Meta is significantly expanding its collaboration with Nvidia, solidifying its commitment to artificial intelligence by purchasing millions of Nvidia's AI processors. The deal, details of which remain undisclosed, underscores Meta's aggressive push into the AI arena, according to a Bloomberg report.

The agreement includes Nvidia's cutting-edge Blackwell series chips, designed for AI data processing, as well as the next-generation Vera Rubin chips and Grace central processing units (CPUs). This marks the first time Meta will utilize Nvidia's Grace CPUs.

Meta's intensified focus on AI technologies is evident in this substantial investment. This move aligns with Meta's broader strategy, which includes exploring the development of its own AI chips, as reported previously.

Earlier this year, Meta announced plans to allocate over $136 billion to AI initiatives, exceeding previous years' investments, according to CNBC.

The reliance on standalone CPUs represents a notable shift, as Meta has historically paired CPUs with graphics processing units (GPUs) in its AI infrastructure.

Nvidia highlighted that this agreement is the first major deployment of its Grace CPUs independently of GPUs.

Meta is planning to construct more than 30 data centers in the coming years, primarily in the United States, including its two largest projects currently under construction: Prometheus and Hyperion.

According to the Bloomberg report, Meta alone accounted for 9% of Nvidia's total revenue last year, translating to $19 billion in revenue for Nvidia in the previous fiscal year.

Meta CEO Mark Zuckerberg expressed enthusiasm about the strengthened partnership, stating his company is eager to broaden its collaboration with Nvidia to deliver superior personalized AI to individuals worldwide.

Nvidia's Vice President of Accelerated Computing, Ian Buck, emphasized that the Meta deal reinforces Nvidia's position as a leader in AI chip manufacturing, uniquely capable of delivering high-performance silicon.

Buck described the Grace CPUs as excellent for data centers, saying that they can deliver twice the performance per watt of power consumption.