Mobileye: The Great AV Impediment

Mobileye CEO Amnon Shashua has begun to sound more and more like Tesla CEO Elon Musk.? Both executives make bold performance claims for their autonomous driving products on behalf and in spite of persistent artificial intelligence ambiguity.? The key difference is Musk’s ability to point to actual “self-driving” performance improvements, where Shashua can only make claims and tout design wins with auto makers.

Listening to Musk on Tesla’s third quarter earnings call last fall one can hear the struggle the company faces in identifying and correcting weaknesses in its current full self-driving offering.

Quoting Musk from the Q3 transcript: “And when you start getting to where it can take 10,000 miles to find a mistake, it takes a while to actually figure out which it is. Is software A better than software B? It actually takes a while to figure it out because neither one of them makes the mistakes.”

Similarly, Shashua’s efforts to explain the progression of Mobileye's self-driving solutions in his annual speech at CES 2025 left analysts nonplussed resulting in a 25% stock plunge.? Shashua wasn’t wrestling with explaining a performance shortfall.? The deficit lay in announcing new OEM design wins.

The relative claims and performance of Tesla and Mobileye systems is significant with Mobileye positioned as the main supplier of driving assist systems to Tesla competitors.? Mobileye has accrued a substantial share of advanced driver assistance system market share thanks to its EyeQ chips and forward facing cameras.? Mobileye represents the tip of the spear for an automotive industry struggling to keep pace with Tesla’s self-driving efforts.

Notably, Mobileye parted company with Tesla back in 2016 after the fatal crash in Florida in which a Tesla drove under a trailer straddling a divided highway.? The separation set up a dynamic where Mobileye virtually leads AV development for the entire legacy auto industry against Tesla, with its superior vertical integration and video capture and processing.

The key problem for the rest of the auto industry, now almost exclusively dependent upon Mobileye, is Mobileye’s business model of controlling the capture and processing of data from forward facing cameras and the resistance of auto makers to sharing or transmitting that data.? In a post-Wejo world (Google it), auto makers are increasingly intent on owning the customer experience and managing any data collection or analysis.? (Regulators are more than casually interested in this data exchange as well.)

The main reason given for the post-CES stock plunge was the failure of Mobileye to announce any new OEM contract wins.? The impression created was of stagnation in spite of Shashua’s assertions of technical progress and the substantial long-term committed OEM relationships expected to contribute to billions in revenue for Mobileye – now approaching a $2B annualized run rate.

Mobileye has helped facilitate the development of hands-free driving systems from multiple auto makers including General Motors and Ford Motor Company.? The long-term plan for Mobileye is to transition these systems steadily toward eyes-free and, ultimately, driverless technology with a corresponding increase in revenue.

Since Mobileye has chosen to pursue its objective entirely as a supplier to auto makers, though, the company is unable to directly demonstrate the superiority of its technology to Tesla’s.? Instead, Shashua has gotten into an almost academic debate making a distinction between its own data fusion process, also adopted by Waymo, and called CAIS for Compound AI System.?

In contrast, Mobileye describes Tesla’s alternative approach as end-to-end: ?“an end-to-end neural network that processes visual data directly ‘from photons to driving control decisions,’ without intermediary steps or ‘glue code.’”? The key difference increasingly being Tesla’s ability to demonstrate and refine its solution in real world circumstances on a daily basis with results that are simultaneously impressive (increasing mean time/miles driven between failures) and troubling (occasional high profile collision events).?

The problem lies in the Mobileye business model.? Mobileye’s introduction of its Road Experience Management platform in 2017 – for data collection and sharing across OEMs – transformed Mobileye’s model from a solution to a service.? At the same time, it called for Mobileye clients to share the data collected by their forward facing cameras.

This business model change has brought Mobileye into the vehicle data sharing business which car makers themselves are desperate to exploit.? The REM platform gathers data including speed limit signs, traffic signals, roadway arrows, and lane markings to support automated and semi-automated driving.

REM does not capture or transmit driving video, which would be even more valuable for the development of self-driving technology.? This is a key advantage of Tesla’s vertically integrated solution which combines static data collection with video collection and data optimization software along with a Wi-Fi-based data sharing strategy.? The Tesla solution is designed with the clear purpose of steadily refining the existing full-self-driving solution for which Tesla owners pay extra at the point of vehicle sale and as a subscription during the life of the vehicle.

Auto makers competing with Tesla are desperately interested in both refining their own self-driving systems and extracting subscription revenue from their connected car systems.? Mobileye is essentially standing in the way, seeking to own the self-driving system as well as the process of refining the technology and extracting revenue.

Car makers are increasingly looking for ways to work around Mobileye.? The challenge derives from the lack of expertise within the typical OEM for building the machine learning necessary to support evolving self-driving systems in house.

In essence, dozens of car makers have handed the keys to self-driving development to Mobileye and Mobileye has hijacked that development locking OEMs out of the process.? Arguably, the Mobileye solution which may now rightly be described as a platform could serve as the ideal solution for cross-OEM data sharing in the interest of collective industry development of enhanced safety systems.

The problem lies in the reality that car makers compete on safety. ?They don’t cooperate or share. ?Safety is not a source of collaboration.? Virtually every safety system in the industry works a little differently with software and hardware variations though they are all focused on the same objective of preventing crashes and saving lives.

Shashua has devolved to being the standard bearer for an inferior self-driving platform dependent upon a data sharing approach that is at odds with auto industry data sharing practices.? To deliver on the promise of full self-driving, Mobileye is requiring ever-increasing investments and commitments from auto makers (along the evolutionary path from eyes-on ADAS to surround ADAS to SuperVision to Chauffeur to Drive/driverless) that ought to be investing in their own in-house systems in order to effectively compete with Tesla while defining and differentiating their own solutions.

There are a variety of paths forward for legacy auto makers.? There is the acquisition path evidenced by General Motors' acquisition of Cruise Automation, now being subsumed into GM’s existing self-driving development activities.? Another example is Stellantis’ acquisition of AIMotive.

There is the traditional Tier 1 swap.? Auto makers could simply replace Mobileye and its semiconductor-centric approach with platforms supplied by Nvidia or Qualcomm along with in-house development.

Or auto makers could stand up their own in-house self-driving teams (i.e. Woven by Toyota), likely already supporting semi-automated hands-free driving solutions to take on the more demanding task of honing full-self-driving systems with the help of partners.? One potential partner uniquely positioned to support this activity is Nexar which has both the Wi-Fi data sideload technology necessary for data extraction while also having mastered the collection of driving video from its hundreds of thousands of deployed dashcams.

Whichever approach auto makers adopt, the need for change has arrived.? The time to take charge of data collection and processing associated with refining self-driving technology is here.? Mobileye dependency is clearly a dead end.