Intel’s AI Strategy: Shifting Gears to Xeon CPUs
Chip giant Intel (INTC) has faced challenges in the competitive AI accelerator market, where Nvidia currently dominates. Intel’s Gaudi family of AI accelerators initially showed promise, with competitive pricing, but sales were hampered by an underdeveloped software ecosystem. After missing its 2024 AI chip sales targets, Intel has adjusted its approach.
Falcon Shores, intended as the successor to Gaudi 3, has been canceled as a commercial product. Intel is now prioritizing rack-scale AI solutions, which are not expected to be ready until 2026. Despite being a minimal presence in the AI accelerator field, Intel’s CPU business could benefit from this shift. As the AI industry matures and workloads change from training AI models to running or ‘inferencing’ these models, Intel’s Xeon server CPUs are presented as a strong asset.
Cost-Effective AI: A Potential Advantage for Intel
The recent news that Chinese start-up DeepSeek developed an AI model comparable to the best U.S. models at a significantly lower training and operational cost could prove beneficial for Intel. While training advanced AI models demands potent accelerators like those from Nvidia, this isn’t necessarily the rule for inference—the actual running of these models. More compact, less complex AI models can already be run effectively on CPUs, especially those with built-in AI acceleration capabilities.
Intel unveiled new members of its Xeon 6 family of server CPUs on Monday, expanding the lineup to cater to lower price points and specialized uses. The Xeon 6500 and 6700 series chips are tailored for data centers. These server CPUs are substantially more efficient and dense than Intel’s previous generation, with the company claiming customers can achieve up to 68% lower cost of ownership compared to systems that are five years old. While these chips can be used with AI accelerators in AI training clusters, Intel also highlighted that they deliver up to 50% greater AI inference performance when compared to AMD’s most recent server CPUs.
Intel also introduced its Xeon 6 for network and edge chips, designed for radio access networks and other edge computing applications. These chips are up to 70% more energy-efficient than their predecessors and incorporate AI capabilities. Intel noted that a 38-core video edge server system can perform AI inference on 38 simultaneous camera streams.
Not every AI use case requires the most advanced AI model. As more capable AI becomes available in smaller, more affordable models, CPUs with integrated AI acceleration technology can become a crucial part of any company’s AI strategy.
A Lucrative Market Opportunity
As the AI industry advances, the initial rush to deploy AI solutions will be followed by realistic assessments of returns on investment. Intel’s original strategy to focus on cost efficiency with its AI accelerators failed, likely due to software issues; however, a similar approach might succeed with the CPUs. In applications involving small, fine-tuned AI models that can be run without expensive AI accelerators, servers loaded with Intel’s latest Xeon 6 CPUs could offer the most cost-effective solution.
Even in cases where AI accelerators are necessary, Intel’s CPUs can still offer value. IDC forecasts that total annual spending on machine learning and analytics will reach $361 billion by 2027, with $153 billion allocated to generative AI spending. As capable AI models become more affordable and efficient, a growing portion of this spending could go to infrastructure without high-end AI accelerators.
While being excluded from the AI accelerator market means that Intel’s overall AI opportunity is smaller than it might have been if Gaudi had been successful, the company is still a player in the AI landscape.
