NVIDIA and Meta announce multi-year strategic partnership to build a new future for AI infrastructure

This is not the first collaboration between the two parties. Meta has been using NVIDIA GPUs for over a decade, and this new agreement elevates their cooperation from a single hardware procurement to a full-stack binding of chips, networks, software, and security. According to the agreement, Meta will deploy millions of NVIDIA chips (including the new standalone CPU, Blackwell architecture GPU, and the next-generation Vera Rubin system) on a large scale in global hyperscale AI data centers, building an efficient full-stack AI computing power system.

The core highlight of this collaboration is the first large-scale independent commercial deployment of NVIDIA's Grace CPU. Meta will become the first enterprise globally to deploy this chip on a large scale independently. Grace CPU is based on the Arm architecture and equipped with 144 Neoverse V2 cores, boasting a power-efficiency ratio five times that of traditional x86 CPUs. It is tailored to meet Meta's data center requirements and will work synergistically with Blackwell GPUs and the next-generation Vera Rubin system. The next-generation Vera CPU is planned for deployment in 2027

The cooperation between the two parties extends to the entire AI infrastructure chain: at the network level, NVIDIA Spectrum-X switches are introduced to address communication latency in AI model training; at the security level, NVIDIA confidential computing technology is integrated into WhatsApp AI functions; at the research and development level, the Meta Avocado large model (successor to Llama) is jointly optimized to accelerate model training and deployment.

Strategic cooperation stems from the alignment of visions and complementary strengths between both parties. Meta plans to invest $600 billion in the United States by 2028 to build 30 hyperscale data centers, and by 2026, its AI investment will reach up to $135 billion. Partnering with NVIDIA can secure high-end chip supply, alleviate stock shortages, and provide infrastructure support for large model research and development.

For NVIDIA, this collaboration marks a pivotal step in its full-stack transformation. Leveraging the commercialization of Grace CPU, NVIDIA has officially ventured into the server CPU market, completing its transition from being a sole GPU supplier to a full-stack solution provider. Analysts estimate the agreement to be worth tens of billions of dollars, and Meta's large-scale adoption further solidifies NVIDIA's position as an industry benchmark.

The collaboration between the two parties has a profound impact on the global AI industry, reshaping the competitive landscape of infrastructure. After the announcement, Meta and NVIDIA's after-hours stock prices rose while AMD's fell, reflecting the market's recognition of NVIDIA's advantages. At the same time, it promotes the transformation of AI infrastructure towards system-level optimization, and the large-scale application of Arm architecture will revolutionize the server chip market.

It is worth noting that Meta has not relied solely on Nvidia as a supplier. It has simultaneously advanced its self-developed chips and adopted AMD products. Previously, it had also planned to adopt Google's Tensor Processing Unit (TPU) in data centers by 2027. Behind this diversification strategy lies the rational consideration of global AI giants regarding supply chain security amid the tight supply of Nvidia chips. However, it is undeniable that Nvidia's full-stack layout in the CPU and GPU fields, as well as its deep collaborative advantage with Meta, are still difficult to be replaced by competitors in the short term.

Looking ahead, as the strategic cooperation between the two parties gradually takes shape, millions of NVIDIA chips will power Meta's global hyperscale data centers. The development of the Avocado large model will continue to accelerate, and AI features in products like WhatsApp will become even more secure and efficient. Meanwhile, the democratization of "personal superintelligence" will also see new breakthroughs. Simultaneously, NVIDIA will use this collaboration as a benchmark to continuously expand the boundaries of the AI computing power ecosystem and drive the iterative upgrading of full-stack computing technology. The full-stack collaboration model explored by both parties will also provide a reference model for AI infrastructure cooperation among global technology companies.

In today's era of rapid iteration in AI technology and explosive growth in computing power demand, the strategic cooperation between NVIDIA and Meta is not only a "powerful alliance" between the two companies, but also an important signal indicating the global AI infrastructure's progression towards high-quality development. This comprehensive collaboration spanning hardware, software, and research and development will continue to unleash the vitality of technological innovation, break the bottleneck of computing power, lay a solid foundation for the large-scale application and inclusive development of AI technology, and jointly write a new chapter in the future of AI infrastructure.