Elon Musk's artificial intelligence startup xAI and tech giant Oracle have reportedly closed talks for a potential $10 billion server deal, in which Grok developers rented Nvidia AI chips from the cloud provider to train AI chatbots. The company rented GPUs to train the Grok 2 AI chatbot.
Musk said xAI contracted for 24,000 Nvidia H100s GPUs from Oracle and used those chips to train Grok 2. The company now offers Grok 1.5, and Musk recently said Grok 2 will debut in August of this year.
“Grok 2 is currently undergoing tweaks and bug fixing and will likely be released next month,” Musk added.
Musk's xAI will build its own data center
The Information reported, citing people involved in the talks, that xAI was buying the chips to build its own data centers, but Musk later confirmed this in a post from X, another of his companies.
He said his AI company is building its own systems using Nvidia's H100 graphics processing unit.
“xAI has built 100,000 H100 systems in-house, [the] “The time to completion is minimal. We are aiming to commence training later this month. This will be by far the strongest training cluster in the world,” he added.
“The reason we decided to develop 100,000 H100s and our next major system in-house is because our fundamental competitive advantage is being faster than any other AI company. That's the only way we're going to catch up,” Musk emphasized. Notably, Grok 3 is scheduled to be released by the end of this year.
Why xAI is terminating its contract with Oracle
A multi-year deal to rent Nvidia processors from Oracle for a planned supercomputer is reportedly in the works, but negotiations have stalled over issues including Musk's demand that the supercomputer be built faster than Oracle envisions.
“Oracle is a great company and there's another promising company involved in the OpenAI GB200 cluster, but when our fate depends on being by far the fastest, we have to take the wheel, not be a backseat driver,” Musk said in the post.
Reuters reported, citing sources familiar with the matter, that the specific capabilities Oracle was discussing with xAI had been contracted to another customer.
“We are constantly in discussions with our customers about future capacity and continue to discuss with xAI about their infrastructure needs,” the source said. Oracle also expressed concerns that there is insufficient power supply in xAI's desired locations.
Musk said xAI contracted for 24,000 Nvidia H100s GPUs from Oracle and used those chips to train Grok 2. The company now offers Grok 1.5, and Musk recently said Grok 2 will debut in August of this year.
“Grok 2 is currently undergoing tweaks and bug fixing and will likely be released next month,” Musk added.
Musk's xAI will build its own data center
The Information reported, citing people involved in the talks, that xAI was buying the chips to build its own data centers, but Musk later confirmed this in a post from X, another of his companies.
He said his AI company is building its own systems using Nvidia's H100 graphics processing unit.
“xAI has built 100,000 H100 systems in-house, [the] “The time to completion is minimal. We are aiming to commence training later this month. This will be by far the strongest training cluster in the world,” he added.
“The reason we decided to develop 100,000 H100s and our next major system in-house is because our fundamental competitive advantage is being faster than any other AI company. That's the only way we're going to catch up,” Musk emphasized. Notably, Grok 3 is scheduled to be released by the end of this year.
Why xAI is terminating its contract with Oracle
A multi-year deal to rent Nvidia processors from Oracle for a planned supercomputer is reportedly in the works, but negotiations have stalled over issues including Musk's demand that the supercomputer be built faster than Oracle envisions.
“Oracle is a great company and there's another promising company involved in the OpenAI GB200 cluster, but when our fate depends on being by far the fastest, we have to take the wheel, not be a backseat driver,” Musk said in the post.
Reuters reported, citing sources familiar with the matter, that the specific capabilities Oracle was discussing with xAI had been contracted to another customer.
“We are constantly in discussions with our customers about future capacity and continue to discuss with xAI about their infrastructure needs,” the source said. Oracle also expressed concerns that there is insufficient power supply in xAI's desired locations.