Musk claims new Tesla AI supercluster will exceed 500 MW
Musk claims new Tesla AI supercluster will exceed 500 MW (Photo: Unsplash)
Tesla's new AI supercluster was finally unveiled by CEO Elon Musk and it is expected to ramp up to over 500 MW, which will make it one of the largest if not the largest in the world. At the same time, Musk emphasised that Tesla had been leading the pack in experimenting with next-generation AI chip in auto.
Back in several months, there were signs that Tesla had been facing issues in laying the groundwork for an expansion at Gigafactory Texas to house a new giant supercomputer for training Tesla’s AI. At first to assemble a 100 MW cluster by August, Musk refocused Tesla’s construction endeavour by halting other projects.
When asked about the expansion, drone footage revealed the expansion, Musk said that it will go up to over 500 MW in the next 18 months. This year, it will be sized to the order of 130 MW of power and cooling capacity, with long-term growth plans of more than 500 MW over the stated period. Musk emphasised a split between Tesla's own AI hardware and Nvidia's technology, highlighting Tesla's competitive strategy: It should be play to win or not to play at all.
Earlier, there was confusion as Tesla was the internal moniker for the project and usually everything associated with Dojo, which is Tesla’s home-grown supercomputing hardware. Some sources pointed that Nvidia’s compute power was also going to be tapped.
Musk has now clarified that Tesla has plans to continue using Tesla-made hardware (Such as the next generation AI5, previously referred to as HW5) alongside Nvidia and other suppliers. However, Musk’s comments do imply some uncertainty about whether Tesla’s HW4 computers, which are employed for in-car uses with their proprietary chips, will be employed in the training clusters or otherwise, to what extent he is referring to Tesla’s computing strategy in general.
On balance, Tesla is set to increase computing power manifold to power its autonomous vehicles, rely on a mix of self-developed chips and those from established players in the domain, including Nvidia.
