news

our reporter yin limei and zhang shuo reported from hefei

the technological high ground of the automobile industry and the strategic competitive fulcrum of automobile companies are rapidly migrating to intelligence driven by ai (artificial intelligence). the competitiveness of the past can hardly support the next development of car companies. future development must be based on intelligent development driven by ai, and strategic migration can be achieved through ai. only in this way can car companies win the future. on the contrary, if they fail to keep up with the changes in the field of intelligence, or do not pay enough attention, then many companies will lose their future.

on september 29, at the global intelligent vehicle industry conference (2024giv) held by the china electric vehicles association of 100, zhang yongwei, vice chairman and secretary-general of the china electric vehicles association of 100, made the above statement. moreover, he pointed out that the development speed of automobile intelligence is too fast, exceeding the industry’s predictions.

intelligent transformation requires huge r&d investment. a reporter from china business news noted that according to official standards, tesla’s cumulative r&d investment in fsd (full-self driving, fully autonomous driving) will exceed us$10 billion in 2024. as of the end of 2023, huawei auto bu (huawei smart auto solutions business unit) has invested 30 billion yuan in intelligent software and hardware. byd also proposed to invest 100 billion yuan in the field of intelligence.

zhang yongwei believes that at this stage, in terms of intelligence, companies are not only competing for knowledge and speed, but also for strength. the threshold for intelligent development has become higher and higher.

ai has become a decisive factor in driving changes in automobiles

"the change cycle of the automobile industry is getting shorter and shorter, and it even has the characteristics of superposition of changes: the past changes have not yet been completed, and new changes are about to begin. this superimposed development has become the new normal of the development of the automobile industry. the century-old automobile industry continues to "the latest driving factor in evolution is artificial intelligence." zhang yongwei said that artificial intelligence, represented by big computing power, big data, and big models, has begun to integrate with automobiles. the biggest change brought to the automobile industry in the artificial intelligence era is that ai has become the driver. the new decisive factor in the automotive revolution.

in this situation, how china's automobile industry adapts to new changes and makes new adjustments has become a must-answer question. zhang yongwei’s solution to the problem is that the development of china’s intelligent connected cars needs to focus on ai technology and data value to shape new competitiveness in the industry. that is to say, on the one hand, we should give full play to the value of data as a production factor and build data competitiveness; on the other hand, we should make good use of ai model capabilities to promote the level of automobile intelligence.

data is a key element in the iteration of intelligent driving technology. our country does not have data advantages in terms of intelligence, and most companies have very limited computing resources.

taking self-driving clips (valid video clips) data as a comparison, domestic companies have less than one million valid video clips, while tesla already has more than 10 million valid video clips. tesla’s computing power level has reached 100 eflops, which is the sum of the computing power possessed by all car companies. huawei's current computing power level is only 7.5 eflops, which is significantly behind tesla.

the reporter learned that most of the data of domestic car companies is distributed in several dense scenes, and the "head effect" is obvious. at the same time, different vehicle models have different sensor configurations, resulting in differences in collected data and poor data reusability. in addition, car companies have insufficient capabilities in data sorting and value mining.

zhang yongwei believes that domestic enterprises must solve two core problems from the perspective of data: first, let data become the core assets and elements of the enterprise, let data create value, and change the lack of data mining capabilities of automobile companies and their lack of data. to take advantage of the unfavorable status quo, data should be turned into assets and assets should generate value.

second, in the field of data, we must solve the problem of data synergy. zhang yongwei said: "in terms of training software and systems, it is difficult for us to have as much data as tesla. for us, relying on the amount of data from a single car company is not enough. in the ai ​​era, competitiveness depends on the data is stacked, so we must solve the problem of large-scale data. this requires us to create our mechanism to promote data aggregation, so that everyone can input data into the platform and use data in accordance with market principles, and solve the problem of the current small scale of data. question."

the structural shortage of intelligent computing power has become the main contradiction

after solving the problem of insufficient data, domestic enterprises still need to solve the problem of large models.

large models are quite popular this year. the automotive industry is still facing a series of challenges before large-scale applications of large models are implemented. these challenges involve models, data, and computing power.

zhang yongwei suggested that automobile companies should develop and use large models now. in his view, to build competitiveness in intelligence, we need to solve the "new five domains" of automobiles: first, use artificial intelligence to solve the design problems of automobile electronic and electrical architecture, second, use artificial intelligence integration to solve power problems, and third, intelligent driving, the fourth is the smart cockpit, and the fifth is the car chassis. because the "new five domains" of automobiles are deeply integrated with artificial intelligence and large models.

"we need to use the logic of artificial intelligence for research and development to form new architectures and solutions. we must not only solve general and vertical models, but also solve models used by automobile companies to form our own ai competition in new development areas. power." zhang yongwei said.

in the information age, computing power is productivity. in the second half of smart cars, major companies are competing on ai and computing power. it can be said that chip computing power determines the intelligence limit of smart cars to a certain extent. the higher the computing power, the greater the potential of the car's intelligence level.

with the deep integration of automobiles and ai, end-to-end intelligent driving and large cockpit models are accelerating in cars, and the automotive industry's demand for intelligent computing power is growing rapidly. however, there is still a structural shortage of domestic automotive intelligent computing power, and there is a large gap in "mature" computing power with a complete software ecosystem.

zhang yongwei suggested that we should build automotive intelligent computing infrastructure and strengthen the co-construction and sharing of computing power. "in the era of artificial intelligence, what automobile companies lack is not production capacity. what the automobile industry currently lacks most is intelligent computing infrastructure. the main contradiction in the domestic automobile industry is the structural shortage of intelligent computing power."

zhang yongwei said that to complete the development and training of end-to-end intelligent driving, the demand for intelligent computing power must reach at least 1 eflops. the current average computing power of car companies is 3 eflops, and the ideal computing power is 100 eflops. we need to invest heavily in computing power, and we must continue to invest to create a scale effect around data, computing power, and algorithms.

according to public data, by the end of 2024, the total amount of computing power resources planned by the three major operators is 53 eflops. however, in terms of an end-to-end large model, the computing power required by an enterprise reaches 100 eflops.

“at this stage, how to solve the demand for computing power for smart driving and artificial intelligence is a top priority. we must not only ensure that computing power is available, but also pursue the low cost of available computing power, and even solve the problem of local computing power. "the problem of developing from immaturity to maturity." zhang yongwei explained that the computing power that nvidia has "has chips and has a software ecosystem" is called mature computing power, and the computing power that "has chips and lacks software" is what we call it. because of immature computing power. "our mission is to speed up the solution to the problem of immature domestic computing power, build mature computing power by enriching software and ecology, and reduce the problem of computing power being 'stuck' in hardware in the future."

in the era of intelligent network connection, the technological development of intelligent driving is changing from a single technical route to a development route of intelligent driving with chinese characteristics that integrates bicycle intelligence and car-road cloud. zhang yongwei said at the conference that my country should be committed to following the integrated development route of intelligent driving, that is, a new intelligent driving solution that integrates bicycle fsd and chelu cloud (c-fsd).

according to zhang yongwei, in the past, the industry generally believed that bicycle intelligence and car-road cloud were two technical routes. however, with the rapid development of ai large model technology, ai training chips and data closed-loop capabilities, the upper limit of bicycle intelligence technology has been greatly improved, and it can cope with the crisis. in most scenarios, bicycle fsd and chelu cloud have actually become two supporting forces towards the goal of autonomous driving. the two are rapidly integrating into a technical route, with bicycle fsd as the "basic score" and chelu cloud as the "additional score". items".

zhang yongwei believes that the new smart driving solution based on c-fsd can realize data sharing, computing power coordination, and model co-construction, lower the research and development threshold of enterprises, solve the current situation of uneven distribution of technology, resources, and markets to a certain extent, achieve equal rights for innovation, and release start-ups. the innovative vitality of enterprises.

(editor: zhang shuo review: tong haihua proofreader: zhai jun)