inDrive Robotaxi Tests in Kazakhstan: What to Know
inDrive plans supervised robotaxi tests in Astana in 2027 with Pony.ai and Nvidia. See what the partnership means, the challenges, and what happens next.
inDrive's Astana Robotaxi Plan: What the 2027 Tests Will Prove
Robotaxis are moving beyond the markets people usually associate with autonomous driving. inDrive, the ride-hailing company known for letting passengers and drivers negotiate fares, says it will test Pony.ai vehicles in Astana, Kazakhstan, in the first half of 2027. Nvidia is also involved, with technology for autonomous driving and safety.
It's a serious plan - but it's not the same thing as a confirmed, commercial launch. The companies have signed a memorandum of understanding (MOU) to explore bringing Pony.ai robotaxis to inDrive's platform. Early rides are expected to include a safety operator, with driverless operation framed as a later goal. That distinction matters. A supervised test can show whether the system works in a specific environment, but it doesn't prove it's ready to run at scale or compete on price.
Astana is a demanding place to run those tests. Long, snowy winters can challenge sensors and how a vehicle makes decisions. At the same time, Kazakhstan is still building its rules for autonomous vehicles, which will shape what can be tested on public roads. Here's what the companies have said, how the partnership is structured, and what needs to happen before robotaxis become a practical option for riders.
What inDrive and its partners have announced
The plan brings together three companies with different roles. inDrive would connect the vehicles to its ride-hailing platform and operate them in Kazakhstan. Pony.ai would provide its Gen-7 robotaxis. Nvidia would contribute autonomous-driving technology, including its Hyperion platform and Halos safety system.
The companies have signed an MOU, which sets out an intent to explore the arrangement. It also means the announcement isn't a full checklist of what's been settled - commercial terms, technical setup, regulatory approvals, and the details needed to move from tests to an actual service. The companies have not said, for example, what the fleet size would be, what areas the robotaxis would cover, how much rides would cost, or how they plan to progress from supervised operations to driverless ones.
A test is not a public launch
According to inDrive CTO Yuri Misnik, testing is expected to start in Astana in the first half of 2027. The early phase would include a safety operator. Fully driverless rides are the long-term objective, but that depends on the results of the tests and the permissions in place at the time.
For riders, that's more than a technical footnote. A vehicle with a trained operator can be monitored and supported in ways a driverless service can't rely on. The test can help teams learn how the vehicle handles local roads and specific conditions - and how the app and day-to-day operations work together. What it can't do is automatically prove the same system is ready to run without anyone on board.
The announcement also doesn't say how riders would access the service. inDrive says Pony.ai's vehicles would be offered through its app, but the companies haven't published details like passenger eligibility, the service area, or the fare structure. Those are practical questions, and they'll matter as much as vehicle specs once the pilot begins.
How the proposal fits Kazakhstan's plans
The project is arriving as Kazakhstan develops its own framework for autonomous-vehicle testing. The country has said it wants autonomous taxi services in 2027, with early trials planned in Astana and two other cities. It has also introduced rules that allow autonomous vehicles to be tested on certain public roads.
That lines up with the idea of a commercial partnership and a national testing effort. But it doesn't mean specific routes, vehicle deployments, or driverless operations have already received all the needed approvals. Local rules, permissions, and safety requirements will determine what the companies can actually do on public roads.
For inDrive, the approach also reflects its operating model: instead of building everything internally, it plans to combine its platform and local operations with partners' vehicles and technology. That could reduce the work required to develop a full autonomous-driving stack in-house. It also means the project depends on multiple groups coordinating responsibilities in practice.
How the Pony.ai and Nvidia partnership is meant to work
A robotaxi service isn't just a self-driving car. It's a package: the vehicle, sensors, computing hardware, driving software, safety processes, fleet operations, rider access, and oversight from regulators. The announced partnership splits those parts across the companies involved.
Pony.ai is expected to supply its Gen-7 robotaxis. inDrive would run the service in Kazakhstan and place the vehicles on its platform. Nvidia is expected to provide technology, including Hyperion and Halos. In broad terms, the goal is to pair robotaxi driving capability with the ride-hailing service riders already use.
Nvidia's pitch: reduce integration work
Nvidia describes Hyperion as a base that can help companies access sensor and computing technology without sourcing and integrating every component from scratch. It also points to Alpamayo and tools for generating synthetic data to test autonomous systems across different driving scenarios.
The practical message is about time, cost, and integration. Reusing a tested hardware-and-software foundation can reduce some of the effort involved in assembling a complete stack from separate parts. But that doesn't remove the need to adapt the system for a new vehicle, local roads, operating requirements, and weather. A platform also doesn't automatically translate into safety or commercial viability. It can help provide building blocks, but the project still has to validate how the pieces work together in the intended environment.
Nvidia vice president of automotive Ali Kani said cost savings are especially important in emerging markets, where ride-hailing is already relatively affordable. That's the core economic tension here: autonomous systems still have to cover operating costs, while the service has to fit a customer base accustomed to lower fares.
The integration work that remains
The companies haven't published enough detail to confirm the exact technical configuration or where each partner's responsibility boundaries will land. That's not unusual at this stage, but it does mean the partnership shouldn't be read as a ready-made deployment plan.
Before a service can operate reliably, the partners will need to coordinate issues like:
- How the vehicle's software and hardware are set up for Kazakhstan's operating conditions.
- How safety operators will monitor the vehicle during the initial phase.
- What happens when the vehicle encounters a situation it can't handle with confidence.
- How the ride-hailing app communicates service availability, pickup points, and any operational limits.
- Who handles maintenance, incident response, and updates to the system.
These aren't minor details. A vehicle that can technically drive can still produce a bad customer experience if it's hard to dispatch, unreliable in bad weather, or confusing for riders. And a well-designed app can't fix a vehicle that frequently stops its route or depends on human intervention.
So the partnership is best understood as a division of work that still needs to be proven. The pilot's value will come from how the whole service performs, not just whether the vehicle can drive autonomously in select conditions.
Why Astana is a demanding test market
Astana can be useful for testing precisely because it isn't an easy environment. The city has roughly 1.7 million people and faces months of snow and ice. Winter conditions can affect what sensors can detect, how visible lane markings and lane edges are, and how a vehicle responds on slippery surfaces and changing road conditions.
inDrive founder and CEO Arsen Tomsky described the opportunity as a way to evaluate autonomous driving in tough winter conditions. If the technology can operate safely and consistently there, the results could inform deployments in other places with similar challenges. But harsher conditions also increase the chances that testing will uncover limitations - or force the service to operate with narrower boundaries.
Testing should measure the service, not just the drive
A meaningful pilot needs to cover more than whether a vehicle can complete a route. It should also show how often the system can run without intervention, how it handles shifting conditions, and whether it stays dependable for passengers over time.
In a winter city, teams will need to track how weather affects both the vehicle and its operating procedures. They'll also need to understand which parts of the environment the system can handle consistently and where it needs extra caution or human support. The public announcement doesn't spell out planned test routes or the metrics the partners will use, so those details are still unknown.
There's a trade-off. Testing in a demanding environment can reveal weaknesses that might not show up in milder conditions. At the same time, results from one city shouldn't be treated as a universal guarantee. Even if the system performs well on a set of Astana routes in winter, it would still need to be evaluated against different roads, traffic patterns, weather, and local rules before expanding elsewhere.
Regulation is part of the technical problem
Autonomous vehicles operate inside a regulatory and operational system, not outside it. Kazakhstan's rules for testing on certain public roads - and its stated plan for 2027 trials - provide a framework. The specific permissions and conditions for this project will still matter.
Misnik has said a favorable and collaborative regulatory environment is one of the main requirements for a successful launch. That's practical: regulators need to set conditions for safe testing, and companies need enough clarity to plan their work and react to what the tests show. If permissions are unclear or operational limits are too tight, it may be hard for a pilot to generate useful evidence. And if oversight is weak, the impact of failures could be more than just a delayed timeline.
The best result wouldn't just be a quick move to driverless rides. It would be a testing process that produces clear evidence, identifies where the system reaches its limits, and gives regulators and operators a solid basis for deciding what comes next.
The business case: affordability, competition, and scale
inDrive built its ride-hailing business around affordable trips and negotiated fares. Its robotaxi plan suggests it wants to keep that positioning as it moves into autonomous mobility. Whether it can do that depends on the actual cost of operating the vehicles and the service levels it can maintain - not just the sticker price of the car or the availability of an AI platform.
Autonomous vehicles may change how costs are allocated over time, but a robotaxi isn't automatically a cheaper taxi. The service still needs vehicles, technology, maintenance, operational support, and compliance with local rules. During supervised testing, safety operators are part of the plan. If the project later shifts to driverless rides, the operating model would change, but fleet management and reliable support would still be necessary.
A capital-light strategy has trade-offs
Misnik described inDrive's approach as "capital-light by design," working with partners in different markets. That can work when a company wants to explore a new capability without building everything inside. It can also make expansion easier if partners supply expertise that would take years to develop on its own.
The downside is dependency. When multiple companies contribute to one service, it only works smoothly with clear responsibilities and strong coordination. If vehicles, software, app features, or local operations are handled by different partners, changes in one area can affect the others. The announcement doesn't lay out the commercial terms or specify how responsibilities will be divided day to day.
For a company serving more than 1,200 cities across 47 countries, a partner-based approach could help in assessing autonomous mobility market by market. But running a single pilot and repeating it across dozens of locations are different challenges. Each deployment has to fit local regulation, customer expectations, weather, road conditions, and economics. One-off testing is a trial; scaling requires a system that can handle variation without turning every launch into a custom project.
Competition is already moving
This proposal also sits within broader momentum among ride-hailing companies moving into robotaxis. Uber has announced plans to test Pony.ai vehicles in London and already offers supervised autonomous rides there with UK startup Wayve. Uber is also working with robotaxi companies in other regions, including the Middle East.
Pony.ai is pursuing international expansion as well. The company has said it wants to operate more than 3,500 robotaxis in over 20 cities worldwide by the end of 2026, up from 1,975 vehicles at the end of June. Those are company targets rather than proof that the broader fleet expansion has already happened.
For inDrive, the strategic question isn't only whether it can match another platform's robotaxi headline. It's whether it can offer a service that fits its customer base and operating model. In a market where affordability matters, reliability and utilization will count as much as autonomy. If robotaxis only cover a small area or frequently pause service, the experience may still fall short of a compelling everyday option.
What to watch before the first rides
The gap between the announcement and the expected 2027 tests leaves several key questions unanswered. The answers will determine whether this turns into a useful operational pilot or stays mostly an outline.
Start with specifics about the pilot itself. The companies haven't announced fleet size, exact routes, passenger access details, or pricing. More clarity on these points would help distinguish a controlled technical test from something designed to gather real rider feedback.
Next, look at how the safety-operator phase will work. Having an operator on board is a reasonable starting point, but it will be more informative if the companies explain what the operator is responsible for and how interventions are logged. The goal should be to measure the vehicle's performance and limits, not to present supervised operation as a substitute for driverless capability.
Third, watch for evidence on winter operations. Astana's weather is part of why the city is attractive, but the partners will need to show how the system behaves across relevant conditions - and what restrictions apply. A successful demonstration on a small set of prepared routes would be a start, but it wouldn't automatically mean the system is ready for every road or every kind of winter weather.
Then there are regulatory milestones. Kazakhstan has set out a direction for autonomous taxi trials and created rules for testing on certain public roads. Details on local approvals, permitted routes, and safety conditions will define the real scope of inDrive's plan.
Finally, pay attention to the economics. No fares or cost targets have been announced yet. For inDrive, the key question is whether the service can eventually deliver dependable rides at a price that makes sense for its users while still covering vehicle, technology, operations, and support costs.
There's also a prior example worth tracking. inDrive signed an agreement with Saudi Arabia's AI Driver to explore autonomous vehicles powered by China's WeRide, starting in Riyadh. The companies targeted a pilot for the first half of 2026, but it's unclear whether testing has started. That project could offer a useful comparison for how inDrive handles partner-led deployments under different regulatory conditions.
Conclusion
inDrive's proposed Astana pilot is an important step in its plan for autonomous ride-hailing, but it's still a test proposal, not confirmation of a commercial, driverless service. The partnership lays out roles at the start: Pony.ai supplies the vehicles, Nvidia contributes technology, and inDrive connects the project to local operations and its app.
The real test will be whether the full system works reliably in Astana's conditions, meets Kazakhstan's requirements, and can eventually support a service that stays affordable. Until the companies publish details on routes, fleet size, passenger access, safety procedures, and pricing, the sensible way to read this is as an early-stage pilot plan.
As 2027 approaches, look for concrete operating details. Those will matter more than the robotaxi label alone.