Tesla failed to compete with Nvidia, is its self-developed chip going to be a failure?

Although Huo Chu emphasized the investment scale of US$500 million at the press conference, Tesla deliberately downplayed this figure on social media X, pointing out that the company's investment in Nvidia hardware will exceed this amount in 2024.

$500 million is equivalent to about 10,000 H100 GPUs.

On April 8, Meta ranked first on the list of Nvidia H100 GPUs hoarded by X platform user "The Technology Brother" with 350,000 units.

Musk expressed dissatisfaction with the ranking of Tesla and xAI at 10,000 yuan, and pointed out that "if the calculation is correct, Tesla should be second and xAI will be third."

This means that Tesla may have 30,000 to 350,000 H100 GPUs, and xAI may have about 26,000 to 30,000.

Musk, who has been at loggerheads with Zuckerberg, exposed the real situation in his unconvinced words: at least for now, Dojo's self-developed chips have failed and it has turned to Nvidia.

Musk said that to remain competitive in the field of artificial intelligence, at least billions of dollars will be needed each year, and the company will expand its purchases of products from Nvidia's competitor AMD.

However, "whydoesthisitch" believes that it will take until 2027-2028 for Dojo's computing power to reach 100 exa flops, by which time the computing power of mainstream cloud service providers such as Amazon will have reached the zettaflop level.

He said that the current chip performance of Dojo can only reach 10%-35% of H100. When it catches up with H100, NVIDIA will have already run a long way on the new generation of Blackwell.

Nelson believes that at least Musk realizes that buying chips is still the most cost-effective.