Analysis: adding platforms while one platform shrinks

The Grubhub partnership and the Uber Eats guidance cut are the same story told from two ends. Serve was spun out of Uber in 2021 and its delivery volume has depended on a small number of marketplaces. The revenue concentration table in the Form 10-Q shows how small: one customer at 22% of revenue in the quarter, another at 11%, and a customer that supplied 39% of revenue in the second quarter of 2025 no longer above the 10% disclosure line at all. A company whose demand arrives through other people’s apps has to keep adding apps, and that is what the August announcement does.

The fleet numbers show where the constraint now sits. Daily active robots and daily supply hours both fell sequentially while revenue rose 9%, so the quarter’s growth came from getting more revenue per robot hour rather than from more robot hours. Advertising is the reason the company gives: it says advertising made up nearly 50% of food delivery revenue and recurring revenue exceeded 50% of the total. That is a real change in the business model, and it means the delivery fleet is increasingly a distribution surface for brand campaigns as well as a courier.

The micro depot is the operational answer to the same problem. Full depots are fixed costs that only pay back at volume; if volume through any one marketplace can be revised away in a single quarter, the ability to open and close a site cheaply matters more than the ability to open a large one. Whether that works is measurable, and the measurement will show up in daily supply hours per market rather than in press releases.

Two constraints bound how fast any of this can move. The first is regulatory and local: the DC permit terms are typical of the category, and each new city sets its own weight, speed and right-of-way conditions, so market count is not a linear function of engineering. The second is financial. Serve is spending at a rate that produced a $113,131 thousand loss in six months against $6,222 thousand of revenue, funded from a $240.4 million cash and securities balance. The reduced operating expense guidance extends that runway; the reduced revenue guidance shortens the distance the same capital covers in commercial terms.

Nothing in the announcement quantifies the Grubhub agreement. No order volume, no revenue share, no term. What a careful reader would watch is the third-quarter fleet metrics, since a partnership that adds restaurants without adding supply hours is adding options rather than deliveries, and the revenue concentration table in the next Form 10-Q, which will show whether the new platform has reached the 10% line.

What the documents say

Serve Robotics Inc. (Nasdaq: SERV) said on August 17, 2026 that its sidewalk robots will begin taking orders from the Grubhub marketplace, and that it has opened two new metropolitan markets. The announcement, filed with the Securities and Exchange Commission as an exhibit to a Form 8-K, bundles a delivery platform partnership, a geographic expansion, a new site format, a hospital robot upgrade, a hardware preview and an advertising product into one release.

The Grubhub deal comes through Wonder, the food technology company that owns Grubhub. Robot delivery starts in Chicago, Los Angeles and Alexandria, with more than 100 participating Grubhub merchants in Chicago and nearly 200 in Los Angeles. Wonder’s Alexandria location will also offer delivery through Serve’s network.

Two markets, and a cheaper way to open them

Washington, DC and San Jose, California are Serve’s seventh and eighth major U.S. markets, both opened with DoorDash. The company puts the combined population of the two metros at 8 million and lists Los Angeles, Chicago, Atlanta, Dallas and Miami as the existing map. San Jose, its first Bay Area market, has completed a first month of deliveries. In Washington the robots will work Dupont Circle and parts of downtown.

Alongside the expansion, Serve is opening its first micro depot, in Miami. The company describes these as small-footprint sites handling staging, charging, dispatch and maintenance without the build-out time of a full facility, and presents them as a repeatable way to enter neighbourhoods faster and at lower cost.

The District of Columbia sets the terms under which the robots may operate there. The District Department of Transportation permits personal delivery devices under a Public Right of Way Occupancy Permit whose conditions were drawn from a pilot programme that ended in 2018. Devices may not exceed 10 miles per hour, may not weigh more than 275 pounds excluding cargo, must stay on sidewalks, crosswalks and alleyways, must follow pedestrian traffic rules while always yielding to pedestrians, and must have a process for swift removal in the event of mechanical malfunction.

The rest of the announcement

Diligent Robotics, which Serve acquired in January 2026, has begun rolling out a next-generation Moxi hospital robot to health systems including Endeavor Health Edward Hospital near Chicago, Providence Saint John’s Health Center in Los Angeles and Children’s Hospital Los Angeles. Serve says the robot perceives and interprets its surroundings up to 15 times faster, carries 10 times the onboard compute, runs for up to 18 hours and charges 30% faster, without changes to hospital infrastructure. The company says the update draws on deliveries completed across more than 25 U.S. hospitals.

Serve also previewed Beacon, a countertop device with built-in cellular that alerts restaurant staff when a robot arrives and requires nothing from the restaurant beyond power, and launched an advertising product called Characters, which adds interactive personalities to the robot wraps brands already buy. The first is Chomp, a talking hamburger-wrapped robot made with Grubhub.

What the quarter showed

The expansion follows second-quarter results reported on August 6, 2026. Revenue was $3,238 thousand, up 9% sequentially and 404% year on year. Fleet services contributed $2,305 thousand and software services $933 thousand, against $330 thousand and $312 thousand in the same quarter of 2025. For the half, revenue was $6,222 thousand against $1,082 thousand.

Operating metrics moved the other way. Daily active robots averaged 792 in the quarter, down from 812 in the first quarter, though far above the 160 of a year earlier. Daily supply hours averaged 9,809, down from 10,295 sequentially and up from 1,723 a year earlier. Both figures now include the indoor fleet acquired with Diligent.

The company cut its full-year revenue guidance to a range of $9 million to $10 million, citing lower than expected delivery volume through its Uber Eats partnership, including a decline recorded in the second quarter and the removal of projected demand for the second half. It also lowered expected non-GAAP operating expenses for the year to $140 million to $150 million from $160 million to $170 million. Chief financial officer Brian Read described the change as concentrating fleet and capital behind the highest-return opportunities.

Net loss was $64,127 thousand for the quarter, against $20,850 thousand a year earlier, and $113,131 thousand for the half. Loss per share was $0.80 against $0.36. Serve ended the quarter with $240.4 million in cash and marketable securities, total assets of $378,624 thousand and approximately 86 million shares outstanding.