深评：这一年Mobileye为何如此失意？
Mobileye, an Israeli chip company owned by Intel, has received many negative comments in 2021 due to the loss of some customers. NIO, Xiaopeng, Weimar, and Ideanext have announced that their next-generation flagship models will adopt NVIDIA DRIVE Orin chips, while "old partner" BMW stated that its next-generation ADAS and autonomous driving systems will use chips designed by Qualcomm.
Despite Intel CEO Pat Gelsinger's statement at the end of last year that Intel's acquisition of Mobileye has been hugely successful and that the EyeQ system-on-chip (SoC) has shipped more than 100 million units and achieved record revenue growth compared to the previous year, with expected revenue growth of over 40% in 2021 compared to 2020, the negative perspective from the outside still exists.
Therefore, in order to ensure the successful completion of its listing plan in the United States in 2022 and achieve the expected market effect, Mobileye needs to make bold moves. At CES recently held, Mobileye announced the launch of its most advanced and powerful system-on-chip EyeQ Ultra designed for autonomous driving, as well as two new EyeQ SoCs for advanced driver assistance systems (ADAS) - EyeQ6L and EyeQ6H.
The launch of Mobileye's new generation of autonomous driving chips can solve its current problems, but can it bring Mobileye back to its previous industry dominance?
Founded in 1999, Mobileye was once the darling of the times. At that time, Amnon Shashua, a professor at the Hebrew University of Israel, and entrepreneur Ziv Aviram, discovered the possibility of vision technology for car safety, and then founded Mobileye. It is committed to providing advanced driver assistance technology, including pedestrian detection, lane-keeping, and adaptive cruise control, using monocular vision.
From 1999 to 2001, Mobileye developed prototype products at a rate of once a year and proposed to solidify its self-developed algorithms into chips and integrate them into cars, which later became the EyeQ series chips that we know today.
In the fall of 2003, Mobileye released the EyeQ1 chip, which supported advanced features such as frontal collision warning, lane departure warning, and intelligent high beam control. At the same time, Mobileye also made great progress in capital. In 2003, Mobileye received a $15 million Series A financing round, with a post-investment valuation of $195 million; in 2007, Mobileye also received a $130 million financing round from Goldman Sachs, with a post-investment valuation of $600 million.
In 2007, Mobileye ushered in a historic change. BMW, General Motors, and Volvo installed Mobileye chips, and the latter officially got on the bus. The following year, EyeQ2 was released, and Mobileye began to enter a stable development period, with shipping volume continuously increasing.
With excellent performance data, Mobileye went public for the first time in the United States in 2014, raising a total of $890 million and achieving the highest record for an Israeli company's IPO in the US stock market, with a first-day increase of nearly 50%.
Later, Mobileye continued to shine with its stock price and shipping volume continuously increasing. It was due to these outstanding data performances that in March 2017, chip giant Intel chose to acquire Mobileye for $15.3 billion, one-third of which is a premium, which is also the largest acquisition ever made by an Israeli technology company.
Mobileye's positioning changed from "Our vision, your safety" to "Sensing the Future" after being acquired by Intel and becoming a system supplier from a single chip manufacturer. At that point, Mobileye still had few rivals due to its excellent solutions capability for hardware and software integration.
根据数据显示，2020年EyeQ系列芯片的出货量达到了1930万片，占据了全行业七成左右的份额，并且其营收规模达到了9.67亿美元，同比增长10%。Mobileye的收入情况也随之增长。
但是，进入2021年，随着Mobileye的增速持续放缓，其危机已经显现。其中一个原因是Mobileye算力的不足。自Mobileye推出第一款芯片EyeQ1以来，其已经推出了五代自动驾驶芯片，其中，EyeQ4芯片应用最广泛，应用于宝马、大众、蔚来、理想等车企的部分车型上。然而，随着自动驾驶技术的不断升级，EyeQ4所具备的算力已经不能满足更高阶自动驾驶需求。
尽管Mobileye于2020年发布了算力达到24TOPS的EyeQ5芯片，采用7nm的FinFET工艺，而该芯片于去年10月才正式搭载在极氪001车型上量产交付，不如英伟达Xavier芯片。而且，2022年，英伟达下一代自动驾驶芯片Orin芯片也将开始批量上车，单颗算力达到254TOPS，为EyeQ5算力的10倍多。
值得一提的是，国产芯片如地平线的征程5、华为的昇腾610都是与Mobileye EyeQ5的同期产品，各方面性能亦优于EyeQ5。因此，Mobileye在今年的CES Expo上宣布了三款产品，分别是EyeQUltra、EyeQ6H和EyeQ6L。
EyeQUltra是Mobileye主打产品，被称为最先进、性能最强的专为自动驾驶打造的系统集成芯片，其算力为176TOPS，且进行了能效的优化，预计将于2023年底供货，并于2025年全面实现车规级量产。眼下单颗就可以支持L4级别自动驾驶。
同时，EyeQ6H是EyeQ5H的进阶版本，算力为34TOPS，采用7纳米制程工艺，相较于EyeQ5多了接近3倍的算力，但是能耗仅多了25%，EyeQ6H将于今年第四季度交付工程样片，2024年正式量产。EyeQ6L则是低配版本的EyeQ6，同样采用7纳米制程工艺，其算力为5TOPS，功耗只有3瓦，半年前已经交付工程样片，预计2023年量产。
综上所述，算力虽然重要，Mobileye认为有效率、软硬结合同样重要。然而，Mobileye需要提高算力水平才能在竞争激烈的自动驾驶市场上立足。
地平线的算力虽然比不上英伟达的1000TOPS，但其专门设计的车规级芯片所达到的300TOPS效率同样不错。而一些车企更偏向英伟达，因为英伟达更加开放。
当前智能驾驶芯片主要分为两大阵营，一方面是以Mobileye为代表的黑盒阵营，提供从芯片到算法的解决方案；另一方面则是以英伟达为代表的开放生态阵营，允许自主开发上层算法。
虽然“黑盒阵营”可以快速为主机厂提供智能化的能力，但长远来看，算法决定了车企的核心竞争力。因此，在多家车企喜欢强调“全栈自研”的当下，算法显然也要掌握在自己手中。
主机厂布局自动驾驶通常分为三种模式：与Tier1和芯片商合作，做整合方案打包出售给主机厂；部分实力强的主机厂自行软硬件全栈自研，包括核心软硬件如芯片、操作系统和中间件；与Tier1分工合作，其中前者负责自动驾驶软件部分，后者负责硬件生产、中间层以及芯片方案整合。第三种模式或许能成为主要的合作模式，车企来控制能够带来用户体验差异化的部分，基础的软硬件开发工作则交给供应商来做。
Mobileye在最近的演讲中也表示，整个行业发展来说，黑盒子方案是有些过于极端了。从EyeQ5开始，Mobileye已经开始尝试与合作伙伴实现开源协作。
地平线的算力虽然比不上英伟达的1000TOPS，但其专门设计的车规级芯片所达到的300TOPS效率同样不错。而一些车企更偏向英伟达，因为英伟达更加开放。
当前智能驾驶芯片主要分为两大阵营，一方面是以Mobileye为代表的黑盒阵营，提供从芯片到算法的解决方案；另一方面则是以英伟达为代表的开放生态阵营，允许自主开发上层算法。
虽然“黑盒阵营”可以快速为主机厂提供智能化的能力，但长远来看，算法决定了车企的核心竞争力。因此，在多家车企喜欢强调“全栈自研”的当下，算法显然也要掌握在自己手中。
主机厂布局自动驾驶通常分为三种模式：与Tier1和芯片商合作，做整合方案打包出售给主机厂；部分实力强的主机厂自行软硬件全栈自研，包括核心软硬件如芯片、操作系统和中间件；与Tier1分工合作，其中前者负责自动驾驶软件部分，后者负责硬件生产、中间层以及芯片方案整合。第三种模式或许能成为主要的合作模式，车企来控制能够带来用户体验差异化的部分，基础的软硬件开发工作则交给供应商来做。
Mobileye在最近的演讲中也表示，整个行业发展来说，黑盒子方案是有些过于极端了。从EyeQ5开始，Mobileye已经开始尝试与合作伙伴实现开源协作。
理想汽车创始人李想近期发表微博，特别强调理想汽车在2020年上市后有资金能力进行智能驾驶全栈自研，所以停止了与Mobileye的合作。这也是很多车企选择英伟达、地平线等的原因。
Mobileye表示，基础ADAS是价格非常敏感的产品，客户不需要在EyeQ芯片上编程，比如EyeQ4芯片，只需要在挡风玻璃后面提供安全辅助功能即可。而当涉及到带有集中计算平台的多摄像头系统时，如EyeQ5设定的设计目标即是可编程的平台。
提高算力、开发平台，在丢失一些客户后，Mobileye痛定思痛，也在寻求转机。而车企能否再次给它机会呢，期待时间的答案。