Sidewalk delivery robots: what Serve's filings say about robots actually working

Serve Robotics is the only sidewalk delivery robot company that files quarterly reports with the SEC, so it is the only one where "2,000 robots" can be checked against audited and reviewed numbers. The check gives three different counts. Serve says it deployed its 2,000th sidewalk robot in December 2025. Its own decks add about 100 Moxi hospital robots from Diligent Robotics, which it bought in January 2026. And its third-quarter 2026 10-Q reports an average of 917 "daily active robots", a count that since January 2026 includes the hospital robots and, by the 10-Q's own description, robots generating branding revenue. On Serve's figures, 917 is about 44% of the roughly 2,100 robots it says are deployed, and between about 41% and 46% of the 2,000 sidewalk robots, depending on how many Moxi units are in the count (Servospace arithmetic).

Revenue per robot is small and does not separate sidewalk work from hospital work. Fleet services revenue, which covers outdoor deliveries, hospital service contracts and robot advertising, was $2.582 million in the quarter, about $31 per daily active robot per day and about $13 per deployed robot per day (Servospace arithmetic). No quarterly delivery count was found in the Serve filings, releases, and decks Servospace reviewed, as of November 12, 2026. Serve's decks repeat a cumulative "1.8M" sidewalk plus hospital deliveries, and Serve's own January 2026 release credited Moxi with over 1.25 million deliveries by itself, while Serve's November 11 release puts its sidewalk robots at "hundreds of thousands" of deliveries.

The Uber relationship is shrinking on the records that name Uber. Serve's filed 10-Q says Uber Eats volume came in below expectations and that it expects reduced demand through the rest of 2026. Serve's CEO said on the August 6 call that Serve does not expect to renew the Uber agreement when it expires in early 2027, a remark reported by both TechCrunch and Zacks. Only the Zacks summary adds that this holds unless the operating model improves, and that talks continue. Uber's own 13F holdings report for March 31, 2026 listed 2,070,629 Serve shares, and its report for June 30, 2026 no longer listed Serve. Starship and Coco, the two private rivals, publish cumulative delivery counts and robot totals but no daily active count and no revenue, so none of their figures can be checked the same way.

Analysis based on Serve Robotics' 10-K for 2025, its 10-Q filings for the first three quarters of 2026, its earnings releases and investor presentations furnished to the SEC from November 2025 to November 2026, Serve's January, August and November 2026 releases, Uber's 13F filings, DoorDash's 2026 shareholder letters filed with the SEC, EDGAR full-text searches, a published transcript of Serve's November 5 call, TechCrunch and Zacks accounts of the August 6 call, and Starship and Coco company releases, all retrieved November 12, 2026. AI assistance was used in research and drafting, and the piece was checked by a separate AI fact-check and AI editorial review. Servospace did not ride along with any robot, interview the companies named, or see operating data. Calculations are Servospace arithmetic on published figures and are labeled as such.

In this analysis

• The short answer
• What kind of evidence each number is
• Built, deployed, active: three different counts
• How the active-robot definition moved
• Revenue per robot
• Deliveries: the number Serve does not report
• Uber: concentration, renewal and the stake
• The newer partners
• How Starship and Coco compare
• Counterevidence and limits
• What a merchant, city or investor should ask for
• What would change this

The short answer

"2,000 robots" is a deployment count, not a working count. Serve's filed metric for robots working on an average day was 917 in the third quarter of 2026, and that figure mixes sidewalk robots with hospital robots. Serve's revenue per active robot roughly doubled after it bought a hospital robot business, which tells you more about the acquisition than about sidewalk demand, because Serve reports one segment and does not split sidewalk revenue from hospital revenue. Cost of revenue was $10.4 million against $2.8 million of revenue in the quarter, so no robot in the filings pays for itself yet. The largest single customer's share of revenue fell from 43% to 10% in a year, Serve's largest delivery partner sold its shares, and the rivals' numbers are vendor claims with no public denominator.

For readers new to the topic, Servospace's beginner's guide to how robots work covers the basics. The quarter itself is covered in our news story on Serve's third-quarter results, and the November 11 developer API in Serve opens its sidewalk robots to developers. This guide is the cross-quarter denominator check those stories do not have room for.

What kind of evidence each number is

Built, deployed, active: three different counts

Serve's sidewalk fleet grew fast in 2025 and then stopped growing. Its November 2025 investor deck set a goal of a "1,000 robot fleet size by end of Q3 2025" and said the 2,000th robot was "expected ahead of schedule in mid-December." Serve's later decks say the 2,000th robot was deployed in December 2025, and the 2025 10-K says that "As of December 31, 2025, Serve's fleet consisted of over 2,000 sidewalk delivery robots." Serve's May 2026 results release then said that "With approximately 2,000 robots deployed, focus has shifted from fleet expansion to increasing revenue per robot."

No count of robots built, as distinct from robots deployed, was found in the Serve filings and releases Servospace reviewed, as of November 12, 2026. Its decks say Magna International builds the third-generation robot, and the 10-Q books robots as property at cost, with a 4 to 6 year useful life. Robot assets were $49.7 million gross at September 30, 2026, after a $4.5 million reduction for tariff refunds, but the filing gives no unit count for that line.

The labels do not agree with each other. The deck counts 2,000 sidewalk robots plus 100 Moxi robots. The November 5 call, as transcribed, puts 2,000 robots across 40 cities and 27 hospitals, and the November 11 release says "more than 2,000" across the U.S. and Australia. The city count also varies: the decks say 44 U.S. cities, the call says 40, and the November 11 release says Serve's robots "have completed hundreds of thousands of deliveries across more than 150 neighborhoods in 20 US cities." Serve does not explain the gaps, and they may reflect different definitions. Treat 2,000 as a round deployment figure that has not moved since December 2025, not as a measured count of robots on the street.

The daily active count is the filed measure of robots at work. It is an average over the quarter, which matters: a fleet can have 2,000 robots and far fewer on duty on a typical day.

Sources: Serve results releases for Q3 2025, Q4 2025, Q1 2026, Q2 2026 and Q3 2026, and the Q3 2026 10-Q. Q3 2025 was the quarter Serve's deck had targeted for a 1,000-robot fleet; the average on duty was 312.

Q4 2025 is the last clean sidewalk-only number: 547 robots active on an average day, in a quarter that ended with over 2,000 deployed. Because the 2,000th robot was deployed in December 2025, a quarter average against a year-end count understates the share, so that ratio is not a fair utilization rate. The jump to 812 in Q1 2026 coincides with the first quarter that counts hospital robots. On the decks' figures, up to about 100 of the 917 in Q3 2026 could be Moxi units, which leaves between about 817 and 917 sidewalk robots active on an average day, or about 41% to 46% of 2,000 (Servospace arithmetic). Serve does not say how many Moxi robots are in the average.

When an analyst asked on the November 5 call how many daily active robots Serve might exit the year with, CEO Ali Kashani said he did not have a number to share and that Serve adds robots when it has demand, according to the published transcript.

How the active-robot definition moved

The definitions sit in footnotes. The branding link was already there in 2025; what changed in 2026 is the wording of the release definition, the supply hours wording, and the fleet scope.

Two practical consequences follow. First, the 2026 series cannot be compared directly with 2025: part of the step up from 547 to 812 is the hospital fleet joining the count. Second, the 10-Q's description ties the metric to delivery services or branding revenue, and branding is a real business: Serve said on the November 5 call that one sports campaign ran for six weeks on 455 robots across Dallas, Atlanta, Miami and Los Angeles, according to the published transcript. How a robot that is wrapped for an advertiser but doing few deliveries is counted is not spelled out.

Revenue per robot

Serve reports one operating segment, and its revenue table has two lines: fleet services (outdoor deliveries, indoor hospital service contracts and branding) and software services (licensing and engineering projects). The 10-Q says the third-quarter rise in fleet services was "due to overall expansion and diversification of the robot fleet and partnerships, including hospital-based robots."

Servospace arithmetic divides each quarter's fleet services revenue by daily active robots and by the days in the quarter (92, 92, 90, 91 and 92). Using the roughly 2,100 deployed robots instead, Q3 2026 fleet services work out to about $13 per robot per day, and total revenue to about $15.

The doubling in 2026 is mostly a change in what is being counted. The sidewalk-only quarters ran at about $13 to $15 per active robot per day. From Q1 2026 the numerator includes hospital service contracts and the denominator includes hospital robots. Serve's March 2026 deck said Moxi robots "serve 25+ hospitals" and generate "over $200K annual revenue per facility." If those deck figures still held, hospital revenue alone would be more than $5 million a year, or about $1.25 million a quarter, close to half of Q3 fleet services (illustrative Servospace arithmetic on a vendor claim, not a finding; Serve does not report hospital revenue separately). Serve's August release also said "advertising made up nearly 50% of food delivery revenue in Q2," so delivery fees are only part of what the sidewalk robots earn.

The cost side is larger. Q3 2026 cost of revenues was $10.39 million against $2.835 million of revenue, a gross loss of $7.56 million, or about $123 of cost of revenue per daily active robot per day across all lines (Servospace arithmetic). The 10-Q also says that for the outdoor fleet, Serve allocates a portion of depreciation and network costs to cost of revenue "based on the percentage of the fleet available for delivery." Read literally, that ties the depreciation in cost of revenue to the share of robots available for delivery, so robots that are not available do not add their full depreciation to the gross loss line.

Serve's November 2025 deck said that "At full utilization, each robot expected to pay for itself in under one year." No filing gives a per-robot cost, and with cost of revenue at about 3.7 times revenue in Q3 2026, the filings do not show a robot paying for itself at current utilization.

Guidance tells the same story from the top down:

Nine-month revenue was $9.057 million, so the $12 million outlook implies about $2.9 million in the fourth quarter, roughly flat with Q3 (Servospace arithmetic).

Deliveries: the number Serve does not report

Serve publishes no quarterly delivery count in its filings, so deliveries per robot cannot be calculated. What it does publish are cumulative figures that combine different fleets and change labels:

Three things stand out. By Serve's own figures, most of the combined total is hospital deliveries: Moxi alone was credited with over 1.25 million in January 2026, while the sidewalk robots are described as "hundreds of thousands" ten months later. The combined figure in the decks has not changed since March, even though the May release said the fleet was "nearing 2 million"; a slide repeating 1.8M through November is stale rather than evidence of a decline. And Moxi's own count moved between labels, from "deliveries" in January to "tasks" in March. These are floors from different documents, so subtracting one from another would not give a sidewalk-only total.

Uber: concentration, renewal and the stake

Serve began as Postmates X, the robotics unit of Postmates, which Uber bought in 2020; the 10-K says Uber approved contributing the project's assets to Serve in February 2021. Uber was its first platform partner. The records that name Uber all point the same direction.

What Serve filed. The Q3 2026 10-Q says: "As previously disclosed, the Company has experienced lower than expected delivery volume through the Company's Uber Eats partnership, which contributed to downward pressure on our revenue during the three and nine months ended September 30, 2026, and we expect reduced demand to continue through the remainder of 2026." Serve's August 17, 2026 8-K said the same under Item 8.01 for the second half of 2026, and the August 6 results release said the guidance cut reflected "the removal of projected demand in the second half of 2026" through Uber Eats.

What Serve's filings do not say. No 2026 filing Servospace reviewed states the non-renewal. The 10-K and 10-Qs do not name the customers in the concentration tables, and the customer letters are not consistent from filing to filing, so the table below shows the largest share in each period without matching any letter to Uber.

Customers above 10% of revenue made up 91% of 2024 revenue and 55% of 2025 revenue, according to the 10-K's risk factors. The drop in concentration has two causes the filings support: lower Uber Eats volume, and a larger revenue base after the hospital acquisition. Which effect is bigger, the filings do not say.

What the CEO said. On the August 6, 2026 call, Kashani said delivery volume through Uber had grown for 17 straight quarters through Q1 2026 before reversing in Q2, which he attributed to "lower-than-expected robot utilization," as TechCrunch quoted him. TechCrunch reported that Serve did not expect it would make sense to renew the Uber agreement when it expires in early 2027. A Zacks summary of the same call said Serve does not currently expect to renew unless the operating model improves meaningfully, and that discussions with Uber are continuing. The two accounts agree on non-renewal at early-2027 expiry. The unless clause and continuing talks appear only in Zacks. On the November 5 call, asked about Uber, Kashani said the relationship continues and that the contracts run, he believed, through part of the first quarter of 2027, according to the published transcript. These are call remarks, not filed statements.

What Uber filed. Uber's own records say nothing about Serve's operations; an EDGAR full-text search found no mention of Serve Robotics in Uber's 10-K or 10-Q filings for 2025 or 2026 as of November 12, 2026. Its holdings reports do show the investment ending. Uber's 13F for the quarter ended March 31, 2026 listed 2,070,629 Serve shares, the same number in its September 2025 Schedule 13D amendment, which put the stake at 3.36%. Its 13F for the quarter ended June 30, 2026, filed August 7, no longer listed Serve. TechCrunch reported on August 11 that Uber had sold its entire stake and that the sale surprised Serve, citing a person familiar with the events. The filings show the position was gone by June 30, before the August 6 call; they do not give sale dates.

The newer partners

Serve's growth claims now rest on DoorDash and Grubhub, and on percentages without a base.

The partners' own filings are silent on Serve. DoorDash's second-quarter 2026 shareholder letter discusses only its own robot: "with Dot, our land-based robot, we have increased the number of robots in operation and the average number of deliveries per robot per day," and it expects Dot "to deliver a high single-digit percentage of orders in our largest test market by the end of the year." Its third-quarter letter says it is "investing in building capabilities like warehousing and in autonomous delivery." EDGAR full-text searches found no mention of Serve, Coco or Starship in DoorDash filings as of November 12, 2026. So Serve's newest growth partner is also building a competing robot, a pattern the guidance history above already shows the risk of.

How Starship and Coco compare

Neither rival files with the SEC, so every figure below is a vendor claim on its date, with no active-robot or revenue denominator.

Starship's 10 million is the largest cumulative delivery figure in the category, built since its founding in 2014, with commercial service on U.S. university campuses since 2019 according to its own timeline, and operations in seven European countries. Dividing it by today's 3,000 robots would mix a fleet that grew over twelve years with today's, so Servospace does not turn it into a per-robot rate. Coco's "over 500,000" appeared in TechCrunch's June 2025 report and again in Coco's February 2026 release boilerplate, so it is at least eight months old as a measure. Starship's press page lists an Uber Eats partnership release dated November 20, 2025, and Coco says it works with Uber Eats and DoorDash, so both of Serve's main U.S. platforms have alternatives.

Counterevidence and limits

• The case for Serve. Its numbers are the only audited or reviewed ones in the category, and they are moving the right way on cost: Serve lowered its 2026 non-GAAP operating expense outlook to $150 million from $160 million to $170 million, and its Q3 gross loss of $7.56 million was smaller than Q2's $8.78 million. Daily active robots rose from 792 to 917 in Q3, the first quarter after the Uber volume decline.
• Hospital growth is real revenue. Serve said it signed 12 multi-year hospital contract extensions and added 4 new hospitals through Q3 2026. That is contracted, recurring business, even if it is not sidewalk delivery.
• Supply hours are steady. Hours per active robot stayed at about 12 a day across five quarters, so the active robots that do run are kept on long shifts.
• What this piece cannot see. Servospace did not review city permits, so permitted versus operating robot counts in Los Angeles, Miami, Chicago or other cities are not covered here. Serve's 10-Q does warn that a city "cap on the number of robots" could limit revenue. Transcripts are machine-produced; quotes from the November 5 call come from one published transcript and are paraphrased where possible.
• No quarterly delivery count. Without it, a merchant or city cannot check deliveries per robot per day, the number that decides whether robot delivery is cheaper than a courier.

What a merchant, city or investor should ask for

What would change this

• Uber agreement expiry, early 2027. Watch for an 8-K or 10-K statement on renewal, replacement or wind-down, and for the Uber share of revenue in the next filings.
• Serve's fourth-quarter 2026 results and 2026 10-K. Last year's results came on March 11, so early March 2027 is a reasonable expectation, not an announced date. Key items: whether Q4 revenue lands near the implied $2.9 million, daily active robots, any new split of indoor and outdoor robots or revenue, and the year-end fleet count in the 10-K.
• A sidewalk-only metric. If Serve begins reporting sidewalk robots or deliveries separately, the per-robot arithmetic above should be redone.
• Rival disclosures. A Starship or Coco filing, a city permit record, or a platform partner naming robot volumes would give the first independent check on the private fleets.

Servospace will update this guide when Serve's 2026 10-K is filed.

For the airborne side of the same last-mile question, see what FAA filings and company counts show about drone delivery volumes.