Why This Obscure Company Snatched the Uber Deal from Momenta

Written by | Liu Ying

Edited by | Huang Dalu

Designed by | Zhen Youmei

On August 26, 2026, Munich is rapidly heating up as the new battlefield for global Robotaxi.

Waymo announced it would launch its first autonomous driving project in Europe here, planning to start commercial operations by the end of 2027 at the latest; while in June prior to this, Uber and Israeli autonomous driving company Autobrains had already made a preemptive layout.

With Baidu Apollo Go, Momenta, and VW MOIA following suit, multiple distinct technical routes are clashing head-on in this city.

On August 6, Igal Raichelgauz (Igal Raichelgauz), founder of Autobrains, publicly questioned mainstream approaches in the industry.

The mainstream view in the autonomous driving industry is that the future of autonomous driving requires more data. As long as enough video is collected, models are trained large enough, and enough vehicles are deployed, the system will eventually learn to drive. This route has shaped today's Waymo, Tesla, and Mobileye.

Autobrains, however, believes that driving AI is like the human brain. Humans do not learn to drive by having seen all scenarios, but because they understand road rules, spatial relationships, and behavioral logic.

This technology is called Thinking AI or Agentic AI by them.

PhD Dropout Founders "Brain Science AI Company"

To understand Autobrains, one must start with its founder Igal Raichelgauz.

He comes from Israel's elite intelligence units. After retiring, he founded a SMS-based instant messaging platform called Figment, then became CTO at speech recognition company LCB, and later held multiple technical positions at Intel.

Every segment was in a different industry, with different technical directions. The only common thread was his obsession with the question of "how machines understand information."

The turning point was at Technion.

In the early 2000s, Raichelgauz came to the Israel Institute of Technology to begin doctoral research on cortical neural networks. His collaborators were neuroscience professor Yehoshua Zeevi (Yehoshua Zeevi), and long-term collaborator Karina Odinaev (Karina Odinaev).

The three wanted to know how the mammalian cerebral cortex interprets meaningful visual worlds from the chaotic light signals received by the eyes?

The deeper the research went, the stranger it became. For a period, the lab was filled with live rats.

Later, the team really inserted electrodes into the rat's cerebral cortex, then converted digital images into electrical signals "written" into them to see how neurons would react. This was not computer simulation; it was real neural tissue responding to digital information.

Unfortunately, this research did not result in a doctoral thesis. Because in 2007, Raichelgauz abandoned his PhD studies and, along with Odinaev and Professor Zeevi, founded Cortica.

Cortica's initial ambition was to translate the way the human brain processes visual information into algorithms.

The logic of traditional machine vision is this. You show the system one million photos of cats, and it learns to recognize cats. You show it one million photos of stop signs, and it learns to recognize stop signs. Essentially, it is "seen enough," reproducing judgment in similar scenarios. Cortica's logic is different. It attempts to let the system extract structural patterns from visual information itself without relying on massive annotation.

Imagine a scenario. A stormy night, visibility less than 30 meters, an autonomous car driving on a suburban road. Suddenly, a construction worker wearing a reflective vest appears by the roadside, standing next to a temporary traffic cone, waving left to indicate detour. However, this combination, the company's system has never seen in training data.

For traditional systems relying on supervised learning, this scenario is a threat. Because it needs to be "seen" to be recognized, and edge scenarios in the open world are, by definition, impossible to collect completely. Pedestrians obscured by dappled tree shadows, animals suddenly crossing rural roads, bizarre roadblocks never seen before... Each requires dedicated collection, annotation, training, and verification.

Autobrains calls this dilemma the "Supervised Learning Wall." The company has long believed that methods relying on manual annotation and supervised training cannot effectively cover long-tail scenarios in open roads.

"After leading autonomous driving development projects for carmakers for many years, I clearly see that the industry's reliance on data annotation and supervised learning methods is insufficient now. It is too expensive, and once encountering scenarios not specifically trained, it will fail."

The words of Autobrains Senior Vice President Joachim Langenwalter (Joachim Langenwalter) serve as an intuitive description of this "wall." He previously served as Director of Automotive Software at Nvidia, Vice President of Software Strategy at VW Group, and Senior Vice President of Engineering at Stellantis.

Autobrains' founder also said in an interview later: "This process of information processing is not generating text or images, nor searching for images, but interacting in real-time with a large number of dynamic participants in extremely complex real-world environments."

He was talking about cars, but also seemed to be talking about the brain. He believes this is the hardest problem in AI. If this method works on open roads, it means it has solved one of the hardest perception problems in the AI field.

Incubated by Joint Automotive Industry Capital

At inception, the company focused on video content understanding, advertising technology, and security fields. In 2019, Cortica migrated to the automotive industry, giving birth to Cartica AI.

Cartica AI had obvious industry capital incubation colors from the start. The usual autonomous driving startup path is to first establish a company, develop technology, then go to OEMs, Tier 1s, and investors, round after round persuading others to believe in you.

Cartica AI's starting point was not like this. From the moment of its birth, it carried the color of joint industry incubation. OEMs, suppliers, and investment institutions sat at the table together, including Continental, Toyota AI Venture, BMW i Ventures, and Israeli crowdfunding investment platform OurCrowd.

The parent companies of these four institutions are in the upstream and downstream of the automotive industry chain. Continental is a global top-tier automotive parts giant. Toyota AI Ventures and BMW i Ventures represent industrial capital from OEMs in Japan and Germany. OurCrowd is the most active investment platform in Israel.

In other words, Cartica AI was not an isolated laboratory project from the first day of its birth. It was more like a "alternative technical route verification project" jointly supported within the automotive industry.

Later, Cartica AI evolved into Autobrains, with Raichelgauz serving as CEO. PitchBook data shows that Autobrains' predecessor names also included "Cortica Automotive," more clearly revealing its technical inheritance relationship with Cortica.

From Cortica to Cartica AI, then to Autobrains, every name change for this company was not a simple brand adjustment, but a process of technology vertically migrating to automotive scenarios.

Cortica solved "how machines understand the visual world," Cartica AI packaged this technology into solutions facing ADAS and autonomous driving. Autobrains then further locked the commercial target on car companies, Tier 1 suppliers, and mobility platforms.

In terms of specific products, this logic corresponds to the landing forms of different periods of Autobrains.

Liquid AI, released in 2024, breaks driving tasks into multiple independently optimized sub-questions, each handled by a dedicated neural network. Back to that stormy scenario at the beginning. The system does not use one large model to handle rain/fog interference, pedestrian recognition, and gesture understanding simultaneously. Instead, it flows information "like liquid" to the most suitable processing unit. Each unit is only responsible for the layer of judgment it is best at.

The Skills product line, released in October 2024, is the modular commercialization form of Liquid AI. It breaks autonomous driving into a series of "Skills." Each Skill is an AI model end-to-end optimized for specific driving scenarios, activated dynamically according to real-time driving situations. This contrasts with traditional large single neural networks, theoretically easier to extend and verify for new scenarios.

In 2026, Agentic AI entered the core narrative, further organizing driving intelligence into a group of mutually collaborating "scenario agents." Each agent is responsible for specific dimensions of perception, reasoning, and decision-making, and can respond in real-time to the complexity of the real world.

Air-to-Road positioning system, released in September 2024, fuses real-time visual data from vehicle-mounted cameras with satellite images, achieving centimeter-level accuracy positioning, completely without relying on GPS or HD maps. This directly solves the problems of high cost, slow updates, and limited coverage of HD maps, especially attractive for car companies needing to quickly enter new markets.

The company website states that all these technologies have four common points: low computing power requirements, hardware agnosticism (agnostic to sensors and SoCs), no need for large-scale manual annotation, and support for modular expansion.

But in reality, these claims still have a lot of parts that need verification in actual mass production. In public data, detailed independent third-party technical evaluations, large-scale real-road safety data, and standardized comparison tests with mainstream systems are currently lacking.

Its Star Rivals and Friends

Born in the same Israel, there was once a company that changed the global automotive industry landscape.

Mobileye was founded by Amnon Shashua (Amnon Shashua) and Ziv Aviram (Ziv Aviram) in Jerusalem in 1999. It took nearly twenty years to turn a camera-based driving assistance system into the world's most mainstream ADAS supplier.

In March 2017, Intel acquired it for about $15.3 billion. This was the largest acquisition case in the Israeli tech industry at that time. In 2022, Mobileye relisted on Nasdaq as an independent company.

EyeQ series chip shipments exceed 100 million units. Most mass-produced cars on the roads globally with ADAS functions, their eyes are Mobileye's.

The market often sees it as a reference frame that Autobrains cannot avoid.

TechCrunch used the expression "Mobileye challenger (take on Mobileye)" directly in the title when reporting on Autobrains' financing in 2022. Israeli tech media CTech has also placed it in a context of comparison with Mobileye multiple times.

Raichelgauz himself also admitted in an interview: "We are undoubtedly Mobileye's competitors. We are a software company and do not need to develop chips, which allows us to cooperate with different chip manufacturers."

But flipping over Autobrains' shareholder list, you will find its total financing amount is not conspicuous. Public data shows the company's total financing so far exceeds $140 million, holding more than 300 patents, and about 100 to 150 employees. Compared to Waymo and Cruise's financing of tens of billions, this scale is quite low-key.

But the composition of the shareholder list is almost hard to find a second similar company in the industry.

When established in 2019, the first batch of shareholders were famous industry investment institutions.

By Series C, the list expanded further. In November 2021, the company announced completing a $101 million Series C financing round, led by Singapore sovereign fund Temasek, with commercial vehicle brake system supplier Knorr-Bremse and Vietnamese EV manufacturer VinFast following suit. BMW i Ventures and Continental continued to add investments.

In March 2022, this financing round completed its final closing, total amount rising to $120 million. The same batch of industry capital was all present. In the list of investors later announced by Autobrains, global automotive parts supplier Magna and automotive electronics company Aogong also appeared.

Spreading out this list, the OEM dimension includes BMW, Toyota, and Vietnamese EV manufacturer VinFast. The supply chain dimension includes Continental, Magna, and commercial vehicle brake system supplier Knorr-Bremse. The long-term capital dimension includes Singapore sovereign fund Temasek.

This group of investors is itself an important potential customer for Autobrains.

How to Truly Enter Automotive Procurement Systems

One early morning in January 2023, a new onboarding document was received at the Autobrains office in Tel Aviv, Israel.

The person signing was Hilla Tavor (Hilla Tavor). For the previous twelve years, she worked at Mobileye in Jerusalem, serving as Vice President of ADAS Business Development, responsible for relationships with global OEMs. Among them, Ford, GM, VW, BMW, all were on her contact list.

Three weeks later, on February 8, another onboarding document also attracted attention. This time it was Joachim Langenwalter, serving as Senior Vice President of Autonomous Driving Business. According to Autobrains' official announcement, his previous job was Senior Vice President of Engineering at Stellantis, before that Vice President of Software Strategy at VW Group, and even earlier Director of Automotive Software at Nvidia.

Two months, two automotive veterans with extremely deep resume endorsements. This was not a coincidence, but a clear signal.

When Raichelgauz announced Langenwalter's addition, he spoke directly: "Hilla comes from Mobileye, responsible for leading ADAS. Now Joachim joins, responsible for leading autonomous driving."

This means for Autobrains, technology proof has reached a stage. Next, truly entering the automotive procurement system is the key node for enterprise development.

These are two different difficulty problems.

In the early technology stage, persuading a company that your algorithm is effective can rely on demos, releases, and media exposure; there is still a chance. But the automotive procurement system is not a place where one can enter with just a Demo. Procurement cycles range from three to five years. In between, there are functional safety certifications (ISO 26262), intended functionality safety (SOTIF), cybersecurity (UNECE WP.29), and other checkpoints. From prototype demo to mass-produced vehicle, an ADAS system involves a long engineering journey invisible to outsiders.

Mobileye took more than twenty years to build these relationships and trust. What Tavor and Langenwalter bring is precisely these industry accumulations that are difficult to build from zero.

Obtaining a Ticket to the Small Car Market

In all discussions about Autobrains, Robotaxi and L4 autonomous driving occupy most of the版面。But if you only look at this line, it is easy to miss the company's another more pragmatic route.

On December 4, 2024, Autobrains jointly announced the launch of a smart camera ADAS solution with JOYNEXT. JOYNEXT is headquartered in Dresden, Germany, a global Tier 1 supplier serving well-known automakers for over twenty years. It specializes in vehicle-mounted infotainment systems and smart connection gateways. Products have entered millions of vehicles globally.

The solution itself is not complex. It is by integrating Autobrains' AI perception software with JOYNEXT's camera hardware and control software into a compact "single box" module, compliant with European GSR general safety regulations and Euro NCAP requirements, and can be directly inserted into existing vehicle electronic and electrical architectures. After announcing the cooperation, both parties completed on-site demonstrations on test tracks in Germany.

JOYNEXT CEO Stavros Mitrakis (Stavros Mitrakis) said at the cooperation release: "Building a strong ecosystem is crucial for providing reliable and future-oriented technology for global customers. Autobrains is a suitable partner. Cooperating with them, we will be able to provide reliable, high-quality, and cost-effective autonomous driving technology in the future."

The target customers are global OEMs for small cars and compact cars.

Behind this positioning is a simple realistic logic, because the largest volume of the global automotive market has never been in luxury cars and high-end flagships, but in Class A, Class B, small SUVs and other mid-range and entry-level models. With the promotion of EU regulations (mandatory equipment of AEB in new cars) and improvement in consumer cognition, ADAS is becoming a standard from an option.

But cost is the threshold that this market cannot avoid. A complete perception solution requiring LiDAR, HD maps, and dedicated chips may increase the cost of a 100,000 yuan small car by 20,000 to 30,000 yuan. This is economically completely unfeasible. Autobrains' camera-first, low computing power, single box integration solution exactly hits the real demand here: low cost, easy integration, regulatory compliance.

JOYNEXT described this cooperation as an ADAS solution facing the mass production market, rather than a Robotaxi project. Compared to high-order autonomous driving, products facing small car and compact car markets are closer to the true sales baseline of the automotive industry.

Chosen by Vietnamese Automaker VinFast for Cooperation

North of Ho Chi Minh City, on a ordinary urban road in Hanoi, a white VinFast VF 8 is driving slowly. There is no LiDAR on the roof, and no dense sensor arrays can be seen front and back. Only a few inconspicuous positions on the body are equipped with cameras.

This car is testing Autobrains' L2++ system.

VinFast and Autobrains' relationship is another key to understanding this company's commercial logic. It is one of Autobrains' investors, and later became a partner.

On January 27, 2026, VinFast and Autobrains jointly announced strategic cooperation, jointly developing L2++ autonomous driving technology, and exploring a new autonomous driving architecture named "Robo-Car."

The technical claims of the Robo-Car solution are quite radical. Through 7 standard cameras plus one high-performance computing chip (processing about 20 trillion operations per second), without using LiDAR, radar arrays, or HD maps. Positioning relies on Autobrains' self-developed Air-to-Road technology, fusing real-time visual data with satellite images to achieve centimeter-level accuracy positioning under no HD map conditions.

Currently, the L2++ system has launched pilot tests in controlled areas in Hanoi on two main electric SUV models, VinFast VF 8 and VF 9, and plans to expand to larger cities and international markets.

VinFast Global Executive Vice President and Head of ADAS/AD Research Nguyen Van Duong (Nguyen Van Duong) said at the announcement of cooperation: "VinFast's strategic cooperation with Autobrains is helping to shape a disruptive, affordable autonomous driving solution, consistent with our long-term vision for global mobility. Our goal is to push advanced driving technology from early users to ordinary drivers."

Why VinFast? Reuters specifically emphasized in the report on both parties' cooperation that the biggest feature of this solution is not pursuing the highest level of autonomous driving, but achieving higher level intelligent driving capability at the lowest possible cost.

This Vietnamese car company's situation has a certain typicality. It is expanding globally, facing dual competitive pressure from Chinese EV brands like BYD and Xpeng, and traditional European and American car companies. Its early layout in the US market encountered sales declines and product quality problems, urgently needing to find a new differentiation fulcrum in intelligent driving.

But it cannot copy Waymo's high-cost route, nor does it have Tesla's vast data fleet of its own.

Autobrains just provided the answer it needed: low cost, camera-first, no LiDAR needed, not relying on HD maps. Some media called this solution "Low-Cost Tesla Route" because both emphasize camera-first and try to avoid expensive sensor stacking.

But in reality, there is an essential difference in the technical capability level between the two, cannot simply compare. Tesla's FSD capability comes from massive data collection fleet composed of millions of mass-produced cars, and multi-year continuous iteration end-to-end neural network system. Autobrains' cooperation with VinFast is currently still in the testing phase.

A more accurate interpretation is that Autobrains is most likely to find a market in emerging car companies that both need intelligent driving capability, are highly sensitive to cost, and are willing to try non-mainstream suppliers.

Uber's Horse Racing Strategy

On September 8, 2025, Uber and Momenta announced that Munich would become the first city for both parties' European L4 testing, originally planned to start testing in 2026.

However, on June 1, 2026, Uber announced it was ready to cooperate with Autobrains, putting the first stop in German Munich. This news once made outsiders suspect that Autobrains replaced Momenta. At that time, German media Electrive directly asked Uber, but received no reply.

On July 29, 2026, Momenta received full L4 testing license for Germany. The Munich project is still advancing. That is to say, Uber is betting on at least two sets of Robotaxi solutions simultaneously in the same city, leaving the choice of technical route to the market.

Momenta's path is: Uber + Momenta L4 technology + Mass production vehicle platform. Autobrains' path is: Uber + Autobrains Agentic AI + NVIDIA DRIVE Hyperion + Unannounced OEM.

On the night of March 18, 2018, in Tempe, Arizona, USA. An Uber autonomous testing vehicle was driving on a road with almost no streetlights. Ahead, a female pushed a bicycle across the road. The system failed to recognize, the brakes did not start. She died on the spot.

This was the first pedestrian death accident in the history of autonomous driving.

Two years later, Uber announced exiting self-developed autonomous driving. It sold its ATG department to Aurora. Uber also invested $400 million in Aurora and obtained about 26% equity, transforming into "an出行 platform for autonomous driving." It no longer makes its own guns, just doing the dock where shooters set off.

The logic is clear, not betting on a single technical route, but extensive layout, waiting for whoever runs the commercialization first, then deepening cooperation with them.

So far, Uber has woven a network of autonomous driving with more than 30 partners, preparing to invest more than $10 billion in the next few years, and plans to expand autonomous driving services to at most 15 cities by the end of 2026. US has Waymo, Nuro, Avride. Europe has Wayve (London), Momenta (Munich pilot), Middle East has Baidu Apollo Go (Dubai). Now Munich has added Autobrains.

Fast Company described Uber in special reports as one of the largest "platform bettors" in the autonomous driving industry. Compared to developing autonomous driving personally, Uber prefers to cooperate with multiple technical routes simultaneously, filtering the final winner through real operation scenarios.

This is a typical platform-side horse racing strategy. Uber does not need to judge who will win, it just needs to stand at the finish line waiting for the champion to appear.

Against this background, the significance of Autobrains entering Munich becomes clearer. It is not Uber's only European Robotaxi partner, but it is the first Israeli autonomous driving company to obtain a clear cooperation framework in the German market.

Munich's choice is very significant. This is the home of BMW headquarters, the spiritual center of the European automotive industry, and also the typical representative of strict regulations and complex road driving scenarios in European cities. Running Robotaxi here, the value in technical credibility and market demonstration effect far exceeds that of an ordinary small or medium-sized city.

Founder Raichelgauz said in the Munich cooperation announcement: "Cooperating with Uber and Nvidia, we are bringing our method into the autonomous driving mobility field, combining Agentic AI with the required mobility platform and automotive computing power, supporting Robotaxi scaled operation across cities, across vehicles, and across real road conditions."

Of course, the project is still waiting for German regulatory approval. The vehicle manufacturer has not been announced. The fleet size, service area, and commercial launch time have not been determined.

Even so, Autobrains is still worth being taken seriously.

Over the past ten years, the mainstream narrative of the autonomous driving industry has increasingly believed in "big," including large fleets, big data, large computing power, multi-sensors, HD maps, cloud training, and complex system engineering, etc.

This logic indeed pushed industry progress, producing heavyweight players like Waymo, Mobileye, Tesla, etc., but also brought high costs, long cycles, and commercialization pressure.

Until today, large-scale commercialization of fully unmanned driving is still on paper.

Against this background, the question Autobrains raised is a real industry pain point.

If autonomous driving can only be promoted by increasingly expensive systems, then a large number of cost-sensitive models, a large number of users in emerging markets, a large number of application scenarios that cannot afford LiDAR plus continuous cloud update costs, may not truly benefit from this technology.

Autobrains aims precisely to make autonomous driving AI smarter under cost constraints, become reliable enough, cheap enough, and easy to integrate enough.

This is a road, or a wall, needs time to judge.

References:

[1] electrive.com: Uber and Autobrains announce Robotaxi partnership in Munich (2026)

[2] Fast Company: Uber’s $10 Billion Bet: Why the Ride-Hailing Giant is Partnering with 30 Autonomous Driving Companies (June 2026)

[3] TechCrunch: Autobrains takes on Mobileye with $120M funding for its AI-based AV technology (March 2022)

[4] Reuters: Vietnam's VinFast partners with Israel's Autobrains on lower-cost self-driving tech (January 2026)

[5] CTech (Calcalist): From Mobileye to Autobrains: Hilla Tavor joins autonomous driving startup (January 2023)

[6] Automotive World: Autobrains launches Liquid AI to break the "supervised learning wall" (2024)

[7] JOYNEXT Official Newsroom: JOYNEXT and Autobrains announce strategic partnership to deliver cost-effective ADAS solutions (December 2024)

[8] VinFast Official Newsroom: VinFast and Autobrains Announce Strategic Partnership for L2++ and Robo-Car Development (January 2026)

[9] Autobrains Official Website: Technical Whitepaper and Funding Announcement (Agentic AI, Air-to-Road, Skills)