Margin Inflects a Quarter Early, and a 38% Rally Has Already Paid For It

Results vs. Consensus

Q2 2026 Scorecard

DoorDash publishes GAAP diluted EPS only and has no company-defined adjusted EPS, which produces the widest vendor spread of any figure in the print. Three of four consensus providers carried $0.46 to $0.47 for the quarter and one carried $0.50, so the $0.46 result reads as in line to marginally light rather than as the 8% miss the outlier implies. The same dispersion appeared last quarter. Revenue estimates were tightly clustered at $4.32B to $4.35B, so the revenue beat is not in question.

Performance Against Company Guidance

Year-over-Year Comparison

Sequential Comparison

Margins: the line that broke Q1 is the line that carried Q2

Last quarter's central finding was that DoorDash's margin compression was not a spending problem. Every adjusted operating-expense line as a share of GOV was flat or better, and the entire 17 basis points of year-over-year decline arose above the operating-expense lines, in adjusted gross profit per dollar of order value. This quarter that relationship reverses cleanly, and the bridge is worth reading in full.

The sequential bridge is more dramatic, because Q1 was the trough. Adjusted gross profit rose from 6.629% of GOV to 7.304%, a 68 basis point improvement and the largest sequential move in this ratio across the five quarters the company presents. Adjusted sales and marketing cost 8 basis points, adjusted research and development cost 14, and adjusted general and administrative cost 8, for a net 38 basis point improvement in Adjusted EBITDA as a share of GOV.

Assessment: the shape of the improvement matters more than its size. A margin recovery driven by gross profit per order dollar is a recovery in the underlying economics of a delivery, which is the thing the equity story rests on. A margin recovery driven by underspending would have been a timing item. Set against that, the single largest cost movement in the quarter was research and development taking 18 basis points more of GOV, and that is the line carrying the platform rebuild, the autonomy program and the AI investment. The company has now told investors twice that all three get bigger in Q4.

Revenue: the wedge closed, and management called it seasonal

Net Revenue Margin recovered to 13.47% from a five-quarter low of 12.77%, and landed within eight basis points of the year-ago 13.55%. In Q1 the gap between 37% GOV growth and 33% revenue growth was the whole story of the quarter. In Q2 the gap is 36.4% against 35.6%, and it has all but disappeared.

The explanation offered on the call is narrower than the recovery. Asked directly how much of the pickup was Deliveroo mix, advertising or fee changes, the CFO attributed the sequential increase largely to the seasonal pattern in Dasher costs, which are lower in the second quarter than in the first, and then guided the take rate flat in Q3 and lower in Q4 on the reverse of that same seasonality. Advertising was named elsewhere on the call as one of two areas where unit economics ran ahead of plan, but it is again given no dollar value.

Assessment: the honest read is that this quarter neither confirms nor refutes the Q1 concern. If the recovery is seasonal, then the underlying take rate is roughly where it was, and the structural question about consumer-fee reductions and Deliveroo mix returns in Q4 when the seasonality unwinds. What we can say is that the year-over-year comparison stopped deteriorating: 30 basis points of decline in Q1 became eight basis points in Q2, and management is guiding to a Q4 level below Q3 rather than to another leg down. That is containment, not resolution.

EPS and the below-the-line bridge

GAAP net income attributable to common stockholders of $200M is 30% below the year-ago quarter on 36% more revenue, which is the number that produced the initial negative reaction to the print. The bridge shows almost none of it is operating.

The $43M swing in other income is the largest single item and it is not a business result at all. The prior-year quarter carried the fair-value gain on the deal-contingent forward contract entered into in Q2 2025 to hedge the Deliveroo purchase, an instrument that settled on closing in October 2025 and no longer exists. The tax line moved from a $13M benefit to an $8M expense on $207M of pre-tax income, a 3.9% effective rate. Interest income fell 29% on lower average rates.

Income from operations declined $7M while absorbing $136M more depreciation and amortization and $67M more stock-based compensation than a year ago. Roughly $450M of the $1.1B to $1.2B of full-year depreciation and amortization is amortization of acquired intangibles from Deliveroo, Wolt and SevenRooms, which will depress GAAP earnings through at least 2027 irrespective of how the business performs.

Assessment: investors anchoring on the 30% GAAP decline are measuring purchase accounting and a hedge that expired. That said, the alternative anchor is not clean either. Adjusted EBITDA excludes $349M of stock-based compensation in a quarter that produced $200M of GAAP profit, and the honest measure of this business sits between the two. Free cash flow is that measure, and free cash flow is the line that improved most.

Cash: the inflection, and the reversal already disclosed

Operating cash flow of $944M grew 87% and free cash flow of $742M grew 109%. Trailing-twelve-month free cash flow reached $2,139M against $1,752M reported three months ago and $1,713M a year ago, so the trailing series has turned decisively higher after two quarters of decline. This happened while capitalized software and website development costs rose to $141M from $83M, so the improvement is in operations rather than in reduced investment.

Buried in the outlook section is a bullet stating that 2026 year-end merchant-payment timing, measured against the end of 2025, will reduce reported 2026 free cash flow by $700M to $800M. First-half free cash flow was $1,162M. A $750M timing reversal landing in the fourth quarter takes the full-year reported figure to roughly $1.9B, against an underlying run rate that would otherwise produce something closer to $2.6B.

Assessment: this is the most consequential number in the release that received zero attention. It does not change the economics of the business, and a working-capital timing item reverses again the following year. It does change what the 2026 reported free-cash-flow figure will look like, and any investor underwriting a cash-flow multiple off the current trailing series is going to be surprised in February. The disclosure itself is to management's credit; the placement, as the third bullet in a list following stock compensation and depreciation, is not.

Segment Performance

DoorDash reports as a single operating segment and publishes no segment profitability. The two disclosures that carry analytical weight are the geographic revenue split in the quarterly filing and the Deliveroo revenue carve-out in the release.

Revenue by Geography

Deliveroo Carve-Out

United States

U.S. revenue of $3,457M grew 22.2%, a five-point acceleration from the 17% posted in Q1 and the single most important number in the filing. The Q1 recap flagged 17% domestic growth on a fee reduction as respectable but the slowest-growing part of a business the market values on 30%-plus consolidated GOV growth, and specifically as the part that would still be there after the Deliveroo comparison laps in Q4. That part just accelerated.

Management attributes the acceleration to membership. U.S. restaurant GOV growth accelerated from Q1 to Q2, and the release makes the point in a way that invites checking: year-over-year growth in the U.S. restaurant category in Q2 2026 was, on the company's account, roughly the same rate the category grew for the full year 2022. Grocery and retail grew faster than restaurants while improving unit economics significantly.

"In the 12 months through Q2 2026, we increased the number of U.S. paid DashPass members by more than we did over the previous 24 months combined, a sign that consumers are finding growing value in our program." — DoorDash Q2 2026 shareholder release

Assessment: a mature domestic business reaccelerating five points while its take rate recovers is a genuinely good outcome, and it is the strongest evidence yet that the Q1 fee reduction bought occasions rather than merely surrendering price. The caveat is the same one that applies to everything DoorDash says about membership: the member count is not published, so the mechanism cannot be verified. What can be verified is the revenue line, and the revenue line moved.

International

International revenue of $997M more than doubled as Deliveroo laps into the base only from October 2, 2025, so the reported growth rate remains an acquisition artifact for one more quarter. Underneath it, the disclosed direction is uniformly positive: Deliveroo year-over-year growth in monthly active users and Total Orders accelerated again, on management's account to the highest rate in a couple of years, while exceeding the profit expectations the company held entering the quarter. Wolt+ recorded a record quarter for paid subscriber growth. The filing again notes that no single country outside the United States reached 10% of consolidated revenue, so no individual market can be sized.

"When you look at the actual performance of Deliveroo itself, to your point, volume growth or MAU growth or subscription growth, actually, when you look at it on a year-over-year basis, it's been the highest that we've seen in the last couple of years. At the same point, to your second part of your question, we've increased the unit economics as well." — Ravi Inukonda, Chief Financial Officer

Assessment: three consecutive quarters of accelerating growth at an asset the market spent 2025 worrying the company overpaid for is the clearest vindication available on the integration. The single-country disclosure ceiling means the entire international story remains a management assertion supported by one consolidated revenue line. Note also what changed in the paperwork: the release no longer restates the roughly $200M Deliveroo contribution to 2026 Adjusted EBITDA that it carried in each of the two prior quarters, and the only reaffirmation is a verbal reference on the call to meeting the profitability targets set out in the last letter.

Unit economics per dollar of order value

Assessment: four of the five per-unit measures improved year over year, and all five improved sequentially. In Q1 the equivalent table showed none improving. That is the cleanest possible statement of what changed in this quarter. The exception is GAAP net income per GOV dollar, which is the purchase-accounting and hedge-expiry effect already bridged above and not an operating signal.

Key Topics & Management Commentary

Overall Management Tone: assured on results and noticeably less forthcoming on the forward frame than a quarter ago. Management answered every operating question with detail and volunteered more product disclosure than usual, then declined to reconfirm the full-year margin commitment when asked for it directly, redirecting to the one-quarter guide instead. There is no defensiveness in the posture and no evasion in the operating answers; the reticence is concentrated entirely on the annual outlook, which is the reverse of the useful pattern.

Note also the format, which is unchanged. DoorDash publishes a shareholder-letter-style release in place of prepared remarks and opens the call directly into questions after the safe-harbor read, so the entirety of management commentary below is responsive rather than scripted. Fifteen analysts asked questions.

1. Where the beat actually came from, in management's own words

Adjusted EBITDA cleared the guide high end by $44M, an unusually wide overshoot for a company that has spent the last several quarters landing inside its ranges. The question of whether that reflected structural improvement or deferred spending was put directly, and the answer was unusually specific about mechanism and unusually candid about intent.

"More specifically, what we saw in the quarter was the unit economic improvement came in ahead of our expectations, specifically in a couple of areas. Ads was one of them, subtotal was the other one. But in the second half of the quarter, those unit economics came in slightly ahead of what we had expected. In addition, ROO beat our own internal volume expectations, ROO's contribution profit positive." — Ravi Inukonda, Chief Financial Officer

The follow-through is the more revealing half.

"There are going to be times like in Q2, where the EBITDA beat comes in later in the quarter. In those times, we just don't have enough time to reinvest back in the business, especially at the levels of efficiency that we desire. Q2 was one of those quarters where we're happy to drop it to the bottom line." — Ravi Inukonda, Chief Financial Officer

Assessment: read plainly, that is management saying the beat happened because upside arrived too late in the quarter to be spent, and that under normal timing it would have been reinvested. The operating improvement is real and the sources named, advertising and order subtotal, are the right ones. But the company has just told investors that its policy is to convert efficiency into investment rather than into margin, which caps how much a single quarter's margin beat should shift a forward estimate. We take the beat as evidence about the direction of unit economics and not as evidence about the run rate.

2. The full-year margin frame was withdrawn without comment

The Q1 release carried an explicit commitment: full-year 2026 Adjusted EBITDA as a percent of Marketplace GOV to increase slightly compared to 2025, excluding Deliveroo from both periods. That sentence is absent from the Q2 release. In its place is a shape statement covering only the second half, and it points down in Q4.

"In H2 2026, we expect Adjusted EBITDA as a percentage of Marketplace GOV to follow a similar pattern to H2 2025, with a Q/Q increase in Q3 2026 followed by a Q/Q decline in Q4 2026. We expect the Q/Q decline in Adjusted EBITDA as a percentage of Marketplace GOV in Q4 2026 to be driven primarily by a seasonal increase in Dasher costs, an annual increase in insurance expenses, and an increase in investments in our global technology platform and our autonomy initiatives, among other areas." — DoorDash Q2 2026 shareholder release

Asked on the call whether the company is still tracking to a higher year-over-year Adjusted EBITDA margin for 2026 excluding Deliveroo, the CFO answered about landing inside the Q3 range and did not address the annual question.

Assessment: a company that has just beaten its own guide by $44M and expanded its margin year over year for the first time in our coverage had every reason to reaffirm its annual frame, and did not. In 2025 the sequential margin step from Q3 to Q4 was a decline of roughly 39 basis points. Applying the same shape to the Q3 guide midpoint of 3.06% of GOV puts Q4 near 2.68%, which is enough to hold the full year at roughly 2.7%, in line with 2025 rather than above it. That may be exactly why the sentence came out. Withdrawing a commitment in a strong quarter is the disclosure event of this release.

3. Research and development is now the margin drag

GAAP research and development rose 52% year over year to $535M and 34% sequentially. On an adjusted basis, which strips stock compensation and allocated overhead, the growth was 66% year over year. As a share of GOV, adjusted research and development rose to 1.016% from 0.833%, consuming 18 of the 23 basis points of gross-profit improvement the quarter generated.

The quarterly filing gives the composition: personnel-related compensation up $187M and third-party software up $70M, partially offset by $78M more capitalized into software and website development costs. Stock compensation inside research and development alone was $189M, up from $141M a year ago and $113M in Q1.

Assessment: this is the cost line that carries the platform rebuild, autonomy and AI simultaneously, and it is the only line in the model that got materially worse per unit of volume. It is also the line management has now flagged twice as increasing in Q4. An 18 basis point drag on a 2.76% margin is not a rounding error; it is roughly two-thirds of the year-over-year gain the rest of the business produced. Watching this ratio quarter to quarter is the cheapest available proxy for whether the investment cycle is scaling with the business or ahead of it.

4. The platform rebuild now completes in the first half of 2027

Last quarter the CFO said duplicate costs for running three technology stacks in parallel would mostly run through 2026, with some portion bleeding into early 2027. The Q2 release puts a date on completion rather than on cost.

"We expect to begin seeing benefits from our new global technology platform once it is fully rolled out, which we currently expect to be in the first half of 2027." — DoorDash Q2 2026 shareholder release

Asked for quantifiable benefits from the modules already live, the CEO gave examples and an analogy rather than a number.

"The theme really is what we're trying to do and why there's a thesis for a return here is because you're taking the best-of-breed feature and literally offering it to all of our 41 markets. And so this is not like taking one stack and then just like copy and pasting everything into all of the brands. It's actually literally taking the best of each and then putting it into a completely new engine, and then running -- and it's like building a new company." — Tony Xu, Co-Founder, Chair and CEO

Assessment: "benefits begin once it is fully rolled out in the first half of 2027" is a later and softer statement than "duplicate costs bleed out in early 2027," and it means the redundant run cost persists through essentially all of 2026 and into 2027 while the return arrives after that. The program is still described without a dollar figure, four quarters after it was first called the largest component of the investment cycle. Conversion wins from search and automation in support flows are real examples; they are not a return calculation. The 2027 margin case rests on this and it remains unmodelable from outside.

5. Autonomy acquired its first dated commitment

DoorDash Dot has moved from a program described in principle to a program with a testable target, and the release states it plainly.

"Based on our current progress, we expect Dot to deliver a high single-digit percentage of orders in our largest test market by the end of the year." — DoorDash Q2 2026 shareholder release

The market is Phoenix, the vehicle travels road, sidewalk and bike lane, and the company also earned Part 135 air carrier certification from the Federal Aviation Administration for DoorDash Air. Asked how quickly Dot expands to other cities, the CEO declined to give a schedule.

"I'm a big believer that you really have to nail something before you scale it, especially in the area of autonomy where you're really solving like the problems of like six separate individual companies, almost, and you kind of have to be great at all of them." — Tony Xu, Co-Founder, Chair and CEO

Assessment: a high single-digit share of orders in a full metropolitan market by December is the first autonomy claim from this company that can be checked, and it is a meaningfully higher bar than the demonstrations peers publicise. The program still carries no disclosed spend, no unit economics and no addressable-volume framing, and it now sits inside a research and development line that grew 52%, with management naming autonomy as a driver of the Q4 margin decline. We continue to value it at zero and note that its cost is no longer invisible even if its economics are.

6. Membership is the disclosed engine, and it still has no level

The release devotes its longest section to the membership programs and produces two genuinely new datapoints. U.S. paid DashPass members grew more in the twelve months through Q2 2026 than in the previous twenty-four months combined, and DashPass members placed approximately 75% of Total Orders in the U.S. grocery and retail categories during the quarter. The company also concedes the trade explicitly: DashPass lowers consumer fees and therefore carries a lower gross margin percentage than non-member orders, a trade it will make so long as engagement gains prove durable.

"We are happy to make this trade as long as we believe the increase to consumer engagement is durable and average consumer lifetime value increases." — DoorDash Q2 2026 shareholder release

Assessment: the 75% figure is the most useful membership disclosure the company has ever given, because it converts a claim about engagement into a structural fact about who actually buys groceries on this platform. It also reframes the take-rate debate. If three-quarters of non-restaurant orders come from members who by construction pay lower fees, then Net Revenue Margin declining as grocery mix rises is arithmetic rather than competitive pressure. What remains missing is the member count itself, so the growth claim cannot be sized and the lifetime-value argument cannot be tested.

7. Merchant services became a numbered business

For the first time, the release quantifies three separate business-to-business lines in a single paragraph: new signed venues at SevenRooms grew over 100% year over year, revenue from the digital ordering service used by more than 150,000 merchants grew over 40% year over year, and reservations booked through the marketplaces rose over 150% sequentially.

"In Q2 2026, we grew new signed venues at SevenRooms by over 100% Y/Y and grew revenue from our digital ordering service, which is being used by over 150,000 merchants, by over 40% Y/Y." — DoorDash Q2 2026 shareholder release

Assessment: three growth rates without a revenue base is still not a disclosure an analyst can model, but it is a marked change from a company that has historically described this portfolio only in prose. The strategic logic, that owning the merchant's first-party ordering, reservations and in-store demand channels makes the marketplace relationship stickier and higher-margin, is coherent and is the clearest articulation of the mix-shift pillar management has offered. Until a dollar figure appears, it belongs in the option column rather than the model.

8. Agentic commerce: the traffic has not arrived

Last quarter this was the dominant theme of the call and the answer was a historical analogy about top-of-funnel platforms. This quarter one questioner asked for an actual traffic update and received a direct empirical answer, which is more useful than the analogy was.

"I mean what I'd say, in short, is no, the volume is quite low, I think, from some of the agentic partners that we've been testing with." — Tony Xu, Co-Founder, Chair and CEO

The reasoning offered is that the large model platforms have concentrated on enterprise and coding rather than consumer commerce, and that consumers care about the end-to-end fulfillment rather than the interface. The company's own answer is DoorDash Ask, an assistant that helps consumers discover restaurants and build grocery baskets.

Assessment: a specific negative datapoint is worth more than a well-constructed argument. Agentic traffic is currently immaterial to DoorDash, which removes it as a 2026 earnings variable and leaves it a 2027-and-beyond structural question. The relevant watch item is unchanged: if discovery migrates, the first evidence is take-rate compression, and this quarter's take rate went the other way. We are less worried about this than we were three months ago and no better informed about the long-run answer.

9. AI internally: efficiency gains are being reinvested, explicitly

Asked whether AI spending has changed hiring plans, the CFO described a company that has built cost controls around AI usage, including internal routing of tasks to the cheapest adequate model and budget caps embedded in team budgets, and then stated where the savings go.

"The sharper question for us is how do you take the efficiency gains and reinvest that back in the business. Look, we are investing in building some large areas, right? We've talked about autonomy. We've talked about AI. We've talked about the unification of the tech stack." — Ravi Inukonda, Chief Financial Officer

Assessment: the posture is unchanged from Q1 and should be taken at face value. Nobody should model AI-driven margin expansion at DoorDash in 2026 or 2027, because management has now said twice, in different words, that productivity gains fund investment rather than earnings. The cost discipline described, particularly routing tokens by required intelligence, is more operationally serious than most companies' AI commentary, and its visible effect is that third-party software expense inside research and development rose $70M year over year anyway.

10. The buyback finally moved the share count

DoorDash repurchased 5.7 million Class A shares in the quarter at a weighted-average price of $156.88 for $887M, taking the year-to-date total to 6.8 million shares for $1,049M and leaving approximately $3,951M of the February 2025 authorization. Weighted-average diluted shares fell to 439.3 million from 442.3 million sequentially, and the year-over-year increase compressed to 0.2% from 1.6% in Q1.

The share count in the release, measured through August 5, is identical to the June 30 figure, which means no shares were repurchased in July while the stock ran from the low $190s to $207.

Assessment: the Q1 criticism, that this was an anti-dilution program failing to anti-dilute, no longer holds. Executing $887M at an average price 26% below the reaction-day close and 34% below the current level is good capital allocation by any measure. The corollary matters for the stock: management stopped buying as the price rose, so the buyback is not a support at these levels. Full-year stock compensation guidance was lowered to $1.2B to $1.3B from $1.3B to $1.4B, which narrows the gap against a year-to-date repurchase spend of $1,049M through August 5. That pace annualises near $1.8B, and it will not be sustained at this price.

11. The legal and regulatory reserve hit a five-quarter high

Certain legal, tax and regulatory settlements, reserves and expenses excluded from Adjusted EBITDA totalled $98M in the quarter, against $45M in Q1 and $29M a year ago, and more than double the previous high of $48M in the five quarters the company presents. The quarterly filing attributes a $108M year-over-year increase in general and administrative expense to legal, tax and regulatory items. The California Employment Development Department payroll-tax assessment on Dasher classification remains under appeal with an accrual recorded inside accrued expenses and other current liabilities, and the amount is not disclosed.

Assessment: more than tripling the quarterly regulatory reserve is a fact that deserved a question and did not get one across fifteen analysts. It may be a settlement provision, a tax reserve, or policy spending, and the release distinguishes none of them. Worker classification is the standing tail risk in this business model and its only quantified expression on the balance sheet is invisible. The offsetting observation is that the exclusion is at least consistent: this same line has been excluded every quarter, so the Adjusted EBITDA series is not being flattered by a change in definition.

Guidance & Outlook

The Adjusted EBITDA guide is the strong half. A $1,025M midpoint is 3.06% of the $33.5B GOV midpoint, which would be the first quarter above 3% since Q3 2025 and 4.8% ahead of the $978M consensus. The GOV guide is the soft half and is doing less work than it appears. Its midpoint is only 1.3% above the Q2 actual, against a 3.2% sequential step from Q2 to Q3 in 2025, and its midpoint growth rate of 33.9% is 2.5 points below the 36.4% just delivered. Some of that is the Deliveroo comparison beginning to normalise, and some of it is a company that has stopped guiding its top line to a beatable number.

Implied second-half ramp: the two reported quarters produce $1,668M of first-half Adjusted EBITDA against the $1,574M the Q1 guide implied, so the year starts $94M ahead of where we modelled it. Taking the Q3 midpoint at $1,025M and applying the 2025 sequential margin pattern to Q4, a roughly 39 basis point step down from 3.06% to 2.68% on an estimated $36.8B of Q4 GOV, points to $986M in Q4 and $3.68B for the year, against $2.78B in 2025.

Street at: consensus for Q3 Adjusted EBITDA was $978M against a $1,025M guided midpoint, so for the first time in three quarters the company guided above the Street on its profit line rather than below it. No sourceable consensus exists for Marketplace GOV.

Guidance style: the behavioural change flagged in the Q1 recap has now been contradicted in one direction and confirmed in the other. On GOV, DoorDash again guided a range it landed inside rather than through, which is the new, less sandbagged pattern. On Adjusted EBITDA it beat the high end by $44M, the widest overshoot it has produced in the periods we cover, and the CFO's own explanation is that the upside arrived too late in the quarter to be reinvested. Investors should treat the GOV range as informative and the EBITDA range as a floor that management defends and occasionally overshoots by accident.

Analyst Q&A Highlights

DoorDash delivers no prepared remarks, so the Q&A is the entire management performance. Fifteen analysts asked questions across roughly one hour, and the distribution is itself informative: more than half concerned product, autonomy or AI, exactly one touched the withdrawn full-year margin frame, and none went near the regulatory reserve or the free-cash-flow timing disclosure.

Whether the Adjusted EBITDA beat is a run rate or a timing accident

The sharpest exchange of the call put the magnitude of the beat directly to management, asked whether investment had simply been pushed into later quarters, and then asked for a reaffirmation of the full-year margin frame. The first two parts were answered candidly. The third was not answered at all.

Q: "So guys, I just want to double-click on the margin question again. Can you -- Ravi, can you maybe just talk about the drivers for the material beat in adjusted EBITDA in Q2? It came quite a bit outside of the guidance range. More recently, you guys have been guiding to, at some point, hit somewhere in the midpoint. This one is dramatically higher. Were there any investments that got pushed back into Q3 and then Q4? And then are we still tracking to show higher year-on-year adjusted EBITDA margin for 2026 ex Deliveroo?"
— Youssef Squali, Truist Securities

A: "To your second point, look, I mean, our focus has always been on landing inside the range. And if you're thinking about the second half from a margin perspective, I would expect us to land inside the range of the guidance that we've given in Q3. There are going to be times like in Q2, where the EBITDA beat comes in later in the quarter. In those times, we just don't have enough time to reinvest back in the business, especially at the levels of efficiency that we desire."
— Ravi Inukonda, Chief Financial Officer

Assessment: the question contained an explicit request to reconfirm the annual margin commitment and the answer substituted a quarterly one. Combined with the sentence disappearing from the release, this is the strongest available evidence that the full-year frame is no longer being defended. The candour about the beat's timing is genuine and useful, and it is also a warning that the Q2 margin is not a base to extrapolate from.

What drove the take-rate recovery and where it goes next

The other exchange that moved our model asked how much of the Net Revenue Margin recovery was Deliveroo mix, advertising or fee changes. The answer named none of those as the primary driver and gave a two-quarter forward path instead, which is more than the company usually offers on this line.

Q: "I know you don't manage for take rate, but I just wanted to get a little bit more color just on net revenue margin and the pickup that you saw kind of in 2Q. Just trying to understand some of the drivers there in terms of how much might have been Deliveroo contribution versus advertising and fee changes and how you think about that going forward?"
— Douglas Anmuth, JPMorgan

A: "The Q-on-Q increase, which is what I think you're referring to, is largely from Dasher. Look, Dasher costs are seasonal for us. When you go from Q1 to Q2, Dasher costs are lower in Q2. That's what gave rise to sort of the tick up in take rate that you saw from Q1 to Q2. And if you're thinking about from a modeling perspective, what I would expect for the rest of the year, I'd expect take rate to slightly be in the similar range in Q3, so flattish from Q2 to Q3, and lower in Q4."
— Ravi Inukonda, Chief Financial Officer

Assessment: a direct question about three named drivers produced a fourth that none of them anticipated, and a two-quarter guide that the company has historically refused to give on this metric. The willingness to guide the take rate down in Q4 is the credibility-building part of the answer. The consequence is that the Q1 concern is not retired: if seasonality explains the recovery, the year-over-year picture in Q4 is the next real test, and management has pre-announced it will be worse sequentially.

Repricing the grocery relationships that carry the affordability cost

The opening question of the call went straight to the least comfortable question in the grocery category, asking whether partners onboarded at effectively zero take rates can be repriced and whether the affordability burden the platform currently absorbs can be shifted back to them. Management confirmed the direction without conceding the premise.

Q: "I was wondering if that's the case and then what the opportunity is to reprice these relationships going forward, maybe pushing some of the affordability burden that's lending on DoorDash right now back to the grocers."
— Michael Morton, MoffettNathanson

A: "In terms of the economic relationships, I'm not going to really comment about anyone in particular. I mean what I will say is that when you are the fastest grower in the market for them, and you are their source of growth. Put a different way, we might be 100% of the growth that they see in terms of their actual business. You certainly have opportunities to grow your business with them as well as improve your relationships with them."
— Tony Xu, Co-Founder, Chair and CEO

Assessment: "we might be 100% of the growth that they see" is a bargaining-power claim, and it is the most direct statement management has made that grocery take rates are a negotiated variable with room to move. The CFO followed by confirming the new-verticals portfolio remains on track to be gross-profit positive in the second half, which is now the single most gradeable commitment on the platform and comes due in the Q3 and Q4 prints. If repricing is where that profitability comes from, the improvement is durable; if it comes from mix, it is not.

What the platform rebuild returns, and when

A questioner pressed for quantifiable evidence from the modules already in production that would support a return on the largest investment of the cycle. The answer described mechanism and examples without attaching a figure to either the spend or the benefit.

Q: "So now that you've got the tech stack sort of built as of the spring, I know you've been testing a lot of modules and new capabilities. Can you give us some examples where you're seeing early signal with actual quantifiable benefits of the new modules that give you confidence you're going to get real return and ROIC on these investments as we go into 2027?"
— Brian Nowak, Morgan Stanley

A: "The better analogy is it's not like a LEGO project where there's like a finishing step in the instruction manual. It's more like you're constantly -- well, first, you have to like replace an engine while you're flying a plane that's growing in speed and climbing in an altitude and then you're constantly making tweaks. That's probably the more appropriate analogy. But we're doing it. And where are we seeing benefits? I mean, a lot of places already. I mean you're seeing conversion wins from improvements in search."
— Tony Xu, Co-Founder, Chair and CEO

Assessment: the same questioner extracted the only phasing commitment of last quarter's call and this quarter received an analogy. Conversion wins from search and automation in customer support are directionally credible and unquantified, which leaves the program without a specific figure on either side of the ledger four quarters after it was first sized as several hundred million dollars. The release separately moved full rollout to the first half of 2027, so the redundant cost persists longer than the Q1 phasing implied.

Deliveroo's acceleration and whether it reaches the bottom line

The question that mattered most for the acquisition thesis asked whether three quarters of accelerating top-line growth at the acquired business is also showing up in profit. The answer confirmed both and, notably, reaffirmed a target that the written release had dropped.

Q: "You've had now 3 quarters in a row of kind of accelerating growth, I think, in orders, yes, and in GOV and then I think in revenue. So just peel that back a little bit there -- how many different opportunities you've had, what you've been able to pull, what you've been able to change in order to deliver that better performance? And it's a little hard to tell, but is it also showing up on the bottom line?"
— Mark Mahaney, Evercore ISI

A: "At the same point, to your second part of your question, we've increased the unit economics as well. But the way in which we're operating the business is very similar, right? We're finding great opportunities to drive investment back in selection. Quality subscription is a big area of focus for us. So we're going to continue to invest back in the business. And my expectation is we'll continue to drive higher top line as well as meet the profitability targets that we set out in the last letter."
— Ravi Inukonda, Chief Financial Officer

Assessment: "meet the profitability targets that we set out in the last letter" is the only reaffirmation of the roughly $200M Deliveroo contribution anywhere in this quarter's disclosure, and it exists only in the transcript. A verbal reaffirmation of a number the written release stopped printing is weaker than the written commitment it replaced, and it is the second instance in the same quarter of a hard figure migrating from the release into the call. The operating news is genuinely good; the disclosure trend is not.

How far autonomy scales, and by when

A recurring line of questioning across two separate exchanges tried to establish both the ceiling for autonomous penetration and the schedule for expanding beyond the single test market. Management gave the year-end penetration target in the release and declined to give a city schedule or a ceiling on the call.

Q: "So Tony, I think you have a few hundred robots in Phoenix. How quickly is the plan to kind of bring this to other cities?"
— Multiple analysts incl. Ross Sandler, Barclays; Josh Beck, Raymond James

A: "And I'm a big believer that you really have to nail something before you scale it, especially in the area of autonomy where you're really solving like the problems of like six separate individual companies, almost, and you kind of have to be great at all of them. ... We are in parallel, of course, securing permits because we work with cities to actually unlock a lot of this. And we do have plans to expand. We'll share certainly more as time comes."
— Tony Xu, Co-Founder, Chair and CEO

Assessment: the discipline is defensible and the answer is still an answer about sequencing rather than economics. What changed this quarter is that the program acquired a falsifiable target, a high single-digit share of orders in the largest test market by year end, which converts autonomy from an unmeasurable narrative into something the Q4 print can be graded against. Cost remains undisclosed and is now explicitly named as a driver of the Q4 margin decline, so the option has stopped being free even though it has not started being valuable.

Whether agentic traffic has materialised

Last quarter's dominant theme returned in a narrower and more useful form, asking for an actual traffic reading rather than a strategic view. The answer was unhedged.

Q: "Just wondering if you can give us any agentic traffic update, if you're seeing any traffic from there. And then given your huge merchant scale, are there ways where you could really capitalize on that traffic as they roll out booking capabilities and maybe even lower your marketing costs?"
— Justin Post, Bank of America

A: "I mean what I'd say, in short, is no, the volume is quite low, I think, from some of the agentic partners that we've been testing with. But it also isn't that surprising, I think, for a couple of reasons. I think, first, I think especially for some of the larger platforms out there, their core focus has been on the enterprise and much less, especially in coding agents and probably less on the agentic side."
— Tony Xu, Co-Founder, Chair and CEO

Assessment: a straight negative answer to a question the company could easily have hedged is worth more to a model than three paragraphs of framing. Agentic disintermediation is not a 2026 earnings variable. It stays on the risk register as a structural question for 2027 and beyond, and the metric that would first reveal it, the take rate, moved the other way this quarter.

AI spending, its return, and what it does to headcount

The final question of the call asked whether AI investment is changing hiring plans. The answer described unusual internal cost discipline and then closed the loop the same way every cost question on this call closed.

Q: "Ravi, I'd just be curious about your big picture view on AI spending and the ROI that you're seeing. And I'm curious if it's impacted your future hiring plans at all."
— Thomas Champion, Piper Sandler

A: "We've built models where internally, the tasks are routed to the right model, depending on what the actual cost, quality and efficiency is. We've put caps in place. We've also incorporated some of the AI budgets into teams' existing budgets. Look, we are seeing gains from the usage of AI. The sharper question for us is how do you take the efficiency gains and reinvest that back in the business."
— Ravi Inukonda, Chief Financial Officer

Assessment: routing tasks by required intelligence and embedding AI budgets inside existing team budgets is more operationally specific than most companies manage on this topic, and the hiring question was left unanswered. The governing sentence is the last one. Efficiency becomes investment, not earnings, which is the same answer given in Q1 and is consistent with third-party software expense inside research and development rising $70M year over year regardless.

What They're NOT Saying

1. The full-year Adjusted EBITDA margin frame. The commitment to increase Adjusted EBITDA as a percent of GOV slightly versus 2025 excluding Deliveroo, carried in both prior releases, is absent from this one and was not reconfirmed when asked for directly on the call. Nothing replaced it except a second-half shape statement that guides Q4 down sequentially.
2. The Deliveroo contribution figure. The roughly $200M contribution to 2026 Adjusted EBITDA, printed in both prior releases, does not appear in this one. Its only reaffirmation is a verbal reference on the call to the targets set out in the last letter.
3. Dasher gas relief. Q1 guided to a gross cost of over $50M in Q2 and management declined then to commit on duration. The release and the entire transcript contain no reference to the program. Whether it ended, continued at cost, or was absorbed elsewhere cannot be determined from the disclosure, and no analyst asked.
4. The size of the platform spend. Four quarters after being sized only as several hundred million dollars, and one after being called the largest component of the investment cycle, the global technology platform still carries no disclosed cost, no disclosed duplicate run rate and no quantified benefit. Completion has now moved to the first half of 2027, which extends the redundant-cost window without extending any figure.
5. Any advertising revenue number. Advertising was named on the call as one of two specific areas where unit economics exceeded plan and is the primary reason the take rate is not falling further. It appears nowhere as a dollar figure in the release or the quarterly filing.
6. Monthly active user and membership counts. The release states that U.S. paid DashPass members grew more in twelve months than in the prior twenty-four combined, that MAUs hit all-time highs, and that DashPass members placed approximately 75% of U.S. grocery and retail orders. No level accompanies any of these. Every engagement claim remains unauditable.
7. Deliveroo's standalone GOV and orders. The company reconciles revenue excluding Deliveroo in dollars and gives ex-Deliveroo growth rates for GOV and orders. That is enough to compute organic growth and not enough to size the acquired business, which matters increasingly as the Q4 lap approaches.
8. What is inside the $98M regulatory reserve. Certain legal, tax and regulatory reserves more than tripled year over year and more than doubled the prior five-quarter high. The release names the category and not the matter. Fifteen analysts asked no questions about it.
9. The California payroll-tax accrual. The quarterly filing confirms an accrual is recorded within accrued expenses and other current liabilities for the state Employment Development Department assessment on Dasher classification, and does not disclose the amount. The standing tail risk in this business model remains invisible on the balance sheet.
10. Why the year-end merchant-payment timing costs $700M to $800M. The release states the effect on reported 2026 free cash flow and explains neither the mechanism nor whether it reverses in 2027. For a company whose cash-generation credibility was the weakest part of its Q1 print, a three-quarters-of-a-billion-dollar cash-flow bullet deserved more than one sentence.
11. Full-year revenue or GOV guidance. DoorDash continues to guide one quarter at a time on GOV and Adjusted EBITDA. That remains long-standing practice rather than a change, but with the full-year margin frame now withdrawn, the annual figure every investor is underwriting is entirely the analyst's own construction.
12. Autonomy economics. Dot now has a year-end penetration target in one market, a multimodal capability description and a partner platform. It has no disclosed spend, no cost per delivery, no addressable-volume estimate and no city expansion schedule, while being named as a driver of lower Q4 margin.

Market Reaction

• Pre-print setup: DASH closed at $207.27 on August 5, down 8.5% year to date against an S&P 500 up 12.8%, down 18.8% over trailing twelve months, and up 10.0% over the trailing thirty days. The 52-week closing range entering the print was $146.60 to $281.74, so the stock came in near the middle of its own range after a strong month.
• After-hours move: the initial reaction was negative, with shares trading down roughly 5% shortly after the release on the headline 30% decline in GAAP net income. The move reversed across the after-hours session as the release body and then the 4:30 PM ET call were absorbed, with the stock closing the extended session roughly 1% higher.
• Next-day session: the stock opened at $214.67, a 3.6% gap up, traded between $207.18 and $218.03, and closed at $213.26, up 2.9% or $5.99. Volume of 8.6 million shares was 2.1 times the 30-day average.
• Peer and market context: the S&P 500 fell 0.2% on the same session. Across the broader on-demand and marketplace complex, one large peer rose 3.4%, another fell 1.9%, and a third fell 4.8%.
• Since the print: the stock has continued higher, closing at $236.74 on August 28, up 11.0% from the reaction-day close and 14.2% from the pre-print close.

The sequence is the exact inverse of last quarter, and the inversion is the analytically useful fact. In May the print rallied after hours and the call gave three-quarters of it back by the next close. In August the print sold off and the call recovered it. Both times the market's considered answer arrived after management spoke, which for a company that publishes no prepared remarks means the Q&A is doing the price discovery.

What the call added is identifiable. The release led with a 30% decline in GAAP net income and a revenue line that beat by 2.5%, which is a mixed headline. The call supplied the composition of the Adjusted EBITDA beat, a two-quarter forward path on the take rate, and confirmation that the new-verticals portfolio remains on track to turn gross-profit positive. Each of those addresses a specific Q1 bear point, and none of them is legible from the release headline alone.

The three weeks since have added a further 11% on no company news, which is the part of the reaction that should give a buyer pause. Sell-side price targets were raised broadly into a range spanning roughly $220 to $260, and several of the desks that raised targets kept neutral ratings, which is the tell: the numbers moved and the recommendations did not. A stock that re-rates 38% from its prior earnings close on an estimate revision of roughly 4% has repriced the multiple, not the business.

Street Perspective

Debate: Is the margin inflection structural or seasonal?

Bull view: the bull case being made on the Street is that the recovery sits in adjusted gross profit per dollar of order value, which rose 23 basis points year over year and 68 sequentially, and that a gross-profit-driven recovery is by construction a recovery in the economics of a delivery rather than a spending deferral. Contribution margin at 36.8% and adjusted gross margin at 54.2% are both at or above the five-quarter high, and management named advertising and order subtotal as the sources.

Bear view: the bear camp contends that the CFO himself attributed the sequential take-rate recovery to lower Dasher costs in Q2 and then guided the take rate lower in Q4, which makes a meaningful part of the margin gain a seasonal artifact that reverses. On this read the year-over-year expansion of six basis points is within noise, the withdrawal of the full-year margin frame is the company conceding as much, and Q4 is where the honest comparison happens.

Our take: both are partly right and the bulls have the stronger case on evidence. Seasonality explains the sequential move in the take rate; it does not explain a 23 basis point year-over-year improvement in adjusted gross profit per GOV dollar against the same quarter last year, which carries the identical seasonal pattern. What the bears get right is the forward path. Management has pre-announced a Q4 margin decline with three named drivers, and it withdrew the annual commitment in the same document. We treat the structural improvement as real and the 2026 full-year margin as roughly flat against 2025.

Debate: Does the disclosure withdrawal matter, or is it housekeeping?

Bull view: the optimistic camp argues that a company guiding one quarter at a time has always been explicit about doing so, that the full-year margin frame was directional rather than numerical, and that the Deliveroo target was verbally reaffirmed on the call. Gas relief disappearing most likely means the program ended when fuel prices normalised, which is good news that does not require a press release. Reading a withdrawn sentence as a signal over-interprets a document that is not a legal filing.

Bear view: the skeptics counter that three separate forward commitments vanished from the same release, in a quarter that beat, and that the one direct question asking for reaffirmation of the most important of them was answered about a different period. A company confident in its annual margin trajectory restates it when it has just beaten by $44M. Withdrawing guidance from strength is how managements create room without creating a headline.

Our take: the bears are closer, with an important qualification. We do not read the withdrawal as concealment of deterioration, because the reported numbers went the right way on every line that matters. We read it as management reclaiming flexibility to spend, which is exactly what the CFO described twice on the call when he said efficiency gains get reinvested and that the Q2 beat happened only because it arrived too late to spend. The practical consequence is that upside to the 2026 margin now accrues to investment rather than to earnings, and forecasts should be built accordingly.

Debate: Has the stock already priced the recovery?

Bull view: a growing consensus view holds that at roughly 20 times 2027 Adjusted EBITDA the stock is not expensive for a business compounding GOV in the low thirties, expanding margin, generating real free cash flow and holding net cash. The platform rebuild completes in 2027 and its duplicate cost comes out; new verticals turn gross-profit positive in the second half of 2026; Deliveroo laps and international operating leverage begins. All of those point the same direction and none is in the current-year number.

Bear view: some sell-side desks note that the stock has risen 38% since the last earnings close while the 2026 estimate rose about 4%, so essentially all of the move is multiple expansion. At 27 times current-year Adjusted EBITDA and 47 times trailing free cash flow, with a $700M to $800M cash-flow timing reversal scheduled for the fourth quarter and a company that has stopped publishing its annual margin frame, the entry point demands that 2027 goes right rather than merely that 2026 finishes well.

Our take: we side with the bears on the entry point and the bulls on the destination. The 2027 case is credible and we would underwrite it; the price now requires it to arrive on schedule, and the two things that would delay it, the platform rollout slipping past the first half of 2027 and the investment load scaling with the business rather than lagging it, are precisely the two items management declined to quantify again this quarter. This is a business we want to own at a price that no longer offers a margin of safety.

Model Update Needed

Valuation. At the August 28 close of $236.74 and 439.3 million diluted shares, the diluted market capitalization is $104.0B. Cash and equivalents of $4,424M, short-term investments of $923M and long-term investments of $869M against $2,727M of convertible notes leave $3.49B of net cash, for an enterprise value of $100.5B. That is 31.4 times trailing-twelve-month Adjusted EBITDA of $3,202M, 47.0 times trailing free cash flow of $2,139M, 27.3 times our FY2026E Adjusted EBITDA of $3.68B, and 19.7 times our FY2027E of $5.10B. At the May 7 close the equivalent forward multiple was roughly 20 times the then-current year.

Valuation impact: we raise the fair-value range to $228 to $252 from $177 to $189, built as 19 to 21 times FY2027E Adjusted EBITDA of $5.10B plus $3.49B of net cash across 439 million diluted shares. The $240 midpoint implies roughly 1% upside from the current price, which is what a Hold looks like in arithmetic after a 38% move. We would go to 22 times and a range toward $270 on three pieces of evidence: a Q4 sequential margin decline shallower than the roughly 39 basis points 2025 produced, ex-Deliveroo order growth holding at or above 17% through the Deliveroo lap, and a disclosed dollar figure for either advertising revenue or the platform spend. We would move the other way if adjusted research and development passes 1.10% of GOV without a named offset, or if Net Revenue Margin falls below 13.0% in Q4.

Thesis Scorecard Post-Earnings

Overall: thesis strengthened, and by more than the rating change suggests. Four of the six pillars moved favourably, including both of the two we said in May would determine whether this business earns a premium multiple. The margin inflection we said we would need to see in the Q3 print arrived in the Q2 print, and it arrived through gross profit per order dollar rather than through underspending, which is the version that counts. The two things that did not improve are the investment load, which got larger and less quantified, and the company's willingness to commit to an annual frame, which went backwards.

Action: hold. This is now a better business than the one we rated in May and a worse entry point. The correct posture is to own the position, decline to add at 27 times current-year Adjusted EBITDA after a 38% run, and use the Q4 print as the decision point, because Q4 is where the take-rate seasonality unwinds, where the new-verticals gross-profit commitment comes due, where the Deliveroo comparison laps, and where the $700M to $800M cash-flow reversal lands. Four falsifiable events in one quarter is unusual, and it makes waiting cheap.