How Intel Can Still Succeed in the AI Market

Chip giant Intel (INTC) has faced challenges in entering the growing AI accelerator market, which is currently dominated by Nvidia. The company introduced the Gaudi family of AI accelerators, which showed promise and were priced competitively, but their sales were hindered by an immature software ecosystem. Intel recently adjusted its strategy, cancelling its Falcon Shores product, which was set to follow Gaudi 3, and is now turning its focus towards rack-scale AI solutions that are not expected to be available until 2026.

Although Intel is largely absent from the AI accelerator landscape, its CPU business could step in to fill some of the gaps. As the AI industry evolves, shifting from training models to executing them, Intel's Xeon server CPUs can serve as a powerful asset in this space.

Cost-effective AI Solutions

The success of a Chinese start-up called DeepSeek, which trained an AI model comparable to those created by top U.S. firms at a significantly lower cost, may work in Intel's favor. While advanced AI models generally require powerful accelerators like those made by Nvidia, running these models, referred to as inference, does not need the same level of computational power. Less complex AI models can already be effectively deployed using CPUs, especially those with integrated AI acceleration capabilities.

In a recent announcement, Intel expanded its Xeon 6 family of server CPUs, which now includes lower-priced options and specialized chips aimed at data centers. The Xeon 6500 and 6700 series claim to deliver up to a 68% reduction in total cost of ownership when compared to systems that are five years old. While these chips can be used in combination with AI accelerators, they also offer up to 50% better AI inference performance than the latest server CPUs from competitor AMD.

Furthermore, Intel recently introduced its Xeon 6 for network and edge computing. These chips, targeting applications such as radio access networks, are up to 70% more power efficient than previous generations and can handle AI tasks effectively. For instance, a video edge server with 38 cores is capable of performing AI inference on 38 camera streams at once.

It's important to note that not all AI applications require the latest, most advanced AI models trained on extensive datasets. As organizations adapt to using smaller, more affordable AI models, CPUs with integrated AI capabilities could play a key role in their AI strategies.

A Promising Market Ahead

As the AI sector matures, there will be a shift from hurriedly deploying AI solutions to evaluating returns on investment. Although Intel's effort to prioritize cost efficiency with its AI accelerators fell short due to software issues, a similar approach could be advantageous for its CPUs.

For applications that utilize smaller, finely-tuned AI models, servers powered by Intel's Xeon 6 CPUs may be the most economical solution. In scenarios where AI accelerators are still needed, Intel's CPUs can still contribute effectively.

According to IDC, machine learning and analytics spending is projected to reach $361 billion annually by 2027, with $153 billion anticipated for generative AI. As AI models become easier to run and manage efficiently, a larger portion of this budget could be allocated to infrastructure without relying on high-end AI accelerators.

Although Intel's exclusion from the AI accelerator market limits its overall opportunities in this realm, it still possesses valuable assets to remain competitive in the AI industry.

Intel, AI, Market