit fell from $3,846 to $3,568 year on year, and on a non-GAAP basis from $3,385 to $3,228.

The stated destination, half-weighted as management framing: "We remain firmly on the path to selling 3 million cars per year and to achieving 13.5% Adjusted EBITDA margin by 2030 to 2035." That is a nine-year window with a four-year spread in the target date, which is a way of saying the company does not know when — and it is being stated in the same quarter adjusted EBITDA margin fell to 10.4%.

4. Balance sheet, the tax receivable agreement, and cash flow that is not a useful lens

Filing-verified at 2026-06-30 ($M):

The tax receivable agreement is the largest thing in this file that appears in no vendor field. At $2,130M it is 41% the size of total debt, $1,645M of it is owed to related parties — the pre-IPO LLC unitholders, principally the Garcia family — and it is a contractual cash obligation that ranks ahead of common shareholders. It is being paid down: $37M of tax receivable agreement payments in the first half of 2026, against $17M in the first half of 2025. At that pace it is a very long-dated liability, and any acceleration in Carvana's use of its deferred tax assets increases it.

The deferred tax asset is $2,968M — 20.4% of total assets. It arose from the FY2025 release of a valuation allowance, which is what produced the $2,785M income tax benefit in the FY2025 income statement. A company whose largest single asset after inventory and property is a deferred tax asset is a company whose book value is contingent on continuing to earn taxable income.

Cash flow is a poor lens for this business and we say so rather than quoting a misleading yield. First-half 2026 net cash provided by operating activities was $345M against net income of $918M, because operating cash flow absorbs both the $833M increase in vehicle inventory and the gross flows of the lending business — $9,005M of finance receivable originations against $9,323M of proceeds from sale in six months. Trailing-twelve-month operating cash flow is $1,120M and trailing capital expenditure is $191M, giving free cash flow of $929M — a 1.25% yield on the vendor's market capitalisation, which reconciles exactly to freeCashFlowYieldTTM of 1.245%. The vendor's arithmetic is correct. The measure is simply not informative for a company whose operating cash flow is dominated by inventory growth and loan warehousing, and we do not use it in the valuation.

5. Valuation — priced in or room?

Two share counts, and the difference is the first thing to settle. The vendor reports a market capitalisation of $74,606,762,069, which at $68.02 implies 1,096,835,667 shares. The 10-Q reports 719,916,415 Class A and 380,547,355 Class B shares outstanding at 2026-07-27 — 1,100,463,770 combined, within 0.33%. But the company states directly in the earnings release: "Assuming full conversion of LLC units and other dilutive effects, there would have been 1.127 billion shares of Class A common stock outstanding." On the company's own fully-converted count the market capitalisation is $76,658M, 2.7% above the vendor's.

Reported EPS is on a different base again. Diluted Class A EPS of $0.42 is $310M of income attributable to Carvana Co. divided by 739,960 thousand diluted Class A shares — it excludes both the 380,647 thousand Class A Units and the 6,487 thousand Class B Units held by the non-controlling interest, which the filing lists as anti-dilutive. So reported EPS is struck on roughly 66% of the fully-converted share count while capturing only 60.4% of consolidated net income, because Carvana Co. bears corporate tax that the pass-through units do not. On an as-converted basis, EPS is approximately 7-10% below the reported figure. Consensus is on the reported basis and so are the multiples below, but the overhang is real and stated.

At $68.02:

Estimate coverage is thin and gets worse fast, and the FY2029 row is internally incoherent. FY2026 rests on 8 analysts for EPS, FY2027 on 11, FY2028 on 6 and FY2029 on 4 — and the FY2029 mean of $2.590 is BELOW the FY2028 mean of $2.753 on revenue that is 12.3% higher. A consensus that has earnings falling on rising revenue two years out is not a forecast; it is a small number of models that do not agree and are being averaged. FY2029 and FY2030 are excluded from every conclusion here, and FY2028 is used only as a cross-check.

est.ebitdaAvg and est.ebitAvg are rejected — twelfth consecutive file. ebitAvg is negative in every one of the eight years present (FY2027: MINUS $381M) while ebitdaAvg is positive and absurdly small ($313M on revenue of $36.2bn — 0.86%, against a company that just guided to $2.7-3.0bn). This trips two of the three documented defect signatures simultaneously. The FY2023 row is worse: epsAvg of $6.205 against a reported split-adjusted actual of $0.15, and netIncomeAvg of $6.76bn against actual consolidated net income of $150M. All forward valuation uses epsAvg for FY2026-FY2028 only, cross-checked against the company's own adjusted EBITDA guidance.

Enterprise value — rebuilt, and the vendor is 5.4% low. enterpriseValueTTM of $77,598M less the vendor's market capitalisation is exactly $2,991M, which reconciles precisely to total debt of $5,236M plus the $385M non-current operating lease liability less cash of $2,630M. So the vendor includes operating leases in debt (see Section 7) and omits the $2,130M tax receivable agreement entirely, while using a market capitalisation 2.7% below the company's own fully-converted figure. Rebuilt:

On the rebuilt EV, EV / FY2026 guided adjusted EBITDA (midpoint $2.85bn) is 28.7x, and EV / FY2025 actual adjusted EBITDA of $2,237M is 36.6x.

The vendor's EBITDA field and everything derived from it are garbage and must be named as such. inc_a reports FY2025 ebitda of MINUS $110M, ratios_ttm reports ebitdaMarginTTM of 0.567%, km_ttm reports evToEBITDATTM of 546.5x and netDebtToEBITDATTM of 21.06x. Carvana's own FY2025 adjusted EBITDA was $2,237M at an 11.0% margin. The cause is documented in Section 7 and it is a single line: $2,250M of "Other expense, net" in FY2025, which drove pretax income to MINUS $890M despite operating income of $1,881M.

5a. What today's price assumes (the inversion)

At $68.02 — 42.2x FY2026 consensus Class A EPS, 30.9x FY2027, 28.7x the midpoint of the company's own FY2026 adjusted EBITDA guidance — the price embeds:

• Revenue of $28.83bn in FY2026 and $36.21bn in FY2027. (Consensus, 14 and 16 analysts.) The first half delivered $13.81bn, so FY2026 needs $15.02bn in the second half — +8.8% sequentially per half, which the September consensus of $7.57bn is consistent with. This is the least demanding assumption in the price and we think it is likely met.
• Class A diluted EPS of $2.198 in FY2027, up 36.3%. (Consensus, 11 analysts.) On revenue growth of 25.6%, that requires roughly 150 basis points of margin expansion — from a base the company has just guided down.
• Adjusted EBITDA margin recovering above the guided 9.4-10.4%. (Our derivation; the guidance is management's.) This is the most fragile assumption in the price. FY2025 delivered 11.0%; the June quarter delivered 10.4%; guidance for FY2026 implies 9.4-10.4%. Every forward estimate in this file requires the margin to turn, and the company's own guidance says it has not turned yet.
• Other gross profit per unit stabilising near $2,666. (Our derivation; the filing supplies the figure.) This line is 77.4% of operating income and management says it is falling on benchmark rates. A 15% further decline is roughly 12% of guided full-year adjusted EBITDA.
• No meaningful dilution from LLC unit conversion. (Our number.) Reported EPS is struck on 740M diluted Class A shares against 1.127bn fully converted. The multiples above would be roughly 7-10% higher on a fully-converted basis.

5b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+36.3%, from FY2026E $1.613 to FY2027E $2.198) + multiple drift (COMPRESSION from 42.2x to 30.9x at a constant price, which is rolldown, not de-rating) + shareholder yield (ZERO — no dividend, no buyback, and net share issuance from equity compensation).

Our base assumes the market pays roughly 31x the then-forward year, unchanged from where it pays today on FY2027. No expansion, no compression. And that assumption is what produces a base fair value equal to the current price. There is no shareholder yield to add, so the entire return has to come from earnings.

State the arithmetic plainly. If FY2027 Class A EPS of $2.198 arrives and the market still pays 31x, the stock is $68.1. To make money from here you need either an earnings beat or multiple expansion, and multiple expansion at 31x forward requires the market to underwrite the 3-million-unit, 13.5%-margin destination that management dates "2030 to 2035". That is a nine-year option, and at 31x forward earnings a meaningful part of it is already in the price.

Downside symmetry, for completeness: at 24x FY2027E the stock is $52.8 (MINUS 22.4%); at 38x it is $83.5 (+22.8%).

5c. Variant perception (where we differ, what would surprise)

• We have NO variant perception on this name and it is the reason for the verdict. Our base of $68 is 20.4% below the street's $85.45, but that gap is entirely multiple choice, not a different forecast — we use the same consensus revenue and EPS rows. Meanwhile the vendor's own consensus label across 46 analysts is "Hold" (21 buy, 22 hold, 3 sell), and the lowest published target of $67 is below the current price. The street is not enthusiastic and neither are we, and we are not going to invent a disagreement to justify a more interesting verdict.
• We think the market is under-weighting the guided margin decline. Management has guided FY2026 adjusted EBITDA to $2.7-3.0bn, which on the consensus revenue line implies 9.4-10.4% against the 11.0% delivered in 2025. A 41.9% revenue increase producing a lower margin is not what "positive scaling properties" describes, and the FY2027 consensus of +36.3% EPS growth requires the opposite. Watchable number: adjusted EBITDA margin in the 2026-10-28 release against 10.4%.
• We think the loan-sale concentration deserves the weight the knowledge base gives it and the sell side does not. 77.4% of operating income from Other gross profit is a fact none of the 46 analyst targets is expressed in terms of, and it is the variable management has told us is currently compressing.
• Where we are more positive than we sound: the capacity is real, paid for, and quantified. 1.5 million units of built-out capacity against a 790,000 run rate, on $147M of annual capital expenditure, is a genuinely unusual asset and the reason the exponential score is 7. Positive surprise that would force a re-rate: a quarter in which adjusted EBITDA per unit rises year on year — it fell $298 this quarter — which would prove the leverage curve is intact while volume compounds.
• Negative surprise that would break the thesis: Other GPU below roughly $2,400, which would take a further 10% out of the largest profit component; or a quarter in which retail units fail to rise sequentially after management guided that they would; or any impairment of the $2,968M deferred tax asset.

Synthos fair values

All three anchors are multiples of the FY2027 consensus Class A diluted EPS distribution (mean $2.198, low $1.997, high $2.489, 11 analysts), each cross-checked on EV / adjusted EBITDA against the rebuilt enterprise value.

• Bear ~$38 — 19.0x the FY2027 consensus LOW of $1.997. Cross-checks: an EV of $47.9bn (on 1.127bn shares plus $5.1bn of net obligations) against an FY2027 adjusted EBITDA of $3.0bn, 16.0x; 32.4% below the 52-week low of $56.26. The scenario: benchmark rates keep rising, Other GPU falls below $2,400, unit growth decelerates as the industry weakens, the 13.5% margin target loses credibility, and a beta-3.46 stock de-rates to a retailer multiple. MINUS 44.1%.
• Base ~$68 — 30.9x the FY2027 consensus MEAN of $2.198. Cross-checks: 42.2x FY2026E; 24.7x FY2028E; a rebuilt EV of $81.8bn against an FY2027 adjusted EBITDA of $3.62bn (10.0% margin on consensus revenue), 22.6x. Sensitivity, stated openly: 24x gives $52.8 and 38x gives $83.5. The base is the price, and the honest reading of that is that we cannot tell which way it goes. The scenario: units compound at the consensus rate, adjusted EBITDA margin stabilises near 10%, Other GPU holds, and the multiple is unchanged. MINUS 0.03%.
• Bull ~$100 — 40.2x the FY2027 consensus HIGH of $2.489. Cross-checks: an EV of $117.8bn against an FY2027 adjusted EBITDA of $3.98bn (11.0% margin), 29.6x; 4.5% above the 52-week high of $95.69; 17.0% above the street's $85.45 and 16.7% below the street high of $120. The scenario: benchmark rates fall, Other GPU recovers toward $2,900, adjusted EBITDA per unit resumes rising while volume compounds above 30%, and the market starts underwriting the 3-million-unit destination rather than the next four quarters. +47.0%.

Base is at spot; asymmetry roughly 1.07:1 (44.1% down, 47.0% up), with no dividend and no buyback. A base case that equals the closing price, on a payoff ratio of essentially one, with no shareholder yield and a beta of 3.46, is a Hold. Stating it as anything else would be dishonest.

6. Knowledge base — two claims, they disagree, and the bullish one is checkable

Raw hits: 15. After a case-sensitive entity test: 2. Name-level claims on Carvana: 2, one BULLISH and one BEARISH. Discarded: 13.

The sweep ran entity terms CVNA and Carvana plus free text on carvana, used car, used-car and auto retail across all 52,021 distilled claims.

The two exact-entity claims, verbatim:

> 2025-11-25 · BULLISH · conviction 80 · horizon: fact · entities: Carvana, CVNA · channel: natalie_brunell · no named speaker · skill 1.0 · categories: turnaround, auto retail, financing

> "Carvana was a left-for-dead 100x turnaround ($3.50→$400+); 80% of profit comes from finance/interest, not car buying/selling."

> 2023-08-11 · BEARISH · conviction 70 · horizon: fact · entities: CVNA · channel: mike_green · speaker: Michael Green · role: independent · skill 0.9 · categories: auto retail, distressed credit

> "Recently put on a Carvana short; its debt-restructuring rally requires issuing $1B of equity, announced 20 days ago with ~10 days left — a hard raise for a structurally unprofitable business."

The bullish claim's factual assertion is testable and it is correct. "80% of profit comes from finance/interest, not car buying/selling." Against the filings: Other gross profit was $526M of $680M of June-quarter operating income — 77.4% — and $1,733M of $1,881M of FY2025 operating income — 92.1%. The claim understates it on a full-year basis. This is the most directly verifiable knowledge-base claim in this batch and it verifies. The price observation in the first half of the claim is retrospective rather than predictive, and it is stated pre-split: adjusted for the five-for-one split of 2026-05-08, the range it cites is roughly $0.70 to $80, against a close of $68.02.

The bearish claim has been decisively and expensively wrong, and we grade it as such rather than excluding it. Written 2023-08-11 as a live short recommendation predicated on Carvana being "structurally unprofitable" and unable to complete a $1bn equity raise: the restructuring completed, the raise cleared, Carvana has reported positive net income in every quarter since, produced $1,895M of consolidated net income in FY2025 and $918M in the first half of 2026, and the stock is a large multiple of where it stood. A knowledge base that only surfaces the claims that aged well is not a knowledge base. It is worth noting that the claim carries a named speaker with an independent role and a skill weight of 0.9 — the best-attributed claim in this file — and it was still wrong. Attribution quality and accuracy are different things and this file demonstrates it.

The discard, and one part of it is more interesting than a discard. Ten of the thirteen text-only hits are macroeconomic claims about used-car PRICES as a component of the consumer price index — arthur_hayes (2024-08-19, on auction data re-accelerating), andreas_steno (2024-08-19), warren_pies (2022-12-16, on Mannheim near MINUS 15%), luke_gromen (2022-12-08 and 2025-01-15), jim_bianco (2021-08-20), michael_howell (2022-10-14) and darius_dale (2021-08-09). These are about an inflation input, not about a retailer, and the collision between "used car" the CPI line and Carvana the company is a homograph failure of exactly the class the store registry documents. They are discarded and none of them is a signal on this equity in either direction.

Three of the thirteen are worth one sentence. On a single day — 2025-11-25 — natalie_brunell used Carvana as the reference class for Opendoor ("Like Carvana earns most profit from financing, Open Door's real upside is bundling mortgage/title/ancillary revenue", conviction 78), for bitcoin miners converting to data centres ("IREN, Cipher are 'next Carvana' 100-bagger candidates", conviction 75), and for small-capitalisation crypto ("'Carvana coins' offer 100x opportunities", conviction 55). Carvana has become this knowledge base's canonical analogy for a hundred-bagger turnaround. That enters no score. It is worth recording because a stock that has become the benchmark against which other people's speculative upside is measured is not a contrarian position, and the current price is 28.9% below its high in a market that rose 24.3%.

Attribution note: the bullish claim carries speaker: null and is channel-attributed — the weakest sourcing the four-lane policy admits. The bearish claim carries a named speaker and an independent role. No claim in this file carries a management speaker_role. No concentration test is meaningful across two claims from two channels.

Conclusion. Breadth 2, claim count 2, net conviction mixed-low. The knowledge base holds one bullish claim whose central factual assertion the filings confirm and one bearish claim that has been comprehensively wrong. Net, it tells us something true about how Carvana makes money and nothing usable about what the stock is worth.

7. Data integrity — what we rejected and why

Nine findings. The Carvana payload has the widest spread of any file in this batch: two fields are exactly right, and one field is wrong by more than $2.3 billion.

1. A five-for-one forward stock split effected 2026-05-08 is absent from every vendor field as an EVENT — though the numbers are correctly adjusted. From the 8-K filed 2026-05-06: "At the Annual Meeting, the Company's stockholders approved an amendment... to effect a five-for-one forward stock split of the Company's Class A common stock and Class B common stock... The Stock Split will become effective on May 7, 2026 at 9:30 a.m. ET." The 10-Q confirms it was "effected on May 8, 2026" and that "All share and per share data throughout this Quarterly Report... have been retroactively adjusted." We checked whether the vendor adjusted, and it did: the 10-K reports FY2025 diluted Class A EPS of $8.45 on 224,277 thousand shares; inc_a reports $1.69 on 1,121,385 thousand — exactly one fifth and exactly five times. FY2024 ($1.59 → $0.318) and FY2023 ($0.75 → $0.15) likewise. The adjustment is correct. What is absent is any flag that the split happened, which matters for anyone comparing this payload to a price history or a screen struck before 8 May.

2. The $2,130M tax receivable agreement liability appears in no vendor field — the largest single omission in this file. Of it, $1,645M is owed to related parties. It is a contractual obligation to pay pre-IPO LLC unitholders a share of realised tax savings, it ranks ahead of common shareholders, and $37M was paid in the first half of 2026. It is 41% the size of total debt and it is absent from totalDebt, netDebt and enterpriseValueTTM.

3. The market capitalisation understates the company's own fully-converted share count by 2.7%. The vendor's $74,606,762,069 implies 1,096,835,667 shares. The earnings release states: "Assuming full conversion of LLC units and other dilutive effects, there would have been 1.127 billion shares of Class A common stock outstanding." On the company's figure the market capitalisation is $76,658M. The 10-Q's outstanding counts at 2026-07-27 — 719,916,415 Class A and 380,547,355 Class B — sum to 1,100,463,770, within 0.33% of the vendor's implied count, so the vendor is counting shares outstanding rather than shares as-converted. Both are defensible; they are not the same number, and the multiples differ by 2.7% depending on which is used.

4. inc_a.ebitda is MINUS $110M for FY2025 against a company-reported adjusted EBITDA of $2,237M — REJECTED, along with every ratio derived from it. The mechanism is fully traceable: the vendor computes EBITDA from pretax income, and FY2025 pretax income was MINUS $890M because of $2,250M of "Other expense, net" sitting below operating income of $1,881M. Consequential rejections: ebitdaMarginTTM 0.567%, evToEBITDATTM 546.5x, netDebtToEBITDATTM 21.06x, interestBurdenTTM 4.254 and taxBurdenTTM MINUS 2.835. All EBITDA figures in this dive come from the company's own reconciliation.

5. inc_a FY2024 gross profit is understated by $165M — partial-row corruption, and the residual matches a named line exactly. The vendor reports FY2024 costOfRevenue of $10,962M and grossProfit of $2,711M; the 10-K reports $10,797M and $2,876M. The vendor reports sellingGeneralAndAdministrativeExpenses of $1,709M against the filing's $1,874M. Both discrepancies are exactly $165M, and $165M is precisely the "Depreciation and amortization expense in SG&A expenses" line in the FY2024 adjusted-EBITDA reconciliation. The vendor has moved that line from SG&A into cost of revenue, understating FY2024 gross margin by 121 basis points and creating a spurious 322-basis-point gross-margin comparison against FY2025, which the vendor reports correctly. The FY2023 SG&A figure of $1,606M likewise disagrees with the filing's $1,796M. FY2025 is exactly right on all four lines. This is the partial-row class: the same field is correct in one year and wrong in the adjacent one, and each must be checked individually.

6. totalDebt is inflated by the non-current operating lease liability, and the residual matches to the dollar — REJECTED. The vendor reports FY2025 totalDebt of $5,521M with capitalLeaseObligations of $406M. The 10-K balance sheet shows short-term revolving facilities $58M, current portion of long-term debt $227M and long-term debt $4,830M — $5,115M — plus a separately captioned "Operating lease liabilities, excluding current portion" of $406M. $5,521M less $5,115M is $406M, exactly the operating lease line, which the vendor has relabelled as a capital lease obligation. Carvana's operating leases are for retail and logistics property, not equipment financing. This is the fourth consecutive batch in which the lease-inflation check has fired, and the second name in this batch.

7. enterpriseValueTTM carries the lease inflation forward and omits the tax receivable agreement — 5.4% understated overall. enterpriseValueTTM of $77,598M less market capitalisation is $2,991M, which reconciles exactly to total debt $5,236M plus operating leases $385M less cash $2,630M at 2026-06-30. It omits the $2,130M tax receivable agreement and uses the lower share count. Rebuilt on the company's fully-converted count: $81,779M.

8. seg_prod drops the wholesale revenue line after FY2022, leaving a $4,052M hole — REJECTED. For FY2025 the vendor reports only Used Vehicle Sales $14,537M and Product and Service, Other $1,733M, summing to $16,270M against total revenue of $20,322M. The missing $4,052M is "Wholesale sales and revenues", which the vendor reported through FY2022 and then stopped. Both reported figures match the filing exactly; the omission is the defect. A reader would conclude Carvana's revenue does not reconcile to its own segments.

9. seg_geo is EMPTY. Carvana operates only in the United States, so a single-line geography would be uninformative — but an empty array is not the same as a correctly reported single geography, and it is recorded as the ninth seg_geo failure in the last two batches.

VERIFIED CLEAN — recorded because a clean check is a finding.

capitalExpenditure is exactly correct in all three years. FY2025 $147M, FY2024 $91M, FY2023 $87M — each an exact match to "Purchases of property and equipment" in the 10-K cash-flow statement. The figure is remarkably low — 0.72% of FY2025 revenue, 53% of depreciation — and the temptation is to flag it as the "capex below 15% of D&A" suspicion signature. It is not a defect. Carvana bought the ADESA estate in 2022 and is integrating rather than building: property and equipment net moved only $45M across the first half of 2026 while units grew 38%, and the company states it already has built-out capacity for approximately 1.5 million units. The low capex is the thesis, not an error.

freeCashFlowYieldTTM of 1.245% is arithmetically correct. Recomputed: trailing operating cash flow $1,120M less trailing capital expenditure $191M equals $929M, on a $74,607M market capitalisation — 1.25%. The number is right; the measure is uninformative for a business whose operating cash flow absorbs $833M of inventory growth and $9,005M of loan originations in six months, and it is not used in this valuation.

The split adjustment is correct, as documented in finding 1.

Non-equity tripwire — checked and passed. CVNA is Class A common stock, $0.001 par value, NYSE-listed, per the 10-Q balance sheet (2,500,000 thousand shares authorised, 718,978 thousand issued and outstanding at 2026-06-30). Price of $68.02 is not par-like; beta is 3.462; there is no dividend; volume was 6.54M shares (~$445M of turnover); the 52-week band of $56.26 to $95.69 is a 70% range. This is common equity. Note the separate 380,547 thousand Class B shares, which are voting stubs paired with LLC Units and are NOT this security, and the 50,000 thousand authorised preferred shares of which none is issued.

Vendor composite rating — noted, not used. The rating block gives B- / 3 overall, with 1 of 5 on discounted cash flow, 1 of 5 on debt-to-equity and 1 of 5 on price-to-book. The discounted-cash-flow sub-score rests on the free-cash-flow measure rejected as uninformative above, and the debt-to-equity sub-score uses a totalDebt figure inflated by operating leases while ignoring a $2.1bn tax receivable agreement. Neither carries weight.

8. Technicals

• Price $68.02. MINUS 28.9% from the 52-week high of $95.69; +20.9% above the 52-week low of $56.26. Position within the annual range: 30th percentile — the lowest in this batch.
• Between the averages, which is the least informative configuration there is: +1.8% above a 50-day average of $66.80, MINUS 5.9% below a 200-day average of $72.28. The 50-day sits below the 200-day, a bearish medium-term configuration that the last few weeks have been working against.
• RSI 45.9 — neutral, slightly soft. MACD MINUS 0.62, negative.
• Current distance from the six-year peak: MINUS 28.9%, identical to the 252-day figure — which means the six-year peak falls inside the last twelve months and the coincidence is correct, not a defect.
• Relative performance is uniformly poor across every window: 3-month MINUS 9.7% against SPY +7.6%; 6-month MINUS 16.6% against SPY +11.1%; 12-month MINUS 7.5% against SPY +24.3% and QQQ +30.8% — a 32-point twelve-month deficit. A company that grew units 38% and revenue 52% has underperformed the index by 32 points over a year. That is the market pricing the margin, not the volume, and it is the single most useful thing the chart says.
• Beta 3.462 — the highest in this batch by a distance. A 10% market move implies roughly a 35% move here. Position sizing on this name is not a detail.
• Sentiment: 21 buy, 22 hold, 2 sell, 1 strong sell across 46 analysts, and the vendor's own consensus label is "Hold" — the only name in this batch where it is not "Buy". Consensus target $85.45 (+25.6%), median $82, high $120 (+76.4%), low $67 — 1.5% BELOW spot. The spread between the high and low targets is 79% of the low, which is a coverage universe that does not agree on what this company is worth.

Today's move and what it does to the entry

CVNA closed 2026-08-04 at $68.02, up 2.86% or $1.89 from $66.13. It opened at $67.57, traded $66.82 to $69.33, and closed in the upper half of the range on 6.54M shares. No company-specific filing is dated 2026-08-04; the last event was the 2026-07-29 earnings release, six days earlier.

The entry is neither a discount nor a chase. The stock is 28.9% off its high, below its 200-day average, and has underperformed the index by 32 points over twelve months — so this is not momentum. It is also not a washout: it is 20.9% above its 52-week low and RSI is a neutral 45.9. What it is, is a stock trading at the point where our arithmetic says it should trade, six days after a quarter that was excellent on volume and mediocre on margin. There is no informational edge in the entry.

9. Insiders — one officer, one day, and options struck at $2.01

The reading, and it is mostly benign. Every transaction in the file is dated 1 July 2026 and seven of the eight belong to one person. Mark Jenkins, the chief financial officer, exercised 63,750 options struck at $2.01, $8.41 and $10.39 and sold 38,360 shares at $65.18 to $68.14 on the same day. That is a textbook exercise-and-sell to fund the strike and the tax, and it is signal-light. He retains 1,054,970 shares — worth roughly $71.8 million at today's price — so the net position after the transaction is far larger than before. The Breaux transaction is an F-InKind tax withholding, the mechanically signal-free type.

Three observations that do carry information. First, the option strikes of $2.01, $8.41 and $10.39 — post-split — are a reminder of what this company was: those are grants from the 2022-2023 near-death period, and they are now worth 33x, 8x and 6.5x. Second, there is not one open-market purchase in the file. Third, the sales were executed at $65.18 to $68.14, which brackets today's close of $68.02 — so the chief financial officer's revealed indifference point is approximately the current price.

What the file does not contain: any transaction by Ernest Garcia III, the chairman and chief executive, or by Ernest Garcia II. The Garcia family controls the company through Class B shares carrying 707,088,671 of 850,082,440 votes at the 2026 annual meeting. The absence of any family transaction is neutral and uninformative.

10. Verdict, kill-criteria and flip conditions

Hold.

We are going to be direct about why this is Hold rather than something more interesting, because the temptation with a 38%-growth company is to manufacture a view.

The arithmetic lands on the price. At $68.02 Carvana trades at 42.2x FY2026 consensus Class A EPS of $1.613, 30.9x FY2027's $2.198, and 28.7x the midpoint of the company's own FY2026 adjusted EBITDA guidance on a rebuilt enterprise value of $81.78bn. Our base fair value of $68 is the closing price to within three cents, with 1.07:1 asymmetry and no dividend or buyback to add. Our number is 20.4% below the street's $85.45, but the gap is multiple choice, not forecast — we use the same consensus rows — and the vendor's own summary of 46 analysts is "Hold", with the lowest published target below the current price.

What is genuinely excellent, stated without hedging. 197,325 retail units, +38%, against an industry that shrank. Ten consecutive quarters as both the fastest-growing and the most profitable automotive retailer. 2% share of used retail. Reconditioning capacity already built for approximately 1.5 million units against a run rate near 790,000, real estate for 3 million, and FY2025 capital expenditure of $147 million on $20.3 billion of revenue. SG&A per retail unit down $278 year on year. This is one of the better operating records in this programme and none of the caution below is a criticism of the execution.

What stops it being a Buy, in three specific numbers. First: adjusted EBITDA per retail unit fell $298 to $3,897 — the leverage curve went backwards in the quarter volume grew 38%. Second: management's own FY2026 adjusted EBITDA guidance of $2.7-3.0bn implies a 9.4-10.4% margin against the 11.0% Carvana delivered in 2025 — the company is guiding margin down in a year revenue grows 42%, and every forward estimate in this file requires the opposite. Third: 77.4% of operating income is Other gross profit — the gain on sale of originated auto loans — and management says it is compressing on benchmark rates. The equity is a leveraged bet on a lending spread wearing a retailer's clothes, the knowledge base's one useful claim says exactly that, and the sell side does not express its targets in those terms.

What stops it being an Avoid. The volume growth is real and industry-beating; the capacity is paid for and quantified; net debt of $2.6bn against $2.7-3.0bn of guided adjusted EBITDA is not a solvency question; and the stock has already underperformed the index by 32 points over twelve months, so the margin disappointment is at least partly in the price.

Pre-registered UPGRADE conditions — what would take this to Buy — Tactical:

• Adjusted EBITDA per retail unit rising year on year in any quarter, from the current $3,897 against $4,195. This is the single number that would prove the operating-leverage thesis is intact while volume compounds.
• Other GPU stabilising above $2,666, or any disclosure that the customer-rate lever has been used to restore the spread without costing volume.
• Full-year 2026 adjusted EBITDA guidance raised above $3.0bn, which would put the implied margin back above 10.4%.
• A price below roughly $55, which on the base of $68 would restore a payoff ratio above 2:1 without requiring any change to the estimate.
• Insider buying — there is not one open-market purchase in the file.

Pre-registered KILL criteria — what would take this to Avoid:

• Other GPU below $2,400, roughly a further 10% decline in the line that is three-quarters of operating income.
• Retail units failing to rise sequentially in the September quarter, after management explicitly guided that they would.
• Adjusted EBITDA margin below 9.0% in any quarter, or FY2026 adjusted EBITDA guidance cut below $2.7bn.
• Any impairment of the $2,968M deferred tax asset, which is 20.4% of total assets and 58% of total stockholders' equity.
• Total debt rising above roughly $7bn, or any new financing that funds working capital rather than capacity.
• A break below the 52-week low of $56.26 on company-specific news.

Where CVNA fits in the Synthos Framework Portfolio. The consumer sleeve, no position today, with a 1% starter contingent on either a price below $55 or a quarter in which adjusted EBITDA per unit rises. Sizing note: beta is 3.462 — the highest in this batch — so any eventual position should be sized on volatility rather than conviction. Logged as a tracked Synthos call (Hold) as of 2026-08-04 at $68.02, with the fair-value anchors, kill criteria and upgrade conditions all gradeable.

Single biggest risk: interest rates, working through the loan-sale line. Other gross profit is 77.4% of operating income and management states it declined "due to increasing benchmark rates" while the company deliberately "kept customer-facing interest rates stable" to protect volume. A 15% further decline in Other GPU from $2,666 is roughly $316M of annual gross profit on an annualised 790,000 units — approximately 12% of the guided full-year adjusted EBITDA, against an equity trading at 31x forward earnings with a beta of 3.46. The offsetting consideration is that this is a chosen compression rather than an imposed one: Carvana has a customer-rate lever it is currently declining to pull, and pulling it would restore the spread at some cost in volume. That optionality is real. It is also unquantified, and we are not paying 31x forward earnings for a lever whose exercise price the company has not disclosed.

Provenance & disclosures

• Traceability: 2 name-level knowledge-base claims name Carvana out of 52,021 distilled claims, and they disagree (raw hits 15, surviving a case-sensitive entity test 2, name-level 2, discarded 13; breadth 2, net conviction mixed-low). The bullish claim: natalie_brunell, 2025-11-25, conviction 80, horizon fact, speaker: null — "Carvana was a left-for-dead 100x turnaround ($3.50→$400+); 80% of profit comes from finance/interest, not car buying/selling." Its central factual assertion is confirmed by the filings: Other gross profit was $526M of $680M of June-quarter operating income (77.4%) and $1,733M of $1,881M of FY2025 operating income (92.1%). Its price range is stated pre-split and corresponds to roughly $0.70 to $80 after the five-for-one split of 2026-05-08. The bearish claim: mike_green, 2023-08-11, conviction 70, skill 0.9, speaker Michael Green, role independent — "Recently put on a Carvana short... a hard raise for a structurally unprofitable business." It has been decisively wrong — the restructuring and raise completed, and Carvana has reported positive net income every quarter since, including $1,895M consolidated in FY2025 — and it is reported and graded as wrong rather than excluded. Ten of the thirteen discarded text-only hits are macroeconomic claims about used-car PRICES as a CPI component (arthur_hayes, andreas_steno, warren_pies, luke_gromen ×2, jim_bianco, michael_howell, darius_dale), a homograph collision between an inflation input and a retailer. Three are recorded without scoring: on 2025-11-25 natalie_brunell used Carvana as the reference class for Opendoor, for bitcoin miners converting to data centres, and for small-capitalisation crypto — Carvana has become this store's canonical hundred-bagger analogy, which is a crowding observation rather than a signal. No claim carries a management speaker_role. All quotes are verbatim from the stored claim text.
• Data as-of: fundamentals — balance sheet, unit economics, segment gross profit, cash flow and the tax receivable agreement through 2026-06-30, all filing-verified from the 10-Q filed 2026-07-29 and the 8-K filed the same day carrying the complete shareholder letter and financial tables as Exhibit 99.1; annual figures for FY2025, FY2024 and FY2023 from the 10-K filed 2026-02-18, including the adjusted-EBITDA reconciliation and the per-unit gross-profit tables · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873598 = 2026-08-04T19:59:58Z ($68.02, +2.86%; 50-DMA $66.80; 200-DMA $72.28; RSI 45.9; MACD MINUS 0.62; beta 3.462) · knowledge-base claims 2026-08-04. Carvana's fiscal year is the calendar year. All figures come from the Synthos vendor data file for CVNA or from the SEC filings in the CVNA archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-18 (fiscal 2025 — the source of the adjusted-EBITDA reconciliation, the FY2025 income statement showing $2,250M of "Other expense, net", the per-unit tables and the pre-split EPS series); 10-Q filed 2026-04-29 (March 2026 quarter); 10-Q filed 2026-07-29 (June 2026 quarter — the source of the tax receivable agreement balance, the share counts, the Up-C structure description and the stock-split note); 8-K filed 2026-04-29 (first-quarter shareholder letter); 8-K filed 2026-05-06, disclosing the five-for-one forward stock split approved 2026-05-05 and effective 2026-05-07/08, together with the annual-meeting vote totals that quantify the Garcia family's control; 8-K filed 2026-07-29, carrying the complete second-quarter shareholder letter with the per-unit gross-profit tables, the adjusted-EBITDA reconciliation, the full-year guidance and the fully-converted share count. All carry preserved [TABLE] statement data.
• Where the filings contradicted or corrected the vendor (detailed in Section 7): the $2,130M tax receivable agreement liability, of which $1,645M is owed to related parties, absent from totalDebt, netDebt and enterpriseValueTTM — the largest single omission in this file; the fully-converted share count of 1.127 billion disclosed by the company against the 1,096,835,667 the vendor's market capitalisation implies, a 2.7% difference; inc_a.ebitda of MINUS $110M for FY2025 against a company-reported adjusted EBITDA of $2,237M, traceable to $2,250M of "Other expense, net" below operating income, and invalidating ebitdaMarginTTM (0.567%), evToEBITDATTM (546.5x) and netDebtToEBITDATTM (21.06x); inc_a FY2024 gross profit of $2,711M against the filing's $2,876M, understated by exactly $165M — precisely the FY2024 "Depreciation and amortization expense in SG&A expenses" line, which the vendor moved into cost of revenue, with the identical $165M error mirrored in SG&A and a further $190M discrepancy in FY2023 SG&A, while FY2025 is exactly right on all four lines; totalDebt of $5,521M against the filing's $5,115M, inflated by $406M which the balance sheet captions "Operating lease liabilities, excluding current portion" and which the vendor relabels capitalLeaseObligations; enterpriseValueTTM, which carries that lease inflation forward, omits the tax receivable agreement and uses the lower share count, leaving it 5.4% understated against a rebuilt $81,779M; seg_prod, which drops the "Wholesale sales and revenues" line after FY2022 and leaves $4,052M of FY2025 revenue unaccounted for; and seg_geo, which is empty. Where vendor and filing AGREED — recorded, because a clean check is a finding: capitalExpenditure is exactly correct in all three years ($147M / $91M / $87M for FY2025 / FY2024 / FY2023), and although it trips the "capex below 15% of D&A" suspicion signature at 53% of depreciation and 0.72% of revenue, the filings show it is genuine — property and equipment net rose only $45M in a half-year of 38% unit growth, because the ADESA estate acquired in 2022 already supports approximately 1.5 million units of annual capacity; freeCashFlowYieldTTM of 1.245% is arithmetically right (recomputed $929M on $74,607M) though uninformative for a business whose operating cash flow absorbs $833M of inventory growth and $9,005M of loan originations in six months, and it is not used here; and the five-for-one stock split is correctly applied to every historical EPS and share-count figure, verified against the 10-K's pre-split $8.45 / 224,277 thousand for FY2025.
• Basis note: analyst estimates are on GAAP diluted Class A earnings per share, verified against earn_cal, whose epsActual values ($0.42 for Q2 2026, $0.34 for Q1 2026) match the 10-Q exactly. Class A EPS is struck on 739,960 thousand diluted shares against a fully-converted 1.127 billion, and captures 60.4% of consolidated net income because Carvana Co. bears corporate tax that the pass-through LLC Units do not — so as-converted earnings per share is approximately 7-10% below the reported figure. All multiples in this dive use the reported basis, which is the basis the market and the sell side use, with the overhang stated. est.ebitdaAvg and est.ebitAvg are rejected outright: ebitAvg is negative in all eight years present while ebitdaAvg is positive and implausibly small (FY2027: $313M on $36.2bn of revenue), and the FY2023 row carries an epsAvg of $6.205 against a reported split-adjusted actual of $0.15. All EBITDA figures come from the company's own reconciliation or its stated guidance.
• Estimate coverage: 8 analysts on FY2026 Class A EPS, 11 on FY2027 — the anchor for all three fair values, with an FY2027 range of $1.997 to $2.489 (24.6%) — and 6 on FY2028, 4 on FY2029, 2 on FY2030. The FY2029 mean of $2.590 is BELOW the FY2028 mean of $2.753 on 12.3% higher revenue, which is incoherent; FY2029 and FY2030 are excluded from every conclusion.
• Peer note: the vendor peer set — Airbnb, Coupang, General Motors, Hilton, JD.com, Marriott, O'Reilly Automotive, Ferrari, Royal Caribbean, Sea Limited — contains no used-vehicle retailer at all. CarMax, AutoNation, Lithia, Penske and Sonic are absent, as is Cars.com. The set appears to be assembled on beta and growth rather than on business model; two of its ten members (Hilton, Marriott) are hotel franchisors and one (Ferrari) is a luxury manufacturer. No peer-multiple comparison is drawn.
• Fair-value caveat: the $38 / $68 / $100 anchors are multiples of the FY2027 consensus Class A diluted EPS distribution — 19.0x the low of $1.997, 30.9x the mean of $2.198, and 40.2x the high of $2.489 — each cross-checked on EV / adjusted EBITDA against a rebuilt enterprise value of $81,779M and against the company's own FY2026 adjusted EBITDA guidance of $2.7-3.0bn. Stated arithmetic, not a discounted cash flow. The base is sensitivity-disclosed: 24x gives $52.8 and 38x gives $83.5. The base assumes the forward multiple is unchanged, there is no shareholder yield to add, and the resulting fair value equals the closing price — which is the entire content of the verdict.
• Guidance note: the full-year 2026 adjusted EBITDA range ($2.7-3.0bn), the sequential unit-growth expectation for the third quarter, the 1.5-million-unit built-out capacity and 3-million-unit land figures, the early-2027 production date for the first full ADESA buildout, and the "3 million cars per year and 13.5% Adjusted EBITDA margin by 2030 to 2035" destination are management statements from the July 2026 shareholder letter, half-weighted under the standing house rule. The audited and reviewed figures — revenue, units, gross profit per unit, SG&A per unit, adjusted EBITDA, the balance sheet and the tax receivable agreement — are not.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.