Meta (META.US) partners with NVIDIA Corporation (NVDA.US) to expand strategic cooperation and accelerate AI data center deployment. AMD (AMD.US) stock price falls in response.

Meta (META.US) announced a blockbuster partnership agreement covering multiple years with NVIDIA Corporation (NVDA.US), deploying millions of NVIDIA Corporation chips in its artificial intelligence data centers, including NVIDIA Corporation's latest standalone CPU and the next generation Vera Rubin system. The deal was announced to the public on Tuesday, but specific financial terms have not been disclosed. Meta CEO Mark Zuckerberg stated in a release that this expanded partnership will continue the company's long-term vision of "providing personal-level superintelligence to everyone in the world," a vision officially presented in July of this year. Following the announcement, Meta's stock price rose over 1% in after-hours trading, while NVIDIA Corporation's stock rose nearly 1%, and AMD (AMD.US) stock fell about 4%. In January of this year, Meta revealed plans to increase AI-related capital expenditures to a peak of $135 billion by 2026. Creative Strategies chip analyst Ben Bajarin noted that this cooperation with NVIDIA Corporation "is likely on a scale that could reach tens of billions of dollars," with a significant portion of Meta's capital expenditures expected to be directed towards NVIDIA Corporation's infrastructure. Although Meta has been using NVIDIA Corporation GPUs for at least a decade, the breadth of this partnership has significantly expanded. The standalone CPU is the major highlight, as Meta will be the first enterprise to deploy NVIDIA Corporation's Grace Central Processing Unit (CPU) on a large scale in data centers without integrating it with GPUs as in the past. NVIDIA Corporation stated that this marks the first large-scale independent deployment of the Grace CPU. Bajarin pointed out that these CPUs are primarily designed for inference and "agent" workloads, serving as key components for the Grace Blackwell or Vera Rubin racks, and Meta's adoption in scaled scenarios validates NVIDIA Corporation's full-stack infrastructure strategy for both CPUs and GPUs. The next generation Vera CPU is expected to be operational in Meta data centers by 2027. This multi-year agreement is also part of Meta's overall infrastructure expansion plan, with the company committing to invest around $600 billion in the United States by 2028 for data center and related infrastructure construction. Currently, Meta plans to build 30 data centers, with 26 located in the United States. Two major AI data center projects are under construction, namely the Prometheus project in New Albany, Ohio with a scale of 1 gigawatt and the Hyperion project in Richland Parish, Louisiana with a scale of up to 5 gigawatts. In addition to computing chips, the agreement also covers NVIDIA Corporation's Network-1 Technologies, Inc., including the Spectrum-X Ethernet switches used for large-scale AI data center internal GPU interconnects. Meta will also utilize NVIDIA Corporation's security technology to support AI features in products like WhatsApp. However, Meta is not placing all its bets on NVIDIA Corporation. Last November, reports indicated that Meta is considering introducing Alphabet Inc. Class C's tensor processing units (TPUs) in its data centers by 2027, causing NVIDIA Corporation's stock price to drop by about 4% at one point. Additionally, Meta is developing its own chips and continues to use AMD products. With the increasing demand for AI computing power and NVIDIA Corporation's continued supply constraints, industry giants are actively seeking "second sources" for supply, as seen in OpenAI's significant partnership with AMD in October. Currently, NVIDIA Corporation's Blackwell architecture GPUs have been backlogged for months, while the next generation Rubin GPU has recently entered mass production. Through this agreement, Meta is believed to have secured a sufficient supply of existing and next-generation computing power. The engineering teams of both companies will engage in deep collaborative design to optimize and accelerate Meta's cutting-edge AI models. At the model level, Meta is developing a new generation basic model called "Avocado," as a successor to the Llama series. However, the latest version released in the spring of last year did not generate much interest in the developer community. In recent months, Meta's stock price has experienced significant volatility, and its AI strategy continues to spark differing opinions on Wall Street. The company faced its worst single-day drop in three years last October due to aggressive AI investments, but after announcing better-than-expected revenue guidance in January, the stock price surged by about 10%.