Intel's attempting to break into the AI market once more, but this time avoiding Nvidia's dominance in training by going for inference

Intel's Strategic Shift: Focusing on AI Inference to Avoid Nvidia's Dominance

Intel, under the leadership of Pat Gelsinger, initially aimed to challenge Nvidia and AMD in the AI market with its Gaudi series of GPUs. However, the limited sales of these products suggested that the chip giant might be stepping back from the race. That assumption, however, is incorrect — Intel is back in the game, this time with a new strategy that targets the AI inference market, a space less dominated by Nvidia.

According to a report by the Financial Times, Kevork Kechichian, Intel’s general manager of its data centre group, shared more details about Crescent Island during this year’s Computex event. While the company expanded on its earlier announcements, the core idea remains the same: Crescent Island is a vastly different approach to machine learning than its predecessor, Gaudi. Unlike the previous attempt, Crescent Island is designed to focus on inference, the stage where AI models are used for real-world tasks after being trained.

Previously, Intel’s Gaudi GPUs were marketed as AI accelerators with promising specs and competitive pricing. Despite that, Nvidia’s dominance in AI training with its Hopper and Blackwell chips led to disappointing sales. This, in turn, caused Intel to cancel its successor, Falcon Shores.

With Crescent Island, Intel is taking a more strategic approach. Rather than directly challenging Nvidia and AMD in the training market, it is aiming to capture a share of the data centres that handle AI inference. These systems are responsible for running pre-trained models, processing tasks and queries from users. This shift in focus has allowed Crescent Island to be designed without the need for large amounts of expensive High Bandwidth Memory (HBM), which is a major cost driver in AI chips.

Instead, Crescent Island GPUs can be air-cooled, eliminating the need for complex and costly liquid cooling systems. If the AI chip market shifts significantly towards inference, this approach could help ease the global memory crisis, as Crescent Island uses cheaper LPDDR5X memory.

Nvidia is also making moves in the inference space, having partnered with Groq (not Grok) to develop a new chip that combines a language accelerator with its Rubin platform. This indicates that the race for dominance in the inference market is heating up.

The big question now is: who will reach the market first? None of the major chip manufacturers — Intel, Nvidia, or AMD — have given official launch dates for their AI inference products. Intel has only hinted that Crescent Island will be available later this year.

If you’ve been waiting for a resolution to the RAMpocalypse, it seems that the relief may still be a while away. The AI chip market is evolving rapidly, and the competition for dominance in inference is just beginning.