With the application of end-to-end and generative large models, automobile companies entering the second half of intelligent electrification are accelerating their transformation into AI companies. In this transformation, computing power and algorithms will become key factors determining the future fate of the enterprise.

"Xpeng will explore future mobility and become a global AI automotive company." As Tesla's FSD is about to enter the Chinese market, Xpeng Chairman He Xiaopeng, after experiencing Tesla's AI capabilities in autonomous driving during a trip to the US, announced in late August the company's upgrade to "Xpeng AI Automobile Company." As planned, Xpeng's AI system includes AI automobiles, AI robots, and flying cars, as well as AI chips.

Tesla is the pioneer in applying end-to-end applications in the automotive industry. Since 2014, Tesla has introduced its self-developed Autopilot first version, followed by the more feature-rich Enhanced Autopilot in 2017. By November 2022, Tesla further opened up the FSD Beta intelligent driving system as an option to North American car owners. The support of AI has also become a key driver for Tesla's high valuation.

In addition to Xpeng, which is not hesitant to rename itself, new energy carmakers like Nio and traditional automakers such as Chery are all focusing on AI. In addition to intelligent cabins and smart driving, AI has entered multiple areas and processes including research, production, supply, and marketing.

"The whole company is now using AI code tools to generate some code, and production efficiency has already increased by 30%. The best application scenario for artificial intelligence is actually in the smart electric automobile industry." Li Bin, Chairman of Nio, recently stated that a successful smart electric car company must have AI research and development capabilities as well as operational capabilities; a successful smart electric car company must also be a successful AI company.

Chery is currently developing Xiongan Technology, Dazhuo Intelligence, and Next AI Research Institute, focusing on three major areas: intelligent cockpit large models, end-to-end smart driving large models, and AI intelligent management. By empowering new products with AI, Chery aims to achieve intelligent commercial applications. "As automobiles upgrade from transportation vehicles to a new generation of mobile intelligent terminals, AI not only redefines cars but also reshapes the management mode and operational boundaries of automobiles." Wang Lang, Deputy General Manager of Chery, said that AI is the new power of the times, an important tool for achieving uniqueness and differentiation in car products."Every era has its main contradictions and new driving forces for that era, just as Chery firmly chose independent engine research and development 20 years ago, AI is definitely the significant opportunity Chery must seize in the next 20 or 40 years."

With the application of AI, cars need to process large amounts of sensory data, decision-making, and control tasks, thus sharply increasing the demand for computing power. Industry experts believe that in the era of AI competition, automotive companies must form core competitiveness in computing power, chips, and algorithms.

To complete the research and training of end-to-end intelligent driving, the initial computing power required is 1 EFLOPS, which requires a substantial investment in computing power and continuous investment to build a scaled team around data, computing power, and algorithms. Vice Chairman and Secretary-General Zhang Yongwei of the China Electric Vehicle 100 People Association pointed out that after entering the AI era, the automotive industry's most lacking infrastructure is intelligent computing, not production capacity. Without clusters of thousands or tens of thousands of computing power cards, without thousands or tens of thousands of algorithm teams, it is difficult for companies to compete in new tracks.

In Zhang Yongwei's view, besides building local computing power, it is necessary to accelerate the resolution of problems such as immature domestic computing power, weak supporting toolchains, and ecological capabilities to reduce the risk of computing power being hardware-constrained.

Data shows that the current computing power of domestic car manufacturers is below 10 EFLOPS. By the end of 2024, the planned computing power for China Mobile, China Telecom, and China Unicom is 17 EFLOPS, 21 EFLOPS, and 15 EFLOPS respectively, totaling 53 EFLOPS for the three major operators. However, for large-scale 'end-to-end' models, the ideal computing power required by an enterprise is as high as 100 EFLOPS. In comparison, Tesla's computing power level is 100 EFLOPS, equal to the sum of all car manufacturers' computing power.

According to Zhang Yongwei's vision, future efforts should be made to strengthen computing power collaboration and sharing from four aspects to enhance resource utilization efficiency. Firstly, concentrate computing power to improve model iteration efficiency; Secondly, share basic models to reduce redundant resource utilization; Thirdly, coordinate software and hardware to enhance compatibility; Fourthly, build ecosystems to enhance software development ecology.

Developing computing power is a long-term systematic project that requires joint construction by the entire industry. Director Li Ming of Jiang Motors Holding Company believes that as the demand for large computing power in smart automobiles continues to grow, Anhui Province should further strengthen the construction of computing power centers and cloud platforms to make them part of public service support and empower business development.