Revenue Records, Margin Slips, EPS Misses — First Real Test of the Sustainability Question; Maintaining Hold
Initial Read. Top-line momentum stayed at the front of the peer set, but every margin and earnings line came in softer than the post-Q2 trajectory implied; the EPS miss + softer Q4 unit guide is the first time this year the print has not cleanly out-run the multiple.
Key Takeaways
- Revenue beat, EPS missed. Retail units 155,941 (+44% YoY); revenue $5.65B (+55%) crushed the ~$5.08B consensus; adjusted EBITDA $637M beat the ~$600M expectation. But EPS came in at $1.03 vs. ~$1.33 consensus — a meaningful miss. Adj EBITDA margin slipped to 11.3% from 11.7% Y/Y on retail GPU pressure (-$77 YoY) and a $139/unit step-up in advertising.
- FY25 guide raised, but Q4 unit guide underwhelmed. Adj EBITDA now expected at-or-above the high end of $2.0–2.2B (effectively ~$2.2B+). Q4 retail units guided to >150K — below the Street’s expectation for sequential growth and against Carvana’s own multi-year history of Q3-to-Q4 unit growth. Mark Jenkins explicitly framed this as “variability in shape” rather than an underlying slowdown.
- Loan-sale platform got materially larger. Three new agreements signed in October totaling up to $14B of loan principal sales: Ally upsized to $6B through Oct 2027 (from $4B through Apr 2026); two new partners at $4B each through Oct/Dec 2027. This is a structural validation of the finance platform’s asset quality — the buyer base is widening, not narrowing — and lengthens the cost-of-funds tailwind.
- Balance sheet inflection: leverage at 1.5x and shrinking. Q3 retired the remaining $559M of 2028 senior secured notes, primarily via $539M of ATM equity issuance. Combined with a subsequent $98M of 2025 senior unsecured notes retired post-quarter, total corporate debt reduced by $1.2B in 2024+2025. Net debt to TTM adj EBITDA now 1.5x — CVNA’s strongest financial position ever — with the company explicitly targeting investment-grade credit ratios.
- Stock fell -9.7% next session. The print was an “everything-record-but-EPS” quarter, and the market read the Q4 unit guide as the binding signal. The reaction is the first time this year a strong CVNA print has not produced a strong CVNA tape, which itself is a thesis-relevant data point.
- Rating: Maintaining Hold. The franchise is intact and the long-term thesis is unchanged; Q3 in fact reinforces several structural items (loan-sale buyer base, deleveraging, ADESA integrations now at 15). What it does not do is move our valuation read — the GPU normalization, advertising step-up and Q4 unit deceleration are exactly the items we flagged at initiation as the catalysts that could compress the multiple. We hold pending Q4’s read on (a) whether retail GPU stabilizes after the tariff-comp lap and (b) whether unit growth re-accelerates into 2026 or settles into a 25–35% range.
Results vs. Consensus
The mixed scorecard is what matters here: a clean revenue + EBITDA-dollar beat, but margin compression and an EPS miss that the post-Q2 trajectory did not telegraph.
| Metric | Q3 2025 Actual | Consensus | Y/Y | Beat/Miss |
|---|---|---|---|---|
| Retail Units Sold | 155,941 | n/a (qual.) | +44% | Beat — new company record |
| Revenue | $5.647B | ~$5.08B | +55% | Beat by ~$570M (~11%) |
| Diluted EPS | $1.03 | ~$1.33 | n/m | Miss ~$0.30 |
| Adjusted EBITDA | $637M | ~$600M | +48% | Record dollars; missed margin expectations |
| Adj. EBITDA Margin | 11.3% | ~12.0%+ implied | -40bps | First Y/Y margin decline in several quarters |
| GAAP Operating Income | $552M | n/a (qual.) | +64% | Record; 9.8% op margin (+60bps Y/Y) |
| GAAP Net Income | $263M | n/a (qual.) | +78% | 4.7% net margin; record dollars |
| Non-GAAP Retail GPU | -$77 YoY | n/a | -$77 | Higher retail depreciation rates |
| Q4 Retail Unit Guide | >150,000 | Sequential growth expected | n/a | Implied flat-to-down sequentially |
Quality of Beat / Miss
- Revenue: The +55% Y/Y beat was magnified by a contract change that put marketplace-partner inventory onto Carvana’s balance sheet (gross-revenue accounting vs. prior net treatment) plus higher ASPs. Underlying organic top-line growth is closer to the +44% retail-unit pace; the +55% headline is partly an accounting optic.
- Margins: Adj EBITDA margin compressed 40bps Y/Y (11.3% vs. 11.7%) for two reasons: (1) retail GPU -$77 on higher industry depreciation rates and a soft Q3 depreciation lap, vs. an unusually strong Q3 2024 figure boosted by additional loan sales; (2) advertising +$139/retail unit, the previewed Q3 step-up.
- EPS miss: The $0.30 gap to consensus is explained by a combination of below-the-line items (higher interest expense on retired notes, share count from the ATM issuance) and modestly softer flow-through from the GPU/advertising piece. None of the components individually are alarming — collectively they explain why the model expectations were too aggressive.
Operational Performance
| Driver | Q3 2025 | YoY Change | Notable |
|---|---|---|---|
| Retail Units Sold | 155,941 | +44% | 3rd consecutive quarter at 40%+ growth |
| Revenue | $5.647B | +55% | Crossed $20B annualized run-rate for the first time |
| ADESA sites integrated | 15 | +3 vs. Q2 | Cadence holding at ~3 per quarter |
| Same/next-day delivery (Phoenix test) | 40% of customers | vs. 10% nationwide | ~2,500 cars available for same-day delivery in Phoenix on any given day |
| Self-serve completion (retail) | >30% of customers | n/a (new disclosure) | Customers complete entire purchase without speaking to advocate |
| Self-serve completion (sell-to-Carvana) | >60% of customers | n/a (new disclosure) | The automation flywheel is showing measurable scale |
| Non-GAAP Retail GPU | -$77 YoY | Higher retail depreciation | Q3 2025 was a soft retail depreciation quarter |
| Non-GAAP Wholesale GPU | -$168 YoY | Retail outgrew wholesale + higher wholesale depreciation | Wholesale-to-retail ratio compressed but stable |
| Non-GAAP Other GPU | +$63 YoY | Cost-of-funds + finance/VSC attach | Record level for Other GPU |
| Non-GAAP SG&A per retail unit | -$319 YoY | -$96 ops, -$314 overhead, +$139 advertising | Overhead leverage doing the heavy lifting |
| Net debt to TTM adj EBITDA | 1.5x | Improving | Strongest financial position in company history |
| Total loan-sale capacity | Up to $14B new commitments | vs. ~$4B Ally only at start of year | 3 new agreements through 2027 |
Key Topics & Management Commentary
Overall Management Tone: Confident on the long-term framework but visibly more measured than Q2 on the near-term cadence. Garcia leaned heavily on the operational-improvement examples (Phoenix same-day, AI agents writing code) to anchor the call’s narrative away from the margin-and-guide softness. Jenkins was more forthcoming than usual on the seasonal quirks of Q3 vs. Q4 retail and wholesale depreciation, which we read as a tactical effort to pre-empt over-reading the Q4 unit guide.
The GPU Walk: Why Margin Slipped
Jenkins broke down each of the three GPU lines in prepared remarks. Retail GPU -$77 on higher retail depreciation rates; wholesale GPU -$168 on wholesale depreciation plus retail growth outpacing wholesale; other GPU +$63 on cost-of-funds and attach-rate improvements offset by a difficult Q3 2024 lap (additional loan sales).
“Looking ahead to Q4, we expect sequential changes in retail GPU, wholesale GPU and other GPU in a similar range to last year, with the latter primarily reflecting sharing fundamental gains with customers through lower interest rates.”
— Mark Jenkins, CFO
The “sharing fundamental gains with customers” framing — explicitly that Q4 other GPU should look like Q4 2024 rather than Q3 2025 — is the first time this year management has explicitly traded near-term reported GPU for customer pricing. Strategically defensible (it’s the same playbook successful consumer brands run); modeling-wise it caps Q4 EBITDA-margin upside.
Read-through: Q3 was the lap of the unusually strong Q3 2024 retail GPU figure, and Carvana didn’t have a tariff-window tailwind to fall back on. The structural piece of the GPU story (recon + inbound transport efficiency) remains; the tougher comps simply made the headline Y/Y number negative for the first time.
Loan-Sale Platform: $14B Of New Commitments Validates The Asset Quality
Sharon Zackfia (William Blair) opened Q&A on subprime concerns; Jenkins took the chance to draw the cleanest distinction the company has yet made between its 2024/2025 vintage and the 2022/2023 industry-wide cohorts that have underperformed.
“Our 2024 and 2025 loan originations are performing extremely well, both in an absolute sense and relative to industry comparables. ... I think most of the industry, ourselves included, tightened credit in late 2023. We certainly did, and we’ve maintained that tightness here through where we are today in 2025.”
— Mark Jenkins, CFO
The October agreements (Ally upsized to $6B; two new partners at $4B each) are the clearest external validation of asset quality. As Jenkins framed it, the move from one-off pool sales to programmatic agreements is “a maturation and a continuation” rather than a structural change.
Assessment: The $14B in new commitments materially extends the cost-of-funds tailwind through 2027 and reduces the operational risk of relying on a single primary partner. We treat this as one of the most thesis-relevant pieces of the print — quietly more important than the headline numbers.
Phoenix Same-Day Delivery: Strategic Optionality, Not Yet Margin Story
Garcia spent significant prepared-remarks airtime on the Phoenix same-day-delivery test (40% of customers; ~2,500 cars available). When pressed by Chris Bottiglieri (BNP Paribas) on what it’s costing in operating expenses, Jenkins explicitly framed it as a technology-and-process investment rather than a cost investment.
“The same-day delivery is really more about a technology investment at this stage and a process investment, making sure that it’s a complex transaction. ... There’s some incremental investment in staffing just to make sure that you have the capacity available to execute same-day delivery. ... But that’s not a very large dollar amount in the grand scheme of things.”
— Mark Jenkins, CFO
Assessment: Same-day delivery is a credible long-term differentiator and a plausible word-of-mouth accelerant; on the financial model it’s probably 6–12 quarters from being a quantifiable tailwind. The strategic option value is real; the near-term P&L impact is not what investors should price.
AI Capabilities: Concrete Examples, Not Just Talk
Garcia’s response to Brian Nagel (Oppenheimer) on AI was unusual for an operational company: he gave specific anecdotes (an “ambient agent” that detected a website bug, investigated the cause, wrote code, and submitted it for human approval — all autonomously) rather than the standard CEO talking points. Combined with the >30% / >60% self-serve completion stats, it suggests the automation flywheel is genuinely scaling.
Assessment: Carvana is structurally well-positioned for an AI tailwind because the entire business was already built on data + automation + vertical integration. The competitive read-through — that traditional auto retailers cannot capture similar AI gains because their data is fragmented across dealers and lenders — was understated but real.
EV Tax-Credit Expiration: Mix Shift, Not Demand Shift
John Colantuoni (Jefferies) raised the federal EV tax credit expiration (CVNA’s EV mix was above industry average). Garcia’s read: aggregate demand basically unchanged, mix shifted away from EVs, system adapted automatically.
Read-through: Modest watch item but not material. Garcia’s deliberate EV-bull framing (“we expect over time that they will make a comeback”) is consistent with not over-reacting to the policy change.
Q4 Unit Guide: “Variability in Shape”
Rajat Gupta (JPMorgan) led on this, noting CVNA had grown units sequentially Q3-to-Q4 in most prior years. Garcia’s response was the most defensive moment of the call:
“I think it’s largely more of the same. ... When you look at the last several years, Q3 to Q4 for us or for other retailers, there’s a decent amount of variability in the shape that you see Q3 to Q4. ... Everything remains the same.”
— Ernest Garcia, CEO
Assessment: The flat-to-down Q3-to-Q4 unit framing is unusual for CVNA and is the principal reason the stock fell 9.7%. Two innocent explanations: (1) management is being conservative on a quarter where wholesale depreciation seasonally compresses GPU and they don’t want to lay out a unit number that requires further GPU give-back; (2) reconditioning capacity has temporary bottlenecks as ADESA-site managers ramp. We accept both as plausible. We do not yet have evidence to assess whether unit growth deceleration is structural or transitory; that is precisely the gating question for the next print.
Guidance & Outlook
| Metric | FY25 / Q4 Outlook | vs. Prior | Reading |
|---|---|---|---|
| FY25 Adj EBITDA | At-or-above $2.2B (high end of $2.0–2.2B) | Effectively raised | Implies Q4 adj EBITDA of ~$435M+ |
| Q4 Retail Units | Above 150,000 | New | Soft vs. Street; flat-to-down sequentially |
| Q4 Retail GPU | Sequential change “similar range to last year” | New (qualitative) | Implies seasonal compression typical for Q4 |
| Q4 Other GPU | “Look more like Q4 2024 than Q3 2025” | New | Lower — deliberate sharing of gains with customers via rate cuts |
| Q4 Advertising | Similar to or slightly higher than Q3 ($139/unit step-up) | Maintained | ~10–15bps potential margin headwind |
Implied Q4 EBITDA math: FY25 at-or-above $2.2B less H1 ~$1.04B less Q3 $637M leaves Q4 at ~$430–480M+ — consistent with retail unit growth in the 25–35% range and GPU seasonality + further advertising spend. The Q4 framework is internally consistent.
Guidance style: Carvana raised the FY25 floor while flagging Q4 unit conservatism. Reading Garcia and Jenkins literally, they want investors to underwrite the FY25 number ($2.2B+) and not over-fit on the Q4-specific seasonal mechanics. Reading them tactically, they had a clean opportunity to raise the unit guide above 150K and chose not to — a tell that ADESA-site execution and reconditioning capacity may be tighter than the Q2 cadence implied.
Analyst Q&A — Notable Exchanges
Q&A skewed toward credit/loan platform mechanics, GPU sustainability, and the Q4 unit framing. Notable threads:
- Sharon Zackfia (William Blair) opened on the topic du jour — subprime loan performance — and got the cleanest distinction yet between CVNA’s 2024/2025 vintages and the 2022/2023 industry cohorts that have underperformed. The $14B new-agreements color was packaged with the answer.
- Marvin Fong (BTIG) probed the sequential operating-expense uptick and Q4 retail GPU framing; Jenkins’ clearest framing of Q3 vs. Q4 depreciation seasonality (Q2 2025 strong tariff-related depreciation; Q3 2025 softer depreciation; Q4 2025 typically higher depreciation).
- Rajat Gupta (JPMorgan) led on the Q4 unit guide. Garcia’s “variability in shape” framing was the call’s most-replayed defensive line. Gupta’s follow-up on ancillary-product penetration drew Garcia’s acknowledgment that “there’s definitely additional opportunity” without quantifying.
- Brian Nagel (Oppenheimer) got the AI-anecdote-rich response — the “ambient agent” framing was the freshest disclosure of the call. Nagel’s follow-up on macro / used-car-demand softness drew the “things look pretty similar at a high level” consistent line.
- John Colantuoni (Jefferies) raised EV tax-credit expiration (mix shift, not aggregate demand shift) and probed the second franchise-dealership acquisition; Garcia explicitly declined to comment on franchise-acquisition strategy as “premature.”
- Chris Bottiglieri (BNP Paribas) drew the Phoenix same-day-delivery cost framing (technology + process investment, not a meaningful cost line) and the deliberate-customer-rate-pass-through framing on Other GPU into Q4.
- Daniela Haigian (Morgan Stanley) asked about competitive threats from new entrants (Amazon framing) and capital intensity to expand beyond the 3M unit goal. Garcia’s answer stayed focused on internal execution rather than gaming out competitor scenarios.
- Jeff Lick (Stephens) drew the most-detailed disclosure of the call on the wholesale-vs.-retail capability matrix at the 74-site network: 41 wholesale-only ADESA sites, 6 retail-only legacy IRCs, 27 dual-capable sites.
- Andrew Boone (Citizens) probed AI scale opportunities; Garcia’s “deepening level of detail of automation” framing was the cleanest articulation of the operating philosophy.
- Michael McGovern (Bank of America) raised the K-shaped economy / lower-income cohort question; Garcia explicitly declined to confirm any cohort-specific weakness.
- Michael Montani (Evercore ISI) got the Q4 advertising and wholesale-GPU sequential framing.
- Chris Pierce (Needham) closed with the 3M unit goal time-line question; Garcia confirmed execution is the primary driver.
What They’re NOT Saying
- No specific Q4 retail GPU dollar guide. The “sequential change similar to last year” framing is qualitative; the implication is meaningful sequential GPU compression but management didn’t put a number on it. The non-disclosure tells us the magnitude is probably toward the higher end of recent Q4 seasonal patterns.
- No quantified Q4 advertising step-up. “Similar to or slightly higher than Q3” on a $/unit basis when units are guided softer means total dollars are roughly flat or up modestly. The lack of a specific number signals flexibility, not commitment.
- No 2026 framework. Same as Q2 — no FY26 unit or EBITDA goalposts even as the long-term target was reaffirmed. By Q4 2025 / Q1 2026 we’d expect an FY26 framework; its continued absence is a notable choice.
- No update on the second franchise dealership acquisition. Garcia explicitly declined to comment despite Colantuoni’s direct question. The pattern (two acquisitions in quick succession + management deflection) is consistent with this being a small-but-deliberate strategic option being kept out of the model. Watch for it to graduate from “too early to comment” to a stated framework over the next 2–3 quarters.
- No detailed cohort data for older markets. Same as Q2. Garcia’s “all the trends remain” framing in response to Gupta on Atlanta cohorts is consistent. The steady non-disclosure of mature-market unit economics is increasingly conspicuous.
- No commentary on ATM equity issuance pacing forward. Q3’s $539M ATM issuance to retire the 2028 notes was a clean deleveraging move; whether the ATM remains active in Q4 (and at what cadence) was not addressed. Investors who model dilution should treat further issuance as possible.
Market Reaction
- Pre-print setup: The stock had run hard YTD on the back of the Q2 print and consistent peer-set outperformance. Expectations entering the print were elevated — consensus EPS at $1.33, revenue at $5.08B, and the implicit assumption of sequential margin expansion.
- Initial reaction: The print landed after the close on Oct 29. Shares fell ~9.7% in the morning session on Oct 30 as the market read the EPS miss + soft Q4 unit guide as the binding signals.
The reaction is the first time this year a strong CVNA print has produced a sharply negative tape, and it confirms our Q2 framing on valuation asymmetry: at the post-Q2 multiple, anything other than a clean beat-and-raise had downside skew. The franchise didn’t deteriorate in Q3 — the multiple did. That’s the rating gate.
Street Perspective
Debate: Is the GPU compression structural or transitory?
Bull view: The Q3 2024 GPU comp was unusually elevated (additional loan sales generated a one-time ~$100/unit benefit that doesn’t recur). Looking at sequential trends and underlying drivers (recon, inbound transport, ADESA leverage), GPU is structurally still expanding. Q4 will be the easiest comp of the next four quarters.
Bear view: Retail GPU has now declined Y/Y, against the narrative of structurally expanding margins. Other GPU is being deliberately given back to customers via lower rates. Wholesale GPU is compressing as retail outgrows wholesale. The all-in result is a model where headline GPU could be flat-to-down for several quarters even as units grow.
Our take: The GPU normalization in Q3 was within the realm of seasonality + tough comp + deliberate customer pass-through. Q4 will tell us more. We’d expect Q4 retail GPU down sequentially and Y/Y, and Q1 2026 to be the first clean lap of the post-tariff environment. The bull case requires Q1 2026 retail GPU to print at or near Q1 2025 levels — achievable, but not yet evidenced.
Debate: Does the 40% unit-growth framework hold or does it normalize toward 25–30%?
Bull view: Three consecutive quarters of 40%+ growth, ADESA capacity ramping, advertising re-investing, Phoenix proof-of-concept on same-day delivery. Long-term goal of 3M units / 13.5% margin requires roughly 40% / 20% CAGR — Q3 print is at the fast end. The Q4 guide is a one-quarter tactical adjustment, not a structural change.
Bear view: Q4 unit guide of >150K implies effectively flat sequentially — the first time this year unit growth has visibly decelerated. Reconditioning is the binding capacity constraint per Garcia’s own framing. Tougher comps + the tariff-window unwind in 2026 mean 25–35% looks like the more realistic FY26 number.
Our take: We move toward a 30–35% FY26 retail-unit growth framework. The 40% number requires both ADESA recon-site execution to accelerate and macro to remain supportive. The Q4 unit guide is consistent with this rebasing — and the multiple should follow.
Debate: How much of the loan-sale platform expansion should the multiple credit?
Bull view: $14B of new commitments through 2027 + cost-of-funds tailwind + asset-quality validation = a multi-year structural advantage that makes Other GPU more durable and reduces the cyclical risk of the financing line. This is meaningfully more thesis-relevant than the headline EBITDA print.
Bear view: Loan-sale agreements are commercial contracts that can be repriced; the buyer base is widening but the cost-of-funds tailwind eventually mean-reverts as ABS spreads normalize. The strategic optionality is real but its quantitative impact is bounded.
Our take: The $14B commitments are quietly the most thesis-positive disclosure of the print and underwrite a modest extension of the cost-of-funds tailwind into 2026–2027. We do not yet credit it as a multiple-rerating event — but we’ll watch how it scales and whether asset-yield commentary remains consistent.
Model Implications
- FY25 adj EBITDA: Move to at-or-above $2.2B; we model $2.20–2.25B with Q4 ~$435–485M.
- FY25 retail units: ~605–610K (vs. our Q2-anchored 530–560K), reflecting the Q3 acceleration even with the Q4 step-down.
- FY26 retail units: Move to a 28–33% growth framework (vs. consensus that may still be at 35%+); implies ~775–810K.
- FY26 adj EBITDA margin: 11.0–11.5%, with the rate-pass-through to customers offsetting the operating-leverage ramp.
- Net debt: 1.5x TTM at quarter end; trending toward 1.0x by mid-2026 on EBITDA growth even with no further debt retirement. Refinancing optionality continues to improve.
- Share count: Watch for further ATM activity. Q3’s $539M issuance retired senior secured notes — a good trade. Future issuance for general corporate use would be different.
Thesis Scorecard
| Thesis Point | Status | Notes |
|---|---|---|
| Bull #1: Vertically-integrated model is structurally more profitable than peers | Confirmed | 11.3% adj EBITDA margin maintains the ~2x peer-set lead even as it compresses Y/Y |
| Bull #2: ADESA integration is the multi-year cost-takeout / capacity engine | Confirmed | 15 sites integrated; cadence holding; Phoenix same-day-delivery proof-of-concept |
| Bull #3: Vertically-integrated finance platform is structurally advantaged | Confirmed + | $14B in new agreements through 2027 materially extends the cost-of-funds tailwind |
| Bull #4 (NEW): Balance sheet inflecting toward investment-grade | New — Confirmed | 1.5x net debt / TTM adj EBITDA; $1.2B of corporate debt retired in 2024+2025 |
| Bear #1: Valuation is stretched relative to operating performance | Active | Market reaction (-9.7%) confirms the multiple was pricing the upside scenario |
| Bear #2: 40% CAGR sustainability is unproven past 12–18 months | Active — Pressuring | Q4 unit guide of >150K implies effectively flat sequentially; growth path under near-term test |
| Bear #3: Reconditioning is the binding operational constraint | Active | ADESA-site management bandwidth and ramp execution likely a Q4 headwind |
| Bear #4: Macro / used-car-cycle exposure | Dormant | Demand stable per management; CVNA outperforming peer-set roughly flat |
Overall: Operating thesis intact and modestly strengthened by the loan-sale-platform expansion and the deleveraging proof-points. Valuation thesis is the rating gate — and the stock’s -9.7% reaction is consistent with our Q2 framing that the post-print multiple was pricing the upside scenario.
Action: Maintaining Hold (initiated at Hold at Q2 2025). The Q3 print does not change our view of the franchise — if anything, it strengthens the structural pieces of the thesis (loan platform, deleveraging, ADESA cadence). What it does is validate the rating gate: at full valuations, mixed prints get re-rated. We continue to wait for either (a) a meaningfully de-rated multiple that re-creates margin of safety or (b) a Q4 print where retail GPU stabilizes and the FY26 framework is articulated. Either path could move us to Outperform; we are not there yet.
Net: Records on top, normalization on the margins, Q4 unit caution under the hood. The franchise is real; the rating gate is the price.