Baidu Reached Kazakhstan First — Pony.ai's Astana MOU Buys a Regulatory Template, Not Robotaxi Revenue

The order of events is the whole point. In late January 2026, Kazakhstan's Ministry of Artificial Intelligence and Digital Development, the Ministry of Transport and the Astana city government signed memoranda with inDrive and Yandex Qazaqstan for a three-year autonomous taxi pilot, with road markings, signage and pick-up zones scheduled for the first quarter. By early August, the ministries had approved the pilot's operating rules, and by late September the deputy prime minister said robotaxis would start in Kazakhstan in 2027. The Pony.ai memorandum arrived after the regulatory scaffolding was already standing.

In its second-quarter results filed on Aug. 18, 2026, Baidu disclosed a memorandum of understanding with Kazakhstan's Turlov Private Holding to jointly explore autonomous ride-hailing. Apollo Go ran roughly one million fully driverless rides in the quarter, had passed23 million cumulative public rides by June, and has accumulated more than 350 million autonomous kilometres across 28 cities, including fully driverless commercial operations in Dubai available through the Uber app and Hong Kong's first fully driverless permits. By the time the Pony.ai memorandum was announced, Baidu had beat Pony.ai to the Kazakh starting line by five weeks and held one to two orders of magnitude more real-world operating data.

Read the announcement for what is missing. Neither the joint statement nor the Pony.ai investor-relations release names a vehicle count, a contract value, a duration, a deployment schedule or even a binding launch date for Kazakhstan. The three-way split is: Pony.ai supplies Gen-7 vehicles, inDrive supplies the platform and runs the service, and NVIDIA supplies the AI infrastructure, with Hyperion as the Level 4 reference sensor and compute platform combined with the Alpamayo stack and Halos safety system. The launch \"would follow agreement with the government of Kazakhstan.\"

Compare that with how Pony.ai frames its European commitments: more than 2,000 robotaxis contracted with Uber across five European cities, and over 4,000 vehicles under agreement or negotiation across international markets. Kazakhstan is the one market where the company has a signed partner, a named city and a reference compute stack — and tells investors the pilot is a pilot. That is not necessarily bad; it is a deliberately cheap option on a regulatory outcome.

The structure is worth stating plainly because it decides who collects what. Pony.ai sells vehicles and software, not rides; inDrive owns the rider relationship and the operating licence; NVIDIA owns the compute and, more importantly, the safety doctrine that a regulator is likely to reference. So a Kazakh pilotharms no listed bottom line and creates no near-term revenue — it is a reference implementation whose value accrues to whoever owns the standard.

Two upstream facts sharpen the point. Gen-7 is described by Pony.ai as the first system built on 100% automotive-grade components, with system cost down 70% against the previous generation, lidar cost down 68% and the vehicle computing unit down 80%, and the company has said it wants per-vehicle breakeven. That cost curve is what makes a fleet of tens of vehicles in Astana plausible, and it is also why the chassis suppliers — BAIC BluePark, GAC Group and Toyota — are the ones absorbing the pressure of supplying platforms at scale for a customer with pricing power.

Robotaxi revenue has grown to a third of group revenue, but the denominator is what constrains the Kazakhstan story. First-half2026 revenue was US$70.5M against an operating loss of US$124.1M, with research and development of US$104.1M, and the company held US$1,390.5M of cash, short-term investments and wealth-management instruments at June 30. Divide the Q2 robotaxi figure by the 1,975-vehicle fleet and revenue works out near US$6,100 per robotaxi per quarter — a rough ratio, since the fleet grew through the period and segment revenue includes more than fares, but the right order of magnitude.

Set the same quarter against the two comparators and the scale gap is stark. NVIDIA reported US$96.2B of revenue for the quarter ended July 26, 2026, with Data Center at US$89.0B and the combined Edge Computing line that now carries automotive, gaming and robotics at US$7.2B — so the Kazakh pilot cannot move NVIDIA's numbers by any visible amount. Uber reported US$14.19B of revenue and US$2.39B of net income for the June quarter, and it collects riders whether the vehicle behind them is Baidu, Pony.ai or WeRide. The Kazakh pilot barely dents any of these profit and loss statements.

Over a one-to-three-year view the question is whether the doctrine travels. If Astana shows that a Chinese stack can run driverless under remote supervision through a Kazakh winter while a Kazakh regulator signs off, inDrive can point at a legal precedent across 1,200 cities in 47 countries — and the same regulatory template becomes a substitute for a home-market champion in places where inDrive already has drivers and users. That is the strategic content of the announcement, and it is also why the Kazakh pilot matters to NVIDIA: a safety stack that a regulator has already blessed in one jurisdiction is cheaper to certify in the next.

Separating the three layers of confidence. Facts: Pony.ai, inDrive and NVIDIA signed a memorandum on Oct. 23, 2026 to explore Gen-7 robotaxis in Astana; Kazakhstan approved autonomous taxi pilot rules in August, including driverless operation with remote supervision; Baidu disclosed its own Kazakhstan memorandum in August and ran about a million fully driverless rides in the June quarter. Inference: the value of Astana is precedent and certification rather than rides, because a 1.7 million-person capital cannot support fleet economics. Speculation: that the precedent migrates into dozens of other lightly governed markets, which nobody can yet observe.

The uncomfortable implication is that this is not a Western-versus-China story. Nothing in the record shows the United States or Europe denying anything: Uber is already running Pony.ai Gen-7 vehicles in Zagreb, plans London road tests, and offers Baidu's fully driverless Dubai fleet in its own app. Kazakhstan is parallel to the UK's 2027 implementation timetable, not ahead of it. The arbitrage is cheaper permission, not earlier permission, and it is captured by the platform and the standard-setter rather than by whoever flags the first ride.

So Pony.ai trades on optionality in a market it has not entered, and its third-quarter print on Nov. 18 is the first place that optionality could become a number. If the company discloses a Kazakhstan fleet, a contract value or a named local partner, the template thesis gains a hard data point; if the only disclosure is a repeat of the memorandum, then the honest conclusion is that a licensing framework was built for an operator that has not yet committed to it — useful for the ecosystem, and not yet an earnings event for anyone.