Intel is planning a new strategy to better compete with rivals Nvidia and AMD in the fast-growing AI chip market. The company plans to begin shipping a new artificial intelligence graphics processing unit (GPU) called Crescent Island by the end of this year. The chip is designed to target AI inference workloads and will use lower-cost memory and cooling technology compared to competitors from Nvidia and AMD, according to Kevork Kechichian, head of Intel data centers.Keshitian told the Financial Times that Intel is taking a more cautious approach after previous attempts to gain traction in the AI accelerator market failed to yield the desired results. An executive said the company: “Start from the basics” The company is working to rebuild its position in the semiconductor field with a focus on AI.“We decided to start rebuilding muscles with AI. [but] I’m not particularly aiming for [the training market] Based on past experience.” Keshichian said.Crescent Island GPUs focus on inference tasks that occur when users interact with AI models and receive responses. This is different from AI training, where Nvidia currently maintains a dominant position in the market. Intel’s early AI training chip Gaudi struggled to catch on, and plans for a successor were scrapped last year.
Intel aims for low costs AI infrastructure To compete with Nvidia and AMD
One of Intel’s key differentiators is our commitment to lower deployment costs for our customers. The Crescent Island chip uses LPDDR5 memory rather than high-bandwidth memory (HBM) commonly used in Nvidia and AMD AI accelerators. The new chip is also air-cooled, eliminating the need for liquid-cooling infrastructure often required for high-performance AI systems.Kechichian said the chip has been in development for the past 18 months and will initially be shipped to customers in limited quantities before being rolled out broadly.The project marks Intel’s first push into large-scale AI infrastructure under CEO Lip Vu Tan, who took over last year following a change at Intel’s leadership. Mr. Tan has focused on reducing costs, refining product strategy, and strengthening Intel’s position in key semiconductor markets.Intel is also evaluating whether versions of Crescent Island can be sold in China while complying with U.S. export regulations. Current trade regulations limit the ability of Nvidia and AMD to sell certain advanced AI chips in the country.“There are stages.” [the chip] That might be fine. . . And over time, we see that it’s clear that there is a demand for that particular price point in that particular market. ” Ketian pointed out.The company also aims to manufacture more of its future data center products in-house, rather than relying on outside foundries. Intel recently launched advanced PC and server chips that it makes in-house, after years of relying on outside manufacturing partners.“For all of our data center products, we’re actively moving to our own foundries. That’s the overall intent.” Ketian added.Intel believes producing chips in-house will further reduce costs and provide another way to compete with rivals that continue to rely on outside manufacturing partners for advanced AI processors.
