Meta expands collaboration with NVIDIA, intensifying industry competition

On February 17th, Meta and NVIDIA, two major American technology giants, announced the establishment of a new long-term partnership. The collaboration not only involves large-scale deployment of chips but also covers full-stack optimization from hardware to software. In an era of increasingly fierce competition in artificial intelligence, this major news has garnered widespread industry attention. Following the announcement, Meta and NVIDIA’s stock prices rose in after-hours trading, while AMD’s stock price temporarily dropped more than 4%.

According to disclosures from both parties, one of the core aspects of this cooperation is that Meta will deploy millions of NVIDIA chips in its data centers, including Blackwell architecture GPUs, next-generation Rubin architecture GPUs, and Arm-based Grace central processors. Notably, this marks NVIDIA’s first large-scale independent deployment of the Grace CPU, challenging the traditional monopoly of x86 architecture. Additionally, Meta plans to introduce more powerful Vera series processors by 2027 to further strengthen its layout in high-efficiency AI computing power.

This collaboration goes far beyond simple hardware procurement. The engineering teams from both companies will collaborate on the software and hardware design of Meta’s next-generation large-scale language models (such as the successor to Llama4, codenamed “Avocado”) to optimize underlying performance. By integrating CPU, GPU, network technologies (such as Spectrum-X Ethernet platform), and software ecosystems, NVIDIA provides Meta with a unified solution covering training, inference, and data processing.

NVIDIA founder Jensen Huang stated that no other company can deploy AI at Meta’s scale. Meta combines cutting-edge research with industrial-grade infrastructure, creating the world’s largest systems for billions of users. “No company worldwide can match Meta’s AI deployment scale, and the ability to co-design will maximize the potential of both parties’ technologies,” he said.

Through deep collaboration in CPU, GPU, networking, and software design, NVIDIA will deliver a complete platform to Meta’s researchers and engineers, helping them build the next generation of AI frontiers. Meta CEO Mark Zuckerberg said that expanding cooperation with NVIDIA and leveraging its Vera Rubin platform to build leading clusters will enable the provision of personal superintelligence to everyone around the world. For Meta, this partnership is not only a supplement to its self-developed chip strategy but also a key move to counter competitive pressures.

Although Meta has actively developed its own AI chips in recent years, this large-scale procurement of NVIDIA chips is interpreted as a long-term lock-in of external computing power needs, ensuring it remains competitive in AI commercialization against rivals like Google and Microsoft. Zuckerberg emphasized that this move is a crucial foundation for realizing the vision of providing personal superintelligence to global users.

Additionally, Meta mentioned in its statement that it will integrate NVIDIA’s security technology into the AI features of its communication app WhatsApp. Zuckerberg said that by introducing NVIDIA’s confidential computing technology, Meta can enhance performance while meeting strict data security and privacy requirements.

Analysts believe that the scale of this cooperation could reach hundreds of billions of dollars. Last month, Meta announced that its capital expenditure in 2026 could reach up to $135 billion, much of which will be allocated to AI infrastructure development. Chip analyst Ben Bajarin of Creative Strategies pointed out, “Meta’s large-scale adoption validates NVIDIA’s full-stack infrastructure strategy, which combines CPU and GPU deployment.” He emphasized that these chips are specifically designed for inference workloads and intelligent agent tasks, as the AI industry shifts from training to inference.

It is noteworthy that Meta has been pushing its own AI chip strategy in recent years, aiming to optimize performance for its unique workloads and reduce costs. However, informed sources reveal that related projects have faced technical challenges and deployment delays, leading the company to continue relying on NVIDIA’s mature solutions. Nonetheless, Meta has not completely abandoned external options; in November last year, there were reports that Meta was considering introducing Google’s Tensor Processing Units (TPUs) into its data centers by 2027 to diversify its computing sources.

This cooperation has directly impacted NVIDIA’s competitors, especially AMD. Following the announcement, AMD’s stock price temporarily fell over 4% in after-hours trading, while Meta and NVIDIA’s stocks rose by 1.5% and 1.8%, respectively, indicating strong market recognition of this partnership.

Market analysts generally believe that this massive order dispels concerns about large clients shifting to self-developed chips. Wall Street experts note that NVIDIA’s bundled approach—combining Grace CPUs, Rubin GPUs, and Spectrum-X networking technology—successfully consolidates its leadership in AI infrastructure. Meanwhile, Meta’s substantial capital expenditure plans highlight its determination to compete for AI dominance. However, some also warn that Meta might face profit pressures due to this scale of investment, though its long-term compute power security strategy is still viewed as an effective defense against competitors.

From an industry perspective, the collaboration between NVIDIA and Meta reflects a shift in AI computing focus from model training to inference. Inference tasks require efficient, low-latency computing, and NVIDIA’s Grace CPU is specifically designed for this need, with performance and power efficiency significantly better than traditional CPUs. As Meta’s AI inference demands—serving billions of users daily—continue to grow, adopting specially optimized Grace CPUs will be key to controlling energy consumption and improving efficiency.

This partnership also signals a major reshuffle in industry competition. For traditional chip giants like Intel and AMD, Meta’s large-scale shift toward Arm-based CPUs is a stark warning, indicating a structural shift in the power dynamics of the hyperscale data center market. According to multiple authoritative reports, leading companies such as Amazon, Alphabet, and Meta are expected to maintain high capital expenditure levels into 2026, with hundreds of billions of dollars flowing into data center construction centered on NVIDIA GPUs. This sustained, highly certain demand effectively alleviates previous market concerns about “insufficient ROI on AI investments.”

(Reported by Li Qiang, Economic Observer)