Mobileye files for IPO, reveals data

Mobileye files for IPO, reveals data

Business news |
By Nick Flaherty

Late on Friday Intel announced it is preparing to sell shares in its loss making Israeli driverless car technology subsidiary Mobileye.

The financial details of the offering for Mobileye were not disclosed, except that Intel will retain control of the venture through a majority of the Class B shares. It had been looking to value the venture, to be called Mobileye Global, at around $30bn, although the proposed share price was not included in the filing with the Securities Exchange Commission (SEC) in the US. The deal does include the Israeli Moovit real time public transport data app business that Intel acquired in 2020.

The company had revenue of $1.4bn in 2021, with a loss of $75m. This was down from a loss of $196m on revenue of $967m in 2020. Revenue for the first six months of 2022 was $854m with a loss of $67m.

The filing highlighted some of the key data for Mobileye for automated driving and driver assistance systems. The company is working with 50 car makers and its technology is included in 800 models.

“Our technology has been deployed in over 117m vehicles. By 2030 we expect Mobileye driver assistance systems to be deployed in another 266m vehicles globally, based on design wins though July 2 2022 alone,” said  Amnon Shashua, CEO of Mobileye

This has required a huge collection of data for the company’s machine learning AI frameworks and mapping software.

“We have assembled a substantial dataset of real-world driving experience, encompassing over 200 petabytes of data, which includes over 23 million clips collected over decades of driving on urban, highway, and arterial roads in over 80 countries,” said the company. “This data, plus proprietary search tools, enables us to develop and continuously improve our advanced computer vision algorithms to fit road scenarios and use cases that our system encounters.

This takes approximately 500,000 cloud CPU cores to process approximately 100 petabytes of data every month with 2D and 3D automatic-labelling methodologies that, together with a team of over 2,300 external specialized annotators, allow for fast development cycles for the computer vision engines. These advanced data labeling infrastructure and data mining tools can unlock significant data-driven insights, says the company..

It has created a separate dataset of 8.6 billion miles of roads driven by 1.5 million vehicles worldwide to create high definition maps. “We apply a series of on-cloud algorithms to build this crowd-sourced data into a high-definition, rapidly updating map that contains a rich variety of information, including road geometry, drivable paths, common speeds, right-of-way, and traffic light-to-lane associations. We estimate that the data we have accumulated covers over 90% and 80% of the approximately 0.8 million miles of motorway, trunk, and primary road types in each of the United States and Europe, respectively.

The company is also working on a software-defined imaging radar with a dynamic range and resolution backed by advanced processing algorithms to enable an independent sensing state. This would eliminate the need for multiple high-cost lidars around the vehicle and require only a single front-facing lidar, significantly reducing the cost of driverless cars.

Other articles on eeNews Europe


Linked Articles