Serve Robotics Opens Its Fleet Data to Other Robot Makers, With 917 Daily Robots and a $59.2 Million Quarterly Loss

Serve Robotics reported $2.8 million of revenue for the quarter ended September 30, 2026, down from $3.2 million in Q2, and a net loss of $59.2 million. The sidewalk-delivery company announced the numbers on November 5 alongside Serve Physical AI Services, a product line that sells the software behind its fleet to other robot makers.

The number that matters here is not the 313% year-on-year growth in the press release but the denominator: an average of 917 daily active robots produced $2.6 million of fleet-services revenue in the quarter. That is about $2,800 per robot over roughly 92 days, or about $31 a day, by this publication's arithmetic from the company's figures.

What Physical AI Services contains

TechCrunch, which covered the launch, lists three products. Serve Link connects robots in the field to AI systems running in data centres. A data curator tool picks out edge cases in real-world robot data for model training. A foundation-model data catalog is built from what Serve's own sidewalk robots collected.

Bedrock Robotics, which builds self-driving systems for construction equipment, is a customer, per TechCrunch, and T-Mobile is a partner on moving AI computing nearer to the robots. Neither the press release nor TechCrunch gives a price, a contract value or a revenue target for the new line. Dr. Ali Kashani, Serve's co-founder and chief executive, told TechCrunch: "We are solving the problem of robots in human spaces."

The 8-K exhibit says the offering makes "the technology behind Serve's fleets" available to other robot makers. It does not break out Physical AI Services revenue, and software services as a whole were $0.3 million of the $2.8 million.

The quarter in numbers

Metric	Q2 2026	Q3 2026
Revenue	$3.2 million	$2.8 million
Gross loss	$8.8 million	$7.6 million
Net loss	$64.1 million	$59.2 million
Adjusted EBITDA	-$44.5 million	-$42.2 million

The loss narrowed in every line, and revenue fell. Serve attributes a fleet margin improvement of more than 100 points sequentially, though a gross loss of $7.6 million on $2.8 million of revenue means each dollar sold cost roughly $3.70 to deliver in cost of revenue. Brian Read, the chief financial officer, said "This quarter was about execution."

Fleet size, and the Uber question

Serve has deployed more than 2,000 robots in the United States since its 2021 spinoff from Uber, supporting more than 4,000 restaurants, per the company. The working fleet is smaller: daily active robots averaged 917, up from 792 in Q2 and 312 a year earlier.

1. Q3 2025312 robots
2. Q2 2026792 robots
3. Q3 2026917 robots

Source: Serve Robotics Q3 2026 results, SEC Form 8-K exhibit 99.1, accessed 2026-11-07

The gap between 2,000 deployed and 917 active is the first place to look when a deployment count is quoted; Serve does not say how many of the rest are retired, in storage or between cities. TechCrunch reports that the delivery business is moving away from Uber toward DoorDash and Wonder. The company says daily volume among its new delivery partners tripled in Q3 and merchant count grew 2.5 times, and that it launched exclusively in its first international city, in Australia, in October.

The same release says Serve acquired Diligent Robotics in 2026, extending it into indoor hospital robots, with 12 multi-year hospital contract extensions signed and 4 new hospitals added so far. Recurring revenue was about 69% of the total.

Guidance and cash

Serve raised its 2026 revenue guidance to $12 million from $9 million to $10 million, and cut its guidance for 2026 non-GAAP operating expenses to $150 million from a range of $160 million to $170 million. It ended September with $198.0 million in cash and marketable securities, against a net loss that has run between $33.0 million and $64.1 million a quarter in the periods the release compares.

At Q3's adjusted EBITDA loss of $42.2 million, $198.0 million covers a little under five quarters before financing; that is arithmetic, not a company forecast. The lower expense guidance is what would stretch it.

The loss works out to $0.68 a share on about 88 million shares outstanding. Branding revenue grew nearly 80% from Q2, per the release, but the release gives no dollar figure for it, so the effect on the $2.8 million total cannot be isolated.

Compare this with the other physical-AI listings. Agility Robotics came to market on Nasdaq with humanoids, and Boston Dynamics sells a warehouse robot with a rated throughput. Serve is the one that already reports a fleet with a daily active count and a revenue-per-robot you can compute, which is also why the number looks thin.

What would change this read

If Serve discloses Physical AI Services revenue as its own line in Q4, with a customer count above Bedrock Robotics, the data business would be real and the per-robot figure the wrong lens. If daily active robots stop rising while the 2,000 deployed figure grows, the sidewalk business is saturating. Serve has not given a date for its Q4 report.