SERV — Serve Robotics, Inc.

Serve Robotics has deployed 2,000 autonomous sidewalk delivery robots across 20 U.S. cities, growing its fleet 20-fold in three years — while revenue per robot has declined from approximately $4,600 annually in 2024 to roughly $1,350 in 2025, demonstrating that deploying more robots is compressing unit economics rather than improving them. Amazon's Scout autonomous delivery robot program operated across multiple U.S. cities for four years before the team was disbanded in October 2022 — ending in the same conclusion every observer of sidewalk robotics arrives at independently: the economics of urban outdoor autonomous delivery have not been solved, and capital deployment cannot substitute for a working unit economic model. Avoid.

---

Autonomous delivery has been the next transformational logistics category for approximately a decade. The pitch is intuitive: eliminate the human courier, whose fully-loaded cost runs $8 to $10 per delivery, with a robot that can execute the same task for pennies once it reaches scale. The arithmetic is compelling on paper. In practice, the category has produced a consistent record of failed or stalled companies, pivots to controlled environments, and one high-profile shutdown by a company that had more capital and data than any competitor will ever have: Amazon. The cautionary weight of Amazon Scout's four-year lifespan — launched January 2019, shut down October 2022 with the team disbanded — is not a data point that advocates of sidewalk autonomous delivery have found a satisfying answer to. If Amazon could not solve the economics with unlimited resources, the question is not whether the technology will eventually work but whether it will produce investable economics within the capital constraints of a public company whose cash runway is measured in years, not decades.

The last-mile delivery market is large and growing. The U.S. food delivery market alone processes tens of billions in gross merchandise value annually. The logistics cost of the last mile represents a structurally significant portion of total delivery cost, and the human driver shortage, minimum wage increases, and gig economy friction create real economic pressure on delivery platforms to find alternatives. The opportunity is genuine. The competitive structure of the companies trying to capture it ranges from well-funded startups to the logistics arms of the largest technology companies in the world, and the category has so far produced one meaningful at-scale commercial player: Starship Technologies, whose fleet of 2,700 robots has executed more than 9 million autonomous deliveries across 270 locations. Starship's location profile is instructive: the overwhelming majority of its deployments are on university campuses and controlled suburban environments — settings where the robot encounters consistent surfaces, predictable pedestrian behavior, manageable weather conditions, and a concentrated delivery demand that allows high daily utilization per robot. The gap between that controlled environment and the urban sidewalk network that Serve Robotics is targeting is not a minor operational detail. It is the central reason Amazon decided the category was not viable.

Serve Robotics spun out of Uber in 2021 to develop and operate autonomous delivery robots for the urban sidewalk environment. Its Generation 3 robot — now the basis of the 2,000-unit fleet the company has deployed across Los Angeles, Atlanta, Dallas-Fort Worth, Miami, Chicago, and Fort Lauderdale — represents a claimed 65% reduction in unit hardware costs versus the prior generation. The company has partnerships with Uber Eats and DoorDash, operates within a network of more than 4,500 restaurant and retail partners, and counts NVIDIA among its investors. In Q4 2025, the company reported 99.8% delivery completion rates across the fleet and crossed the 100,000 total delivery milestone. The company is also exploring international expansion into Toronto, Sydney, Tokyo, Madrid, and London, and acquired Diligent Robotics in 2025 to add recurring revenue from hospital service robots.

The competitive moat thesis for Serve Robotics rests on three claims: data compounding (more deliveries create better autonomy models that create more deliveries), network effects (platform integrations with Uber Eats and DoorDash create demand density that smaller competitors cannot access), and hardware scale (Gen3's lower unit cost creates a cost advantage as the fleet expands). Each claim is a story about advantages that will exist in the future, contingent on the fleet reaching a scale and utilization level that the current numbers do not yet support. The data compounding thesis is the most interesting, and the most difficult to evaluate: it depends on urban outdoor delivery being a tractable autonomy problem, on Serve's deployment data being sufficiently diverse and dense to improve meaningfully, and on improvements in autonomy actually translating to lower cost per delivery rather than simply better performance metrics that do not affect the P&L. These are large conditional bets stacked on one another.

The competitive table that matters for this business is not a comparison of gross margins or customer retention — the business is too early for those metrics to be meaningful. The table that matters shows whether the unit economics are improving as the fleet scales, which is the fundamental question on which the entire investment thesis rests.

Year	Fleet (year-end)	Revenue ($M)	Revenue/Robot/Yr ($)	Adj. EBITDA ($M)	Cash ($M)
FY2023	~100	~$0.2	~$2,000	est. neg.	est. $300+
FY2024	~600	~$1.85	~$4,600 (avg. fleet)	est. neg.	est. $300+
FY2025	2,000+	$2.7	~$1,350 (avg. fleet)	-$28M (Q4 run rate)	$260
FY2026E	2,500–3,000	$26 (guidance)	~$10,000+ (required)	TBD	est. $150–180

The revenue-per-robot column is the diagnostic. From FY2024 to FY2025, the fleet grew by more than 3× — from approximately 600 robots to 2,000. Revenue grew only 46%, from roughly $1.85 million to $2.7 million. The result is that average annual revenue per robot declined from approximately $4,600 in 2024 to approximately $1,350 in 2025 — a 70% compression in per-unit productivity as the fleet scaled. This is the opposite of what a working network effect or data compounding story produces. In a business whose moat thesis depends on scale creating better economics, the data shows scale creating worse economics. The management explanation is that newly deployed robots operate below steady-state efficiency during their ramp-up period and that fleet cohort maturity drives improving economics over time. This may be true. But 2,000 robots deployed across six metropolitan areas, with over 100,000 total deliveries completed, is not an early ramp-up sample. It is the product after four years of operation and substantial capital investment.

The FY2026 guidance of $26 million is the number that requires the most scrutiny. It represents a 10× increase from FY2025 revenue of $2.7 million — an acceleration that the Q4 2025 run rate ($0.9 million, or approximately $3.6 million annualized) does not support. Approximately $7 million of the $26 million is expected from Diligent Robotics, the hospital robot company acquired in 2025, and approximately $7 million from software and advertising. The remaining $12 million in delivery robot revenue — against a Q4 2025 annualized run rate of roughly $3.6 million — implies a 3× improvement in delivery robot productivity during calendar year 2026. For context, the company went from $1.85 million to $2.7 million in FY2025 — 46% growth — while tripling the fleet. To reach the implied delivery revenue in 2026 at the same fleet size would require per-robot revenue to improve from $1,350 per year to over $5,000 per year, which represents the largest annual improvement in per-robot productivity the company has ever demonstrated, against a backdrop where the trend has been moving in the opposite direction.

The Q4 2025 adjusted EBITDA loss of $28 million on $900,000 in revenue represents a 3,100% EBITDA loss ratio — ninety-three cents of operating loss generated for each dime of revenue. Annualized, this implies over $110 million in EBITDA burn per year. Against the $260 million cash balance reported at end of Q3 2025, the runway at current burn rates is approximately two to two and a half years before the company requires additional equity capital, assuming no improvement in the EBITDA trajectory. The comparison to NNDM, which consumed $330 million in cash over nine months on $67 million in revenue, is not exact — Serve Robotics is at a much earlier commercial stage — but the structural dynamic is comparable: a capital-intensive technology company deploying hardware at scale without yet demonstrating that the underlying unit economics justify the capital required.

CEO Ali Kashani, PhD, co-founded the company and has led it through its Uber spin-out, its public listing, and the Gen3 robot launch. The Diligent Robotics acquisition represents a strategic expansion into healthcare delivery robots — Moxie hospital robots that transport materials within hospital facilities — and is expected to contribute $7 million of the $26 million 2026 guidance. The acquisition adds recurring revenue from a more controllable deployment environment than urban sidewalks, which is directionally rational. It is also an implicit acknowledgment that the delivery robot business at its current scale is insufficient to build a company on. A company confident in its primary business does not typically diversify into a different vertical robot category within its first full year of commercial operation. The Moxie revenue is real; the question is whether it represents a path to profitability or a revenue bridge that sustains the delivery robot narrative while the core thesis remains unproven.

The growth runway argument that Serve Robotics' advocates make rests on the 2,000-robot fleet as an inflection point — the level at which pilots become repeatable, data density compounds, and per-robot utilization reaches the levels required for positive unit economics. The company has stated that sub-$1 per-delivery cost is achievable at scale, versus the $8 to $10 human courier cost. At $79 monthly subscription revenue per robot (an approximation implied by current revenue levels), sub-$1 delivery economics at, say, 5 deliveries per day would generate approximately $1.50 in delivery revenue against $0.30 in direct cost — a 5:1 gross revenue-to-variable-cost ratio. That is the model. It is plausible as a mathematical abstraction. It requires: sufficient delivery density per robot per day, weather and operational conditions that allow consistent utilization, restaurant and consumer acceptance of robot delivery as a primary channel, and no safety incidents that trigger regulatory restrictions. Amazon Scout's operational experience across four years suggests that the real-world friction between that abstraction and actual deployment economics is substantially larger than the model implies.

Serve Robotics has captured approximately 100,000 total deliveries since inception against Starship Technologies' 9 million — approximately 1% of the leading competitor's volume — while operating in a more demanding urban environment that produces more operational failures per deployment than the campus settings where Starship is dominant. Starship's 270 campus locations give it a 9× geographic advantage in deployments that generate high-density, high-utilization economics. The urban sidewalk environments where Serve operates — where robots have blocked firetrucks, gotten stuck on train tracks in Miami requiring emergency train stops, blocked wheelchair accessibility ramps in Chicago, and broken down in minimal snow conditions — are structurally harder to operate in than the campus settings where sidewalk robotics has actually achieved commercial viability. The 4,500-plus restaurant partners are an impressive headline number; the caveat from the research is that several restaurants have dropped partnerships or have not scaled beyond initial testing, which is the leading indicator of what the 4,500 number actually represents in durable commercial relationships.

At approximately $9.75 per share with roughly 74.7 million shares outstanding, the market capitalization is approximately $728 million. Against $260 million in cash, the implied enterprise value for the business operations — the delivery robots, the Diligent Robotics healthcare robots, the software platform, the DoorDash and Uber Eats integrations, and the international expansion pipeline — is approximately $468 million. That $468 million is being paid for a business that generated $2.7 million in FY2025 revenue, is burning over $110 million annually in EBITDA, and whose per-robot productivity is declining as the fleet scales. The ratio of implied business value to trailing revenue is approximately 173×. Even accepting the FY2026 guidance of $26 million at face value — which the Q4 2025 run rate does not support — the implied enterprise value is 18× one-year-forward revenue, for a business that is nowhere near profitability and requires 3× revenue growth in a single year that it has never demonstrated the ability to achieve. This is a venture capital valuation dressed in public market clothes.

The intelligent bull on Serve Robotics argues that the data compounding thesis is real, that NVIDIA's involvement signals genuine technology credibility, and that the delivery economics at scale are qualitatively different from the current run rate in the same way that any infrastructure business looks terrible during the capital deployment phase before utilization reaches break-even. This argument is the best version of the bull case. The answer is that Amazon Scout operated for four years in multiple markets and reached neither utilization break-even nor NVIDIA-grade autonomy levels. The difference between Amazon's Scout and Serve Robotics is not that Amazon lacked technological sophistication — it was among the most AI-capable technology organizations on earth during the Scout program. The difference is that the urban outdoor delivery environment is harder than it looks, and no amount of data compounding has yet solved the weather, the wheelchairs, the train tracks, and the economics of a robot that needs to deliver food profitably at $79 per month per unit.

The conclusion does not require a price-dependent qualifier. Revenue per robot declining at scale is not a feature of the ramp-up period — it is the data speaking. Amazon said the same thing after four years and walked away. At any price that assigns meaningful value to the delivery robot thesis above the cash balance, the risk-reward is unfavorable.

The data said one thing in 2024 and the same thing in 2025, more clearly: more robots, less revenue per robot. The thesis requires that trend to reverse by 7× in a single year. It has never reversed before.