Intel | Fundamental Analysis

Nvidia currently dominates the AI accelerator market, boasting a market share estimated to exceed 90%. Despite the high price of its AI chips, customers are willing to invest to stay at the forefront of the AI revolution.

Intel is set to launch the third generation of its Gaudi AI accelerators later this year. Although the company had a $2 billion backlog for AI chips a few months ago, it expects only around $500 million in Gaudi 3 sales for 2024, possibly due to supply constraints.

Nvidia's strong lead and the mature software ecosystem built around its AI chips present a significant challenge for competitors. To counter this, Intel is collaborating with industry partners to develop a standardized alternative to Nvidia's proprietary CUDA platform, though this will require time. Additionally, Intel aims to offer substantial cost savings compared to Nvidia's products as part of its AI strategy.

At Computex this week, Intel announced several AI-related developments, including updates on its upcoming Lunar Lake laptop chips and its new lineup of server CPUs. The company also detailed the pricing and availability of its Gaudi AI accelerators.

Intel is expanding its network of system providers to include Asus, Foxconn, Gigabyte, Inventec, Quanta, and Wistron, in addition to existing partners Dell, Hewlett Packard Enterprise, Lenovo, and Supermicro. This expansion is expected to increase the availability of Gaudi 3 systems later this year.

Intel also introduced competitive pricing for system providers to highlight the value of its Gaudi chips. An AI kit with eight Gaudi 2 AI chips and a universal baseboard will be priced at $65,000, while a kit with eight Gaudi 3 AI chips will cost $125,000. Intel estimates these kits are priced at one-third and two-thirds, respectively, of the cost of comparable competitive platforms.

While Intel is underpricing Nvidia, it also expects its chips to deliver robust performance. Intel claims that a cluster of 8,192 Gaudi 3 chips can train an AI model up to 40% faster than an equivalent cluster of Nvidia H100 chips. Moreover, Gaudi 3 is reported to offer up to twice the AI inferencing performance of the H100 when running popular open-source models.

Nvidia's H100 will soon be replaced by the Blackwell GPUs, which promise significant improvements in performance and efficiency. Although Gaudi 3 may not match these new GPUs in raw performance, Intel's pricing strategy is designed to provide a strong value proposition.

The year 2025 will be crucial for Intel's AI accelerator business as the company is expected to better manage supply issues and ramp up Gaudi 3 production. With the ongoing demand for AI chips, Intel could see a substantial increase in Gaudi chip sales as customers look for cost-effective alternatives to Nvidia's leading products.

While Nvidia is expected to maintain its dominance in the AI accelerator market, Intel has the potential to gain significant market share if its aggressive pricing strategy appeals to prospective customers.