Intel shifts toward selling its own AI chips for data centers

In short: Intel is leaning more on selling its own AI-focused chips to data centers, not just making chips and designs for other companies.

What's going on

Intel has long been known for making processors and also for making chips for other companies. Now it is putting more of its effort into selling its own chips that are built for AI work in large data centers (big buildings full of computers that run online services).

One current product is Gaudi 3, an AI accelerator that Intel says has been shipping since late 2024. An accelerator is a special chip that helps with AI tasks, a bit like adding a dedicated engine to handle heavy lifting. Intel says Gaudi 3 is available through cloud partners, meaning customers can rent access to it instead of buying hardware.

Intel is also leaning on its Xeon server processors, including Xeon 6, which it says is seeing unexpected demand heading into late 2025. These CPUs are not the same as the AI chips many people hear about, like graphics processors. But Intel argues CPUs still matter for AI because they help feed data to the AI chips and manage large clusters of machines (like a factory line where one part sorts and moves materials while another part does the specialized work).

What to watch

Intel has more chips on its roadmap, including an inference-focused GPU called Crescent Island that is expected to be tested by customers in the second half of 2026, and a new Xeon generation called Diamond Rapids planned for the second half of 2026. Intel is also advancing its manufacturing processes, including 18A for upcoming chips and 14A later in 2026 for outside customers. The big question is whether Intel can produce enough chips on time as demand rises and competition in AI data centers stays intense.

Source: NYTimes