Nvidia has secured a long-term supply agreement with Meta Platforms that covers the sale of millions of Artificial Intelligence chips over several years, spanning both current and future processor generations. The deal brings Nvidia into more direct competition with established CPU providers such as Intel and AMD, as it now includes standalone central processors alongside accelerators. Nvidia disclosed the partnership in mid-February 2026 but did not reveal a specific contract value, underscoring both companies’ reluctance to quantify the financial scope publicly.
The agreement covers Nvidia’s current Blackwell architecture and the upcoming Rubin generation of Artificial Intelligence accelerators, along with Grace and Vera Arm-based CPUs that were first introduced starting in 2023 as companions to its Artificial Intelligence chips. Nvidia is now positioning these CPUs for broader roles, including running Artificial Intelligence agents and handling everyday data center tasks such as database management. Meta, for its part, is pursuing an aggressive infrastructure buildout, including an Artificial Intelligence data center called “Hyperion” in Holly Ridge, Louisiana, planned for approximately 10 billion dollars and intended to reach a size described as a quarter of Manhattan. This push has contributed to a strategic split with long-time Artificial Intelligence chief Yann LeCun, who left to co-found AMI Labs after arguing that Transformer models will not deliver superintelligence and that new World Model architectures are needed.
Meta’s leadership has concluded it must massively increase computing capacity to train and serve its Artificial Intelligence models, especially if billions of users worldwide use Meta AI multiple times daily, which would require corresponding computing power for inference. Industry analysts estimate the total value of the Nvidia agreement at approximately 50 billion U.S. dollars, and Meta is cited among the four largest customers that accounted for approximately 61 percent of Nvidia’s revenue in the most recent business quarter. Nvidia promotes its Grace CPUs as offering approximately half the power consumption for certain common tasks, with successor Vera expected to improve efficiency further, a key factor as U.S. states scrutinize the energy impact of new data centers. The deal follows unfulfilled reports in November 2025 about Meta’s billion-dollar investments in Google Tensor Processing Units and comes as Meta continues to develop its own Artificial Intelligence semiconductors while still relying heavily on external suppliers.
Energy efficiency has become a central constraint on Artificial Intelligence infrastructure growth, and some U.S. states are already weighing limits on new data centers, prompting Meta to stress performance per watt as a strategic priority. In addition to accelerators and CPUs, Meta has ordered extensive quantities of Nvidia Ethernet switches to optimize data center networking. The announcement was timed one week before Nvidia’s quarterly earnings release, signaling sustained demand for its products despite intensifying competition and in-house chip efforts by hyperscalers. Meta chief executive Mark Zuckerberg framed the initiative around an ambition to provide personal superintelligence to every person in the world using the ordered technology, while the contract helps Nvidia reinforce its dominant position in the Artificial Intelligence chip market against both alternative processor architectures and customers’ own chip programs.
