Carvana's Record Q2 Isn't a Demand Story — It's a Subprime Credit Story

On July 29, 2026 Carvana reported its best quarter ever: $7.376B in revenue (+52% YoY), $513M in net income (+67%), $769M in adjusted EBITDA, and 197,325 retail units sold (+38%). Full-year adjusted EBITDA guidance was held at $2.7B–$3.0B. The headline read like a clean execution story. The market read it differently — shares fell ~20% in after-hours trading before partially recovering, then sat down roughly 10% before management's call.

On the earnings call, CEO Ernie Garcia broke the GPU decline into two pieces: a $105 drop in retail GPU (largely lapping a ~$100 tariff benefit), and a $192 drop in Other GPU, the line that captures financing income. The cause of the second drop is the whole story: Carvana passed back more than 100 basis points of rate to its customers against a backdrop of higher benchmark rates.

That trade-off — give up per-unit margin, buy more units, hold the credit box — is what makes this a credit-cycle story rather than a demand one. A retailer chasing share by cutting prices is a textbook margin trap; a lender buying receivables on terms it can resell is a credit-cycle bet. Carvana, post-Bridgecrest integration and ~$6B of rated securitization capacity, is now structurally the second one.

The traditional 'law' — new-car supply drives used-car prices, which drive loan performance — is the single most useful rule of thumb in the auto finance stack. In June 2026 that law broke in both directions: the Manheim Used Vehicle Value Index climbed to 212.9 (+2.1% YoY, +0.1% MoM), with retail used-vehicle listing prices back near $27,000. At the same time, subprime 60+ day auto loan delinquencies sat at 5.49% in May 2026 — the third-highest May on record going back to 1994, and only ~140 bps below the 32-year peak of 6.9% set in January 2026.

Ally's Q2 2026 print, reported July 21, 2026, is the cleanest cross-check on the Carvana thesis. Ally received a record 4.6 million retail credit applications in the quarter (+17% YoY), originated $13.3B in retail auto loans (+21% YoY), and — most importantly — used auto loans made up 63% of those originations (up from 61% a year earlier), with used originations +23.9% YoY. The auto segment generated $410M of pre-tax income on $1.4B of total net revenue, and Ally booked a $442M provision expense — $55M higher YoY — to fund CECL reserves against portfolio growth.

The critical detail is the divergence between credit quality (Ally's charge-offs are improving, S-tier mix is rising) and credit volume (used originations +24% YoY, record applications). That's exactly the bifurcation that makes Carvana's record print possible: there is a deep pool of subprime demand that Ally won't touch because of FICO discipline, and Carvana — via Bridgecrest servicing and ~$6B of rated non-prime securitization capacity — is the marginal buyer of it.

Carvana has completed 40 securitizations to date, including 14 under its non-prime shelf, with ~$6.0B of receivables purchase capacity through October 2026 (per S&P on the 2026-P2 deal). The Fitch presale on Carvana 2026-P2 flagged the pool as '96.5% extended-term (61+ month) loans' — exactly the high-LTV, long-duration paper that defines the current subprime cycle. Bridgecrest, Carvana's servicer for 30+ years, originates prime, non-prime, and subprime loans on Carvana's behalf and is rated by KBRA and Fitch on each transaction.

The credit-market signal is the one to watch, not Carvana's reported EBITDA. Subprime ABS losses at 8.99% annualized are still inside stress-case ranges but well above 2022 vintage norms; KBRA upgrading Carvana deals while the on-bank subprime 60+ rate sits near 5.5% is a credit-spread call, not a fundamentals call. If 60+ DPD rolls over back toward the 6.9% January peak, the $6B of forward purchase capacity is the first thing that closes.

Carvana trades at ~$68B market cap on TTM revenue of $25.1B — that's an EV/Sales of ~2.85x and an EV/EBITDA TTM multiple distorted by tax-accounting items but north of 30x. The market is paying for the 2030–2035 target (3M units × 13.5% EBITDA margin ≈ $5–6B EBITDA at scale) more than for the 2026 print. On guide midpoint of $2.85B 2026 EBITDA, the stock is priced at roughly 24x forward EBITDA — rich versus Ally (~10x P/E, sub-prime auto comps typically single-digit P/E) but consistent with a 25% unit-growth-plus-300bps-margin-expansion story.

Revenue nearly doubled from FY 2024 ($13.7B) to TTM Q2 2026 ($25.1B), and net income flipped from $210M to $1.57B. But EBITDA margin slipped 200 bps YoY at the Q2 level — the same GPU dynamic, in margin form. The stock's valuation only works if EBITDA margin re-expands toward 13.5%; the 2030–2035 timeframe means investors are underwriting a 5–10 year view, which is why a single quarter's GPU compression matters less than the trend.

Three operating decisions from the Q2 letter tell you everything about how management is positioning the next 12 months. First, Carvana is leaning into volume over GPU — units +38%, GPU -5.5% — to compound the loan book while subprime credit is still bid. Second, it's choosing to pass rate cuts through to the customer (>$500 of Other GPU headwind avoided) rather than retain financing margin, which signals confidence in cost-of-funds continuing to drop. Third, it's reiterating the 2030–2035 target — 3M units at 13.5% EBITDA margin — without revising the path. The implicit message: the credit cycle will still be friendly in 2027.

Strip away the used-car-retail narrative and Carvana's Q2 2026 looks like the income statement of a specialty consumer lender that happens to retail cars as an origination channel. Net interest income (i.e. the gap between what Carvana earns on its loan book and what it pays to fund it) is the lever that drives Other GPU. The 'record profit' framing comes from unit growth + cost-of-funds compression + a benign subprime delinquency backdrop — three legs of a credit-cycle stool, not a demand story.

The fact that the Manheim index is up 2.1% YoY at the same time 60+ subprime delinquencies sit near a 32-year high tells you used-car prices are being supported by lender balance sheets, not buyer paychecks. That's the same configuration that broke Carvana in 2022 — when used prices fell, the loan book became the constraint, not the opportunity. The 2026 print looks nothing like 2022, but the mechanism that breaks it is identical. The investor question isn't whether Carvana can sell 3M cars — it's how long Ally, the ABS market, and Bridgecrest will keep underwriting the marginal subprime buyer.

The actionable read: own the credit-cycle exposure through Carvana only if you can underwrite the subprime delinquency trajectory. Otherwise, the cleanest cross-asset bet is to fade Carvana against Ally — Ally is buying the same borrowers' loans at prime credit quality, while Carvana is buying the residual non-prime paper Ally won't touch. If delinquencies normalize down, both work; if they spike, Ally's prime book holds up while Carvana's Other GPU breaks. The 20% after-hours drawdown is the market pricing this risk, not the GPU decline.