Intel and Google deepen AI tie

Google just bet that the most valuable chip in an artificial intelligence data center is not always the graphics processor. On April 9, Intel and Google said Google Cloud will keep using Intel Xeon central processors across future systems while the two companies also co-develop custom infrastructure chips together. (intel.com) A cloud server does two jobs at once. One part runs your application, and another part handles the plumbing like moving data, encrypting traffic, and talking to storage drives. (cloud.google.com) Google already has a name for that plumbing hardware: Titanium. Google says Titanium offloads networking and storage work onto dedicated hardware so the main processor can stay focused on the customer’s code. (cloud.google.com) Intel’s matching term is infrastructure processing unit, which means a chip built to take those background chores away from the central processor. Intel says an infrastructure processing unit is meant for infrastructure management in cloud networks rather than for running the customer application itself. (builders.intel.com) That is what changed this week. Intel and Google said they are expanding co-development of custom application-specific integrated circuit infrastructure processing units, while Google Cloud also keeps deploying multiple generations of Xeon processors in its fleet. (intel.com) Google did not frame Xeon as a legacy chip hanging around for old workloads. The companies said Xeon 6 processors are already powering Google Cloud’s C4 and N4 instances, which Google uses for jobs ranging from large artificial intelligence training coordination to latency-sensitive inference and general-purpose computing. (intel.com) That detail matters because artificial intelligence systems need a traffic cop as much as they need an engine. Graphics processors train models, but central processors still schedule work, feed data, run databases, and keep the whole cluster from stalling. (cnbc.com) Google has been building more of its own silicon for years, including Tensor Processing Units for artificial intelligence and Axion central processors based on Arm designs. This Intel deal shows Google is still mixing in outside chips where it wants specific performance, supply, or integration advantages. (cnbc.com) Intel needs that message badly. The company has been trying to prove that the artificial intelligence boom is not only about Nvidia graphics processors, and Intel chief executive Lip-Bu Tan said in the announcement that “scaling AI requires more than accelerators” and that central processors and infrastructure processing units remain core parts of the system. (intel.com) So this is less a single chip launch than a blueprint for how big clouds now build computers. Instead of one general-purpose server doing everything, Google and Intel are carving the machine into specialist parts: one chip for applications, one chip for networking and storage, and separate accelerators for artificial intelligence math. (cloud.google.com, intel.com)