Nvidia pushes into Intel and AMD's turf with a 'multigenerational' Meta deal

Meta is significantly enhancing its partnership with Nvidia through a groundbreaking agreement described as 'multigenerational.' This deal, unveiled on Tuesday, entails the construction of data centers that will utilize millions of Nvidia's cutting-edge chips, both current and forthcoming, for artificial intelligence training and inference. This strategic move highlights Meta's increasing dependency on Nvidia, even as the tech giant develops its own custom chips and collaborates with competing manufacturers like AMD. Recent reports indicate that Meta has also considered using Tensor Processing Units (TPUs) designed by Google, which could lead to speculation regarding the potential for Nvidia's partnership to overshadow its TPU discussions, according to Patrick Moorhead, chief analyst at Moor Insights & Strategy. The timing of this agreement comes as competition in the AI infrastructure sector intensifies. While Nvidia currently leads the market, other players such as Google, AMD, and Broadcom are actively working to challenge its supremacy. The deal is particularly significant as it allows Meta to leverage both Nvidia's Graphics Processing Units (GPUs) and Central Processing Units (CPUs). Traditionally dominated by Intel and AMD, CPUs are crucial for general computing tasks and serve as the backbone of modern computing systems, while GPUs address specialized needs requiring higher computational power, such as AI and gaming graphics. By supplying both types of processors, Nvidia stands to gain a larger share of Meta’s budget and solidify its position within the company's AI framework. Despite the increased competitive pressure this may create, experts like Moorhead suggest that the soaring demand for infrastructure means Nvidia's competitors are unlikely to experience significant downturns in the immediate future. Nvidia's ambitions in the CPU market have become clearer, with plans to market its upcoming Vera CPU as an independent product. This shift illustrates the growing importance of CPUs as AI workloads evolve from model training to inference tasks. Rob Enderle, principal analyst at Enderle Group, noted that CPUs offer cost-effectiveness and greater energy efficiency for inference processes. Furthermore, both Moorhead and Enderle point out that Meta’s choice to procure GPUs and CPUs from a single vendor streamlines operations, a preferred strategy for many chief information officers seeking simplified problem-solving channels. As part of this partnership, Meta will also incorporate Nvidia's networking technologies within its data centers and utilize its confidential computing solutions for AI functionalities in WhatsApp. The collaboration will also extend to deploying Nvidia's next-generation Vera CPUs, moving beyond the existing Grace CPU model.