Physical AI's uneven week: robotaxis scale, weeding robots retrench

Robotaxis scale

Pony.ai CEO James Peng laid out the company's scaling playbook: the company provides tens of thousands of rides daily, with robotaxis completing around 20 to 25 rides per vehicle per day on average. Peng, who has worked on autonomous driving for nearly 20 years, positions Pony.ai as the Virtual Driver component of the taxi ecosystem — and argues scaling is not just a technical challenge but also involves regulation, public acceptance, branding, and ecosystem partnerships. m:…2360

The cost story favors scale: autonomous vehicles rely on components like LiDARs, cameras, and sensors whose costs track manufacturing volume, and Peng credits the rapid growth of the electric vehicle industry with building a mature supply chain. The company began fully driverless operations on public roads in early 2023, is expanding through projects in Zagreb and Luxembourg, and has announced plans to bring robotaxis to London in partnership with Uber. The urban thesis: shared autonomous vehicles serve many passengers throughout the day while privately owned cars sit parked, which could reshape parking and urban space. m:…2360

The retrenchment

The other end of physical AI had a harder week. Carbon Robotics conducted layoffs, attributing the move to realigning costs amid broader economic conditions. The cuts followed a period of rapid headcount scaling, with reports that some staff were let go just weeks after starting — a whiplash pattern that suggests the company's cost structure ran ahead of its revenue durability. m:…9854

Delivery robots' report card

Serve Robotics will report third-quarter 2026 financial results on November 5, 2026, after market close, with a conference call at 2 p.m. PT. Since spinning off from Uber in 2021, the company says it has deployed more than 2,000 robots across the US, reaching a population of approximately 3 million and supporting delivery for more than 4,000 restaurants. In 2026 it acquired Diligent Robotics, expanding beyond sidewalk delivery into indoor service robots used in hospitals — and the company designs both the hardware and software behind its robots, which it says enables safe operation in public and private environments at scale. m:…0040

The training ground

Underneath both stories, the simulation layer keeps improving. NVIDIA expanded its Cosmos platform with training tools, simulation checkpoints, and Data Factory blueprints for physical AI world models; Porsche expects its Cosmos-based AV data flywheel to cut repeated physical test drives by up to 90 percent, and ANYbotics generated over 1,500 synthetic video clips with the blueprints. Cheaper simulation compresses development cycles — but it does not solve deployment economics, which is where this week's stories diverged. m:…8988

What it means

This is my read, not a reported fact: physical AI is bifurcating. Scaled operators with utilization economics — Pony.ai's 20 to 25 rides per vehicle per day — keep expanding into new cities with ecosystem partners, while capital-intensive hardware startups face cost realignment. For founders, the moat is shifting from the model to deployment density and partnerships: Uber for distribution, hospitals for indoor robots, simulation for development speed. Serve Robotics' November 5 results will show whether the delivery-robotics side of the ledger is converging on the same utilization story. Shared autonomous vehicles can serve more passengers than privately owned cars because of higher utilization, Peng noted — and that utilization math is the business model. m:…2360 m:…0040