Pony.ai, inDrive and NVIDIA plan Kazakhstan robotaxi feasibility project

Pony.ai, inDrive and NVIDIA announced a memorandum of understanding on October 23 to explore autonomous ride-hailing in Kazakhstan, with Astana selected as the proposed first city. The partners will assess whether Level 4-ready vehicles can be offered through the inDrive app. The agreement establishes a feasibility project across vehicle technology, service operations, regulation and commercial requirements; it does not announce that a passenger service has already opened.

The proposed division of work gives Pony.ai responsibility for supplying its seventh-generation robotaxis, while inDrive would provide the app and operate the service. NVIDIA would contribute AI infrastructure, with the companies identifying Hyperion and the Halos safety system as part of the project’s technical foundation. This brings together a vehicle developer, a mobility operator and a computing supplier rather than placing the entire service inside one company.

Gen-7 is an existing Pony.ai technology generation. The company introduced the system at the Shanghai auto show in April 2025 alongside vehicles developed with Toyota, BAIC and GAC. That announcement emphasized automotive-grade components and a modular architecture that could be adapted across different vehicle models. It provides background for the Kazakhstan proposal, but does not identify which of those vehicle models would be used in Astana or establish their local operating approval.

NVIDIA describes Hyperion as a reference architecture joining vehicle sensors and onboard computing. Its current platform documentation discusses a standardized sensor suite, dual DRIVE AGX Thor chips, and integration across cameras, radar, lidar and other sensing hardware. The purpose is to give developers a common foundation for building and validating autonomous systems. Those general specifications describe NVIDIA’s platform offering; the memorandum does not disclose a final bill of materials for the proposed Kazakhstan fleet.

The supporting safety architecture extends beyond the driving model. NVIDIA describes Halos as covering vehicle architecture, system software, algorithms and the partner ecosystem, with evaluation and inspection tools used during development and deployment. In a robotaxi project, that distinction matters because generating a plausible driving trajectory is only one component of a complete vehicle system. A platform designed to support safety work cannot, by itself, establish that a particular service is authorized on local public roads.

Alpamayo supplies another part of the technical context. NVIDIA’s product documentation describes a family of reasoning models and software that combines interpretation of road scenes with explanations of driving decisions. Its open development platform includes models, simulation frameworks and datasets, with support for adapting systems to proprietary driving data. Such tools are intended to help developers examine unusual situations and customize behavior, rather than assume that one unchanged model will handle every operating environment.

The Alpamayo software stack also separates reasoning from runtime checks. NVIDIA describes safeguards that assess candidate driving actions, along with simulation-based safety assessment and system software integrating perception and planning capabilities. That layered design helps explain why the new collaboration includes both AI infrastructure and a safety framework. However, the partners have not identified a particular Alpamayo model, training configuration or completed validation result for the Astana project, leaving those implementation details outside the announced agreement.

A launch would follow an agreement with Kazakhstan’s government, making that agreement the next operating prerequisite in the announced plan. The official statement leaves the opening date, initial fleet size and passenger fares unspecified. The partners intend to assess expansion to other Kazakh cities and additional inDrive markets after initial deployment in Astana, using the first project to evaluate whether the service approach could work elsewhere.