Artificial intelligence no longer only needs graphics cards (GPU) to train models, it also needs processors (CPU) to deploy them. It is this shift which explains the multi-year agreement announced between Google and Intel.
A Google and Intel partnership for AI
Google Cloud is already deploying Intel’s Xeon 6 processors on its C4 and N4 instances which cover a broad spectrum: large-scale AI training coordination, latency-sensitive inference and general-purpose computing. The deal commits Google to several future generations of Xeon chips, with an explicit goal of improving performance, power efficiency and total infrastructure cost. “Scaling AI requires more than accelerators: it requires balanced systems. CPUs and IPUs are essential to delivering the performance, efficiency and flexibility that modern AI workloads demand.”a déclaré Lip-Bu Tan, directeur général d’Intel.
Beyond Xeons, Google and Intel reaffirm their co-development of IPUs (infrastructure processing units), programmable accelerators based on ASICs (dedicated integrated circuits) whose joint work dates back to 2022. Their role is precise: offload CPUs from network, storage and security functions to free up computing power for AI-related workloads.
Google maintains a dependence on Intel processors that dates back to its very first server racks, almost 30 years ago. This new agreement is therefore part of a continuity, but in a radically different context: Intel seeks to regain credibility in AI after several years dominated by Nvidia. The company is increasing signals in this direction with participation in Elon Musk’s Terafab project with SpaceX and Tesla and a planned repurchase of its stake sold to Apollo Global Management in its Irish factory where Xeon servers are manufactured.
