MuskOn July 9, it announced that its artificial intelligence companyxABuilding a 100,000-chip Nvidia H100GPUThe move marks the end of xAI's talks with Oracle to expand its existing agreement to lease more Nvidia chips.
Mask stressed that this would be “the most powerful training cluster on a global scale, with a huge advantage in the lead”. He stated that the core competitiveness of xAI lies in speed, “it is the only way to close the gap”。
Prior to this, xAI had rented the computing power of 24,000 H100 chips from Oracle for trainingGrok2Musk revealed that Grok2 is currently in the final polishing stage and is expected to be released as early as next month.
Despite the termination of the expanded cooperation, Mask praised Oracle as “a great company”. He stressed that when the fate of the company depended on speed, “we had to take control of the wheel ourselves and not just sit in the back seat”。
It is worth noting that in May this year, it was reported that xAI and Oracle were close to reaching an expanded cooperation agreement worth approximately $10 billion. However, this announcement seems to indicate that xAI is shifting to a strategy of building its own AI infrastructure.
This decision reflects the fierce competition in the AI field and the key role of top computing resources in this competition. With the deployment of xAI's new supercomputer, the industry will pay close attention to its effect on improving the performance of AI models.