NVIDIA’s Autonomy Strategy, Baidu goes driver-out in Dubai, Lucid/Uber/Nuro unveil Robotaxi
Meanwhile: BYD Launches New Robotaxi Brand and Xpeng starts Robotaxi Test
👋 Welcome back everyone,
This week, the autonomous vehicle industry converged on Las Vegas for CES 2026.
The show delivered a wave of announcements, partnerships, and strategic repositioning across the stack—from foundation models to manufacturing commitments to new market entries.
But CES wasn't the only story. Europe saw moves toward vertical integration and manufacturing sovereignty. China accelerated deployment timelines and launched new robotaxi programs.
We have a lot to discuss.
Let’s break it down.
⏱️ ~4000 words, 18 minute read
NVIDIA's Autonomy Strategy and the Manufacturing Race
NVIDIA CEO Jensen Huang stood on stage at CES and declared: "The ChatGPT moment for physical AI is here." He wasn't talking about chatbots. He was announcing Alpamayo, an open-source stack for autonomous driving spanning models, simulation, and datasets, designed to bring reasoning-based autonomy to vehicles.
The core release is Alpamayo-1, a 10B-parameter Vision-Language-Action (VLA) model designed to reason through complex and rare driving scenarios using chain-of-thought logic.
Jensen Huang put it during his CES keynote like this: “Not only does [Alpamayo] take sensor input and activate steering wheel, brakes, and acceleration, it also reasons about what action it’s about to take. It tells you what action it’s going to take, the reasons by which it came about that action. And then, of course, the trajectory.”
For nearly a decade, NVIDIA sold the infrastructure. The DRIVE platform. The chips. The simulation tools. The reference architectures. But the actual “brain”, the end-to-end model that decides when to brake or turn, that was left to OEMs, Tier-1 suppliers, or AV developers.
What Alpamayo actually is
Alpamayo 1 is a VLA model trained on 80,000 hours of multi-camera driving video across 25 countries and 2,500 cities. It’s designed to reason through rare and ambiguous scenarios rather than react purely on pattern recognition.
Instead of splitting perception, prediction, and planning into rigid modules, it leans into the same end-to-end direction that has gained traction across the industry.
Rather than running directly in-vehicle, Alpamayo models serve as large-scale teacher models that developers can fine-tune and distill into the backbones of their complete AV stacks.
As Grayson Brulte noted on the Road to Autonomy podcast NVIDIA has released open-source models for autonomous driving before. The difference? Licensing terms.
Previous models came with restrictive contracts. You could only use NVIDIA GPUs. You were subject to audits. Alpamayo uses Apache 2.0 licensing. Developers can use it commercially, modify it, and build on it without the same restrictions. The model has already been downloaded over 15,000 times on Hugging Face within days of release.
NVIDIA is betting that democratizing access to a high-quality foundation model will create an ecosystem effect. More developers build on Alpamayo. More OEMs adopt NVIDIA's stack. More demand for NVIDIA GPUs for training and inference. It's the Android playbook applied to autonomous vehicles.
NVIDIA wants to vertically integrate automotive, and the goal is straightforward: sell more GPUs. Automotive and robotics revenue hit $592 million last quarter, barely 1% of NVIDIA's total haul. That is a rounding error for NVIDIA.
Source: App Economy Insights
But CEO Jensen Huang has been unusually explicit about where he sees growth. Robotics, including self-driving, is now described as NVIDIA’s second most important growth category after AI infrastructure.
If a future world really includes hundreds of millions of autonomous vehicles, then selling a few thousand dollars of compute per vehicle quickly becomes meaningful.
Alpamayo is best understood as a vertical integration move in service of GPU demand.
Open models reduce friction. Reduced friction increases adoption. Adoption pulls hardware.
Even capturing a fraction of that market would transform NVIDIA's automotive business from a rounding error to a major revenue pillar.
OEMs Are Making Their Move
For years, OEMs faced an ugly choice.
Either:
Build autonomy in-house and accept multi-billion-dollar annual burn, or
Buy stacks and remain dependent on partners for core intelligence.
Recently, that balance has started to shift. And maybe NVIDIA's timing is no accident.
Some OEMs are bringing autonomy development in-house, and they need foundation models to accelerate progress without spending billions on R&D.
General Motors brought in Sterling Anderson to lead an in-house autonomy effort after shutting down Cruise. Ford announced at CES it will launch Level 3 autonomy in 2028, starting with its $30,000 all-electric pickup truck.
Alpamayo lowers the barrier to re-entry. It does not make autonomy easy. But it makes it organizationally survivable. Instead of starting from scratch, they can fine-tune NVIDIA's foundation model on proprietary fleet data, integrate it with DRIVE AGX Thor compute, and validate performance in simulation before deploying to production vehicles. It compresses development timelines.
This could put pressure on players like Mobileye, which has positioned itself as the pragmatic choice for OEMs wanting affordable Level 2+ systems. But if NVIDIA offers a customizable foundation model for free, Mobileye's value proposition weakens.
Mercedes-Benz as the first real test
But does the model actually work? Mercedes-Benz will provide the first real-world answer. The automaker's next-generation CLA, launching in late 2026, will use NVIDIA's technology for point-to-point navigation in U.S. cities.
Early test drives in San Francisco suggest the system is competitive with Tesla's Full Self-Driving. One reviewer described it as feeling "like the car is on rails" during a 40-minute drive through heavy traffic, handling four-way stops, double-parked cars, and unprotected left turns without intervention.
But to be clear: like FSD, this is also still L2+ and requires driver supervision
The Tesla Dynamic
Elon Musk noticed. Hours after Huang’s keynote, the Tesla CEO posted on X in response to a transcript of the announcement: “Well that’s just exactly what Tesla is doing 🤷♂️.”
What followed was a rare exchange between two tech industry heavyweights that went viral for its civility rather than conflict. Both acknowledged the other’s technical achievements while staking out different strategic territories.
The relationship between the two companies is layered. Tesla depends on NVIDIA’s data center GPUs for training its autonomous driving models, with spending projected to reach $10 billion cumulatively by year-end. Yet Tesla designs its own inference chips for in-vehicle compute. Meanwhile, Musk’s separate AI venture xAI buys heavily from NVIDIA, which has also invested in the startup.
Their technical philosophies diverge sharply. Tesla has committed to camera-based perception, betting that vision-only systems offer the best path to economic scalability. NVIDIA’s platform incorporates lidar, radar, and ultrasonic sensors, prioritizing redundancy and safety margins. Tesla vertically integrates everything and manufactures vehicles. NVIDIA provides enabling technology to the industry.
Yet their consumer strategies are converging. Tesla markets Full Self-Driving (Supervised), an advanced assistance system that handles point-to-point navigation but requires continuous driver attention. NVIDIA now offers automakers similar capabilities through platforms like the one powering the upcoming Mercedes CLA.
Both see robotaxis as the ultimate prize. For Tesla, today’s FSD becomes the foundation for a captive fleet it would own and operate. NVIDIA envisions the same outcome through partnerships with companies like Uber, Lucid, and Nuro, targeting commercial robotaxi deployments by 2027.
During a Bloomberg interview at CES, Huang was candid about where things stand today. He called Tesla’s autonomous stack “the most advanced AV stack in the world.” But he also positioned NVIDIA’s open ecosystem as a way to accelerate progress industry-wide, not just for a single manufacturer.
Musk, characteristically, pushed back on the idea of near-term competitive pressure. He wrote on X that bridging the gap between systems that “sort of work” and systems dramatically safer than human drivers takes years. Meaningful rivalry, he suggested, remains half a decade out.
Interest beyond Mercedes
NVIDIA has been careful with its wording. Only Mercedes is confirmed as a production program.
That said, Lucid Motors, Jaguar Land Rover, Uber, and research groups like Berkeley DeepDrive have all publicly expressed interest.
That brings us to the second big CES story.
The Manufacturing Race No One Expected
Waymo unveiled its rebranded Zeekr robotaxi, now called Ojai (pronounced "oh-hi"), at CES. The vehicle has been in development for three years and is finally ready for commercial deployment. It features 13 cameras, four lidar units, six radar sensors, and NVIDIA DRIVE compute.
Waymo currently operates several thousand vehicles across five U.S. markets. Scaling to tens of thousands requires manufacturing capacity that Waymo doesn't control. The Ojai is built by Geely's Zeekr in China, which introduces geopolitical risk and supply chain complexity. Waymo needs a production partner that can deliver at scale, and it's unclear whether Zeekr is that partner.
Waymo also had its Hyundai partnership represented at CES. The IONIQ 5 is still part of Waymo’s next wave, and Hyundai vehicles have started gradually testing with human drivers, but there’s no public launch day yet.
Source: Autonomy central
Meanwhile, Lucid, Uber and Nuro unveiled a production-intent robotaxi at CES that stole the show. Based on the Lucid Gravity SUV, the vehicle integrates cameras, solid-state lidar, radar, and NVIDIA's DRIVE AGX Thor compute directly on the assembly line at Lucid's Arizona factory. This is a critical advantage. Waymo currently has to retrofit the vehicles with autonomous hardware, a time-consuming and expensive process. Lucid builds the autonomy package into the vehicle from day one.
The reaction from people who saw it in person was consistent. High build quality. Clean integration. Purpose-built without looking experimental.
The vehicle looks impressive. Spacious interior. High-quality materials. Integrated LED halo for rider identification. A rider interface that mirrors Waymo's isometric graphical display.
Source: Sean O'Kane
Lucid doubled production in 2024 and hit new sales records despite earlier software struggles. Once final validation is complete on the robotaxi later this year, true production versions will start rolling off Lucid’s factory lines. If Lucid can maintain quality and ramp production, the Uber/Lucid/Nuro partnership could reach commercial deployment in San Francisco by late 2026, with production vehicles rolling off the line shortly after.
That would put Lucid/Nuro in direct competition with Waymo for manufacturing capacity.
Of course, when discussing manufacturing at scale, Tesla holds the clearest advantage. The company could produce fleet vehicles in significant volume without the partnerships, retrofitting, or supply chain complexity that others face. But Tesla first needs to crack FSD unsupervised. Until then, manufacturing capacity sits idle while the software catches up.
The industry is no longer asking whether robotaxis work. The question is: who can manufacture them at scale, and how fast? It seems Lucid could be in a good position.
What This Means for the Competitive Landscape
NVIDIA’s move creates fragmentation, which is exactly what Uber wants. The ride-hailing giant needs multiple autonomous vehicle providers to avoid dependence on any single supplier. NVIDIA’s open-source approach enables more players to enter the market, whether that’s OEMs like Mercedes and Ford or new entrants using Alpamayo as a foundation.
But fragmentation also creates challenges. If ten different OEMs ship ten different autonomous systems, each with different capabilities and safety records, consumer trust becomes harder to build. Regulatory approval may become more complex. The industry needs both competition and standardization, a difficult balance to strike.
The robotaxi manufacturing race introduces another variable. Waymo has a multi-year head start in operations and safety validation, but Lucid/Nuro could leapfrog them on production capacity. Waymo’s partnership with Uber appears strained, though the two companies remain in multiple markets together. If Uber shifts volume to Lucid/Nuro vehicles, Waymo’s path to scale becomes more difficult.
The Road Ahead
NVIDIA’s Alpamayo won’t be the “Android of self-driving” overnight. The model needs to prove itself in production vehicles, starting with Mercedes in 2026. OEMs need to demonstrate they can build safe, reliable autonomous systems without spending a decade on development. And the robotaxi manufacturing race will determine which companies can scale from thousands to tens of thousands of vehicles.
The winners will be determined by execution, not announcements. Manufacturing capacity, safety validation, regulatory approval, and operational efficiency will separate the leaders from the laggards.
2026 will provide the first real answers.
🔗 NVIDIA / NVIDIA (2) / CNBC / / Reuters / Robonomics / Autonomy Central / TechCrunch / TechCrunch (2) / The Verge / Road to Autonomy / Constellation Research / Bloomberg / Bloomberg (2)
Enjoying the AV Market Strategist?
If you find value in these deep-dives and enjoy the updates on autonomy each week — you can support my work.
Your contribution helps me access better data sources, industry reports, and research tools to keep every issue sharper and more insightful.
The newsletter will always remain free to read, whether you choose to support or not — but every bit of backing helps me make it even better.
Europe and China Aren't Waiting
Europe Builds Its Own Full-Stack Provider
While NVIDIA dominated headlines in Las Vegas, a quieter but strategically significant deal closed in Europe. BENTELER Group acquired ioki, Deutsche Bahn’s mobility software subsidiary, creating what the companies are calling “Europe’s first full-service autonomous mobility provider.”
The transaction brings together three European companies under one roof: HOLON (autonomous shuttle manufacturer), ioki (dispatch and on-demand platform), and Benteler Mobility (fleet operations and financing). It’s the European version of the Zoox playbook: own the hardware, software, and operations to reduce dependency and capture more margin.
Why does this matter? Because it puts a European player in position to compete with Chinese imports and American technology providers without relying on either. BENTELER now controls the entire value chain from vehicle production to fleet management, targeting cities and transit agencies that want turnkey autonomous mobility solutions.
ioki brings immediate market access. Founded in 2018, the company has deployed software in over 200 transport services across Europe, reaching nearly 10 million passengers. Its data-driven route planning and simulation tools are already integrated with public transport systems in Germany, Austria, and Switzerland, the most regulation-ready autonomous public transport markets in Europe.
The strategic logic is again vertical integration. Cities don’t want to piece together vehicles from one supplier, software from another, and operations from a third party. BENTELER is betting that a fully integrated European solution will win contracts that fragmented competitors can’t touch.
This also sets up a collision course with Volkswagen’s MOIA and the ID.Buzz autonomous shuttle program.
BENTELER is also building a production facility in the U.S., which means the company can offer its integrated stack to American cities without depending on U.S.-based software providers. It's a deliberate move to avoid the regulatory and supply chain vulnerabilities that come with Chinese manufacturing or reliance on American tech giants.
Applied Intuition Brings Silicon Valley to Stuttgart
A very different European signal came from Applied Intuition.
Applied Intuition used CES to underline its expansion in Europe, with a particular focus on Germany and Stuttgart. The headline was the European rollout of its Self-Driving System (SDS), working directly with European automakers on Level 2+ ADAS and automated driving features for production vehicles.
Applied Intuition is explicitly positioning itself against black-box autonomy solutions. Its SDS is marketed as a white-box, end-to-end autonomy stack that OEMs can inspect, customize, and integrate into their own software organizations. That framing resonates strongly in Europe, where OEMs remain deeply uncomfortable with ceding control over core vehicle intelligence.
The Stuttgart expansion is therefore not just about local hiring or road testing. It is about embedding autonomy development into European OEM workflows, validation processes, and safety cultures. Applied Intuition is effectively saying: autonomy does not need to be outsourced wholesale to Silicon Valley or China. It can be built collaboratively, on European roads, under European standards.
In a market where regulators, works councils, and internal governance matter as much as raw performance, that positioning is powerful. It also explains why Applied Intuition increasingly shows up not as a “tool vendor,” but as a quasi-co-developer for OEM autonomy programs.
China Accelerates: Baidu, BYD, and Xpeng
Baidu’s Apollo Go received Dubai’s first fully driverless testing permit from the Roads and Transport Authority, allowing the company to operate autonomous vehicles on public roads without safety drivers. The permit marks a critical step toward commercial operations in Q1 2026, with plans to scale to over 1,000 driverless vehicles.
Baidu opened its first facility outside China, Apollo Go Park, a 2,000-square-meter operations hub in downtown Dubai integrating charging, maintenance, and command infrastructure. The company has already logged over 240 million autonomous kilometers globally, with 140 million in fully driverless mode. Weekly rides exceed 250,000 across 22 cities.
The RTA is actively partnering with Baidu to test safety, reliability, and customer experience at scale, with the explicit goal of integrating autonomous vehicles into the city’s public transport network.
And Baidu is not alone.
BYD used CES-adjacent events to underline just how aggressively it is pushing intelligence into the mass market. By assembling a 5,000-person assisted-driving team and committing more than 100 billion yuan to intelligence, BYD is turning ADAS into a volume game. Its strategy is not to jump straight to robotaxis, but to flood the market with advanced driver assistance across even its lowest-priced vehicles, generating enormous data volumes in the process.
That data flywheel is the real asset. Millions of vehicles, generating hundreds of millions of kilometers per day, feeding back into increasingly capable models. BYD is building the largest vehicle-cloud database in China, almost as a byproduct of selling cars.
The goal isn’t to replace the driver but to create “the strongest guardian of safe travel.” It’s a pragmatic strategy that sidesteps regulatory uncertainty while building the data foundation for future Level 4 systems.
But that they also harbor Level 4 ambitions became clear through the next news. BYD filed four new models under a brand-new sub-brand called Linghui, specifically targeting China’s ride-hailing and robotaxi market. The models—Linghui e5, e7, e9, and M9—are rebadged versions of existing BYD vehicles (Qin Plus EV, Sealion 06 EV, Han sedan, and Xia MPV) adapted for autonomous operations.
This is brand architecture strategy. BYD’s growing association with affordable ride-hailing could hurt its luxury positioning through Yangwang and Denza divisions. Linghui solves that problem by creating a B2B-only marque that separates fleet operations from consumer brands. It’s low-cost differentiation: rebadge existing platforms, avoid cannibalization.
Whether Linghui vehicles use the same God’s Eye sensor suite or have upgraded hardware for robotaxi durability remains unclear. But the move signals BYD expects robotaxi sales to become a significant revenue stream. The company has invested in LiDAR supplier Robosense while developing proprietary autonomous solutions in-house.
BYD joins an increasingly crowded field of Chinese OEMs staking claims in robotaxis
Xpeng is pursuing full autonomy. The company announced it will begin public road testing of robotaxis equipped with VLA 2.0 software, a vision-language-action model enabling entry-level Level 4 capabilities. Xpeng plans to launch three robotaxi models in 2026, each powered by four in-house Turing AI chips delivering 3,000 TOPS of compute.
The vehicles require no modifications, don’t rely on high-definition maps, and are designed for low-cost mass production. VLA 2.0 will also roll out to Xpeng’s consumer vehicles starting in March, claiming to deliver near-Level 4 capabilities. CEO He Xiaopeng declared that 2026 “will mark the true dawn of fully autonomous driving in both China and the United States.”
And the latest news from China comes from Pony.ai. The company deepened its strategic partnership with BAIC BJEV, one of China’s leading EV manufacturers, to accelerate robotaxi mass production and global deployment.
Under the new agreement, Pony.ai and BAIC BJEV will collaborate more deeply on the forward design and development of purpose-built robotaxi models, jointly optimizing vehicle architectures and in-cabin systems to better support autonomous operations and passenger experience. This collaboration establishes a scalable foundation for large-scale robotaxi deployment, while also exploring potential applications of Pony.ai’s autonomous driving technology in certain passenger vehicle programs over the longer term.
In addition to vehicle development, the two companies will also strengthen collaboration across the autonomous mobility value chain, including user acquisition, fleet operations, and vehicle maintenance. Leveraging BAIC BJEV’s strengths as one of China’s leading EV manufacturers and its mature OEM-grade supply chain, the partnership is expected to further reduce the bill of materials (BOM) and long-term operating costs of autonomous vehicles. At the same time, the collaboration will improve vehicle performance, maintenance efficiency, and full lifecycle management—key enablers for sustainable robotaxi commercialization at scale.
The partnership is already producing results. The Arcfox Alpha T5 Robotaxi, their first joint model, debuted in April 2025. Over 600 units have since rolled off production lines and are now operating commercially in Beijing and Shenzhen.
Taken together, this week was clearly dominated by news out of the United States. NVIDIA’s moves and CES shaped much of the industry conversation.
But beneath that headline narrative, Europe and China continued to push forward in their own, distinct ways.
Autonomy is advancing unevenly, region by region, stack by stack, use case by use case.
The pace is picking up.
And I’m looking forward to tracking, analyzing, and commenting on these developments for you, week by week.
🔗 Ioki / CNEVPost / PR Newswire / PR Newswire (2) / Barak Sas / Ralf Göttel / Applied Intuition / Gasgoo / Automotive World
💡 Quick Takes
RoboSense and CIDI Partner on Autonomous Mining and Logistics
Chinese lidar maker RoboSense and autonomous driving developer CIDI announced a strategic partnership spanning unmanned mining trucks, driverless logistics vehicles, and smart city infrastructure. RoboSense supplies products to over 310 automakers and 3,400 industrial customers globally. The partnership focuses on closed-environment applications where safety and efficiency demands are highest, including mines, ports, and industrial parks.
🔗 Gasgoo
Caterpillar and NVIDIA Collaborate on AI-Powered Heavy Industry
Caterpillar is deploying NVIDIA Jetson Thor for real-time AI inference on construction and mining equipment, enabling autonomous operations and intelligent in-cab experiences. The company debuted its Cat AI Assistant, built on NVIDIA Riva speech models, which provides voice-activated support for equipment settings, troubleshooting, and maintenance. Caterpillar is also using NVIDIA Omniverse to build digital twins of its factories for production optimization.
Waymo Rebrands Zeekr Robotaxi as "Ojai"
Waymo's Chinese-made Zeekr RT robotaxi is now called Ojai, named after a California arts community. The rebrand distances the vehicle from its Chinese manufacturing origins ahead of commercial deployment. The Ojai features 13 cameras, four lidar units, and six radar sensors, with a steering wheel included in the production version.
Amazon and Aumovio Partner on Autonomous Truck Deployment
Aumovio is integrating AWS generative AI tools into its development workflow for Aurora's autonomous truck platform. The partnership aims to streamline testing and validation by allowing engineers to query millions of driving scenarios using natural language. Aumovio is co-developing the Aurora Driver with large-scale production set for 2027, including a safety-redundant backup computer.
🔗 Transport Topics
Kodiak AI Partners with Bosch to Scale Autonomous Truck Platform
Kodiak AI announced a partnership with Bosch, the world's largest automotive supplier, to develop a production-grade safety-redundant autonomous platform for commercial trucks. The deal aims to integrate specialized hardware, firmware, and software interfaces that can be installed either on factory assembly lines or by upfitters.
Mobileye Acquires Mentee Robotics for $900M
Mobileye announced it will acquire humanoid robotics startup Mentee Robotics for $900 million in cash and stock. Mobileye co-founder Amnon Shashua also co-founded Mentee in 2022. The deal signals Mobileye's expansion into "physical AI" beyond automotive applications. Mentee will operate as an independent unit within Mobileye, leveraging the company's AI training infrastructure and automotive revenue pipeline of $24.5 billion over eight years.
📚 Worth Reading/Listening
Alex Immerman: Self-driving cars: social robots that save lives
🔗 a16z.news
📊 Weekly Performance
Note: Stock performance data as of January 11, 2026. Past performance does not indicate future returns
Thanks for reading!
If you found value in this newsletter, please consider sharing it with a friend or colleague who might benefit from these insights.
And if you haven’t already, subscribe to stay updated on the latest developments in the autonomous driving industry.
Disclaimer: This newsletter is for informational purposes only and should not be considered investment advice. All opinions are my own.
Great write-up and summary as always! Keep it up. 🤗
Ecosystem plays matter when the market is this early.