On April 28th, Tesla CEO Musk, who had just arrived in Beijing in the afternoon, stated on social media that "Tesla will invest approximately $10 billion this year in AI training and reasoning, with reasoning mainly used for automobiles."
Musk also added that any company that does not achieve an annual expenditure of $10 billion or cannot efficiently deploy cannot compete in the market.
Tesla just added $500 million in new investment in January this year, purchasing approximately 10000 Nvidia H100 GPUs for autonomous driving training. According to data, Tesla may have over 30000 H100 GPUs.
Musk recently stated that Tesla may be the second company with Nvidia H100 GPUs, second only to Meta, and third in xAI.
Domestic electric vehicle manufacturers have previously stated that they will increase investment in autonomous driving. Earlier this year, Xiaopeng Motors announced plans to invest 3.5 billion yuan in artificial intelligence fields such as autonomous driving this year; According to publicly available data, Xiaomi's intelligent driving team is currently expected to invest 1.5 billion yuan in research and development expenses per year.
However, according to the plan, Xiaomi will invest 11 billion to 12 billion yuan in new business research and development, including automobiles, in 2024. At the same time, it will further increase investment in key areas such as underlying operating systems, and plans to expand the scale of its intelligent driving team.
Tesla's heavy investment in AI is aimed at the future business opportunities of autonomous driving. Musk stated that a self driving taxi (Robotaxi) will be launched on August 8th this year.
Recently, Tesla has gradually started large-scale push to North American users based on end-to-end FSD V12. In short, FSD V12 achieves the goal of autonomous driving through a large amount of video training data, which allows Tesla to save hundreds of thousands of lines of code. However, training such AI autonomous driving requires a lot of computing power. Musk announced in a live broadcast in August last year that Tesla will soon launch a GPU cluster consisting of 10000 Nvidia H100 units to train the new FSD system.
AutoX founder and CEO Xiao Jianxiong told First Financial reporters that Tesla FSD uses end-to-end deep learning algorithms, which have improved its performance significantly compared to before. In the past, it relied on manually writing rule-based algorithms, but now it uses a large model network to train the network with data.
"Tesla, as a cost control expert, also realizes that selling electric vehicles alone is difficult to make a profit, especially in the context of price wars. Now, we must rely on selling cars with technological content to make money," said Xiao Jianxiong.