Serve Robotics Inc. (SERV)

Sidewalk delivery robots at real fleet scale and triple-digit revenue growth — burning roughly six dollars for every one it earns.

Investment Research · Photoncap-style deep dive · v1 of "Serve Robotics" · May 14, 2026

What Serve Robotics physically does

Serve Robotics operates a fleet of autonomous, four-wheeled, cooler-sized delivery robots that travel on sidewalks to carry restaurant and convenience orders the "last mile" — typically under a couple of miles — from a merchant to a consumer's door. The robot is a low-speed (walking-pace) electric vehicle with a lockable insulated cargo bay, a sensor stack of cameras, ultrasonics and lidar, and an onboard compute system running Level 4 autonomy within mapped service areas. It is purpose-built for one job: moving a small payload along a pedestrian path without a human driver and without a car.

The business model is the part that matters for the theme. Serve is not selling robots; it operates them and charges per delivery, increasingly blended with a software-services layer. In Q1 2026 software services contributed roughly one-third of revenue and just under half of total revenue was recurring, per the company's May 7, 2026 results release — a deliberate shift from "we own and run robots" toward "we also license the autonomy and operations stack." The binding-constraint argument for Serve is weak relative to the supply-chain names in this theme: sidewalk delivery is a service business with thin unit economics, not a chokepoint that everyone else depends on. Serve's edge, such as it is, is operational density and a multi-year head start on permitting, mapping and merchant integration in specific US metros.

The robot is general-purpose only within a narrow envelope — it does not manipulate, it does not enter buildings, it does not handle large or heavy loads. It is a logistics endpoint, and its economics live or die on deliveries-per-robot-per-day and the cost of the human remote-supervision overhead behind the "autonomous" fleet.

Product roadmap

Serve's hardware has progressed through generations of its sidewalk robot, with the current-generation unit (the third major design, scaled through 2025) being the platform underpinning the ~2,000-robot fleet referenced on the Q1 2026 call. The 2025 build-out was a fleet-expansion phase — the explicit Q1 2026 message was that the company has now "shifted from fleet expansion to increasing revenue per robot," per the May 7, 2026 release. That is the roadmap pivot: 2024-2025 was about putting robots on sidewalks; 2026 is about utilization.

The geographic roadmap: the operating footprint reached 44 cities across 14 states as of Q1 2026, driven by new metro launches, hospital-network deliveries (an indoor/campus extension of the model) and densification of existing markets. Cumulative deliveries across indoor and outdoor environments are approaching 2 million, per the company. The software roadmap — the more interesting one — is the productization of Serve's autonomy and fleet-operations stack as a licensable layer, which is what drove the software-services revenue to ~one-third of the Q1 mix.

What Serve does not do: it does not make humanoids, it does not do drone delivery, and it does not sell hardware to third parties at scale. It is a single-modality, single-use-case operator. The Uber relationship — Serve's roots trace to Uber, which remains a strategic shareholder and a demand channel via Uber Eats — is a distribution asset, not a product.

The financial print

Serve reported Q1 2026 on May 7, 2026: revenue of $3.0 million, up 238% sequentially and 578% year-on-year. The eye-catching line is that Q1 2026 revenue alone exceeded the company's entire FY2025 revenue — real growth, off a tiny base. Fleet revenue was nearly $2 million (roughly a 10x YoY increase) and software services made up the rest, with just under half of total revenue now recurring. The company reiterated FY2026 revenue guidance of $26 million and held non-GAAP operating expense guidance of $160-170 million for 2026.

That guidance pairing is the whole story: ~$26 million of revenue against ~$165 million of operating spend. Serve is spending roughly six dollars for every dollar it earns. Forward P/E is negative and not meaningful; this is a cash-burn-and-scale name. The offsetting fact is the balance sheet — Serve reported $197.4 million of liquidity at Q1 2026, which at the guided burn rate funds roughly a year-plus of operations before another raise is needed. There is no named broad sell-side consensus of the Hanmi variety here; coverage is thin and the stock trades on fleet-growth and revenue-per-robot headlines.

One-year return: the stock has been volatile and is roughly flat-to-down over the trailing year, with RSI 47.8 and price 1.8% below the 50-day moving average — a neutral, un-extended tape, unusual for this theme. Next earnings: Q2 2026, expected on or around August 6, 2026 — the binary on whether the "revenue per robot" pivot is actually lifting utilization.

Customer mix today

Serve's "customers" are best understood in two layers, and the company does not disclose a clean named-customer percentage breakout, so the figures below are reporting-basis. The demand layer is restaurants and merchants whose orders Serve fulfills — aggregated substantially through Uber Eats (Serve's largest demand channel and a relationship inherited from its Uber origins) plus a direct merchant network and, increasingly, hospital and campus clients. The revenue layer splits differently: roughly two-thirds fleet revenue (per-delivery economics) and one-third software services as of Q1 2026, per the May 7, 2026 release.

The structural shift worth flagging: in 2024-2025 Serve was almost entirely a per-delivery fleet operator; by Q1 2026 software services had grown to ~one-third of revenue and recurring revenue to just under half. That mix shift toward higher-margin, more predictable software is the most genuinely encouraging trend in the print — it is the difference between "low-margin gig logistics" and "autonomy platform." Whether it sustains is the open question; one quarter is not a trend.

What's actually happening at the merchant / Uber Eats demand layer

The mechanism that decides Serve's economics is deliveries-per-robot-per-day, and the Q1 2026 pivot language — "from fleet expansion to increasing revenue per robot" — is management conceding that utilization, not fleet size, is now the constraint. With ~2,000 robots deployed across 44 cities, the fleet is large enough; the question is whether each robot is busy enough to cover its capital and supervision cost.

The Uber Eats channel is the key demand pipe: it gives Serve order volume without Serve having to build consumer demand, but it also caps Serve's pricing power and margin — Serve is a fulfillment subcontractor inside someone else's marketplace. The hospital-network and campus additions in Q1 2026 are an attempt to diversify into denser, more repetitive, higher-utilization environments where a robot can run more deliveries per shift. The honest read as of May 2026: Serve has proven it can deploy robots and grow revenue 578% YoY off a tiny base; it has not yet proven the per-robot economics work at scale, and the ~$165M-to-$26M spend-to-revenue ratio says the unit economics are nowhere near self-funding. The August 6, 2026 Q2 print is where "revenue per robot" either starts climbing visibly or it doesn't.

The competitive threat

Serve's competitive set is fragmented and the threat is less a single named rival than the structural question of whether sidewalk delivery is a defensible business at all. Direct sidewalk-robot competitors include Coco (backed by Sam Altman, operating in similar US metros), Starship Technologies (campus-focused, large cumulative delivery count), and Avride and others. None has obvious decisive scale over Serve, and Serve's ~2 million cumulative deliveries and 44-city footprint put it among the leaders by operational density.

The more serious threat is from above and from the side. From above: the delivery platforms themselves — Uber, DoorDash — could in-source or multi-source autonomy, and DoorDash has its own autonomous delivery efforts; Serve's dependence on Uber Eats as a demand channel is also a dependence on a party that could become a competitor. From the side: drone delivery (Zipline, Wing) attacks the same last-mile job with different economics, and conventional gig drivers remain the cheap incumbent. There is no material IP litigation driving the thesis. The competitive bottom line: Serve is a credible operator in a niche that may simply be a low-margin niche, and its biggest "competitor" is the possibility that no one makes good money doing this.

The terminal risk

The terminal risk for Serve is that autonomous sidewalk delivery is structurally a low-margin, capital-intensive service that never earns its cost of capital — not that a new technology obsoletes it, but that the business itself is the wrong shape. A sidewalk robot is a depreciating physical asset that needs maintenance, remote human supervision, permitting, mapping and insurance, to move a single small payload at walking pace. If deliveries-per-robot-per-day plateaus below the level that covers all of that, scale makes the losses bigger, not smaller.

The secondary terminal risk is channel: if the delivery platforms in-source autonomy or squeeze fulfillment margins, Serve becomes a commoditized subcontractor. And the form-factor risk — drones for some routes, cheaper gig labor for others — means sidewalk robots may end up serving only a thin slice of last-mile geography. Serve has a credible roadmap toward the higher-margin software-services layer, which is the only real answer to this risk; if that layer scales, Serve becomes an autonomy platform rather than a robot operator. As of May 2026 that pivot is one quarter old. The multiple you can pay is constrained by the genuine possibility that the core service never reaches positive unit economics.

Bull / Gap / Optionality

Bull

1. Real, verified scale and growth. Q1 2026 revenue of $3.0 million was up 578% YoY and exceeded all of FY2025 revenue, with a ~2,000-robot fleet across 44 cities and nearly 2 million cumulative deliveries, per the May 7, 2026 release. This is no longer a science project — it is an operating fleet at meaningful scale, which is more than most early robotics names can claim.

2. The software-services mix shift is the right kind of growth. Software services reached ~one-third of revenue and recurring revenue just under half by Q1 2026. If Serve can keep shifting toward licensing its autonomy/operations stack, the margin profile improves structurally — this is the path from "gig logistics" to "platform."

3. The balance sheet funds the experiment. $197.4 million of liquidity at Q1 2026 against ~$165 million of guided 2026 opex gives roughly a year-plus of runway. Serve is not at a financing cliff, and the un-extended tape (RSI 47.8, slightly below the 50-day MA) means you are not chasing.

4. Distribution via Uber Eats is a genuine asset. Serve gets order volume without building consumer demand, and Uber remains a strategic shareholder. For an early operator, having a demand pipe already plumbed is a real de-risking of the go-to-market.

Gap

1. The spend-to-revenue ratio is brutal. FY2026 guidance pairs ~$26 million of revenue with ~$160-170 million of opex — roughly six dollars spent per dollar earned. Even with triple-digit growth, the path to self-funding is years away and depends on utilization economics that are unproven at scale.

2. Utilization, not fleet size, is now the binding constraint — and management said so. The Q1 2026 pivot language ("from fleet expansion to increasing revenue per robot") is an admission that the ~2,000-robot fleet is under-utilized. Until deliveries-per-robot-per-day visibly climbs, the fleet is a cost, not an engine.

3. Channel dependence cuts both ways. Reliance on Uber Eats for demand caps Serve's pricing power and exposes it to a partner that could in-source autonomy or become a competitor. Serve is a subcontractor inside someone else's marketplace.

4. The niche may simply be low-margin forever. Sidewalk delivery faces drones from one side and cheap gig labor from the other, and no operator in the category has demonstrated attractive unit economics. The terminal risk is that the business is the wrong shape, not that Serve executes it badly.

Optionality

The trade

Serve is a small, speculative, watch-it-prove-the-economics name, and the trade reflects that. Entry zone is current ±5%, roughly $8.51–$9.41 — and the tape genuinely cooperates here, with RSI 47.8 and price slightly below the 50-day moving average, so this is one of the few names in the theme you are not chasing into a melt-up. Size small, around 0.75% of risk capital, treating SERV as a high-variance call option on sidewalk-delivery economics rather than a core holding; the spend-to-revenue ratio means a wrong thesis here is a slow bleed, not a sudden gap. Stop at roughly $7.10, below recent structural support. The defining catalyst is the Q2 2026 print around August 6, 2026 — specifically whether the "revenue per robot" pivot translates into visibly rising utilization, because fleet scale is already proven and utilization is the only thing left to prove. If the thesis is "autonomous last-mile logistics is real," the cleaner expression is a diversified or platform-layer exposure rather than a single-modality sidewalk operator; Serve is the purest sidewalk-robot bet available, with all the concentration risk that implies. Conviction: 5 / 10.

Sources referenced inline throughout. Reference v1 of this template format: _Watchlist/hanmi-photoncap-style.md.

\newpage

454910 — Doosan Robotics / 두산로보틱스 (454910.KS) · SKIP / WAIT (Tier-3) · Conv 4/10 · Bucket C

ticker: 454910 name: Doosan Robotics / 두산로보틱스 (454910.KS) theme: Robotics bucket: C conviction: 4 entryzonelo: 101460 entryzonehi: 112140 currentprice: 106800 pricedate: 2026-05-14 positionsizepct: 0.5 stoploss: 88000 thesisoneline: A leading cobot maker whose revenue is shrinking and whose humanoid roadmap is still a press release, valued as if both are growth stories. catalystnext: Q1 2026 earnings catalystdate: 2026-05-15 deepdivepath: Theme -- Robotics/454910/454910-deep-dive.md lastupdated: 2026-05-14T00:00:00Z rsi: 61.5 vs50ma: 16.2 forwardpe: 0.0 themecycleposition: early customermixsummary: Diversified global cobot channel — distributors and SME end-users across North America, Europe, Korea; no single dominant customer; humanoid customers do not yet exist. terminalriskoneline: Cobots get commoditised by lower-cost Chinese makers while the humanoid pivot never produces a shipping product, leaving Doosan stuck between two losing positions. bulldriverscount: 5 gapriskscount: 4 optionalitycount: 6 lastearningsdate: 2026-02-12 nextearningsdate: 2026-05-15