Intel Q2 Preview: Gaudi 3 AI Accelerator Availability for AI Inferencing

Intel is set to release its Q2 preview, with a focus on the Gaudi 3 AI accelerator for AI inferencing. The availability of this new technology is key for the company's success in the AI server and data center market. Intel has already unveiled several new graphics cards at Computex, and the Q2 preview is expected to showcase its capabilities.

Intel is set to release its Q2 preview, with a significant focus on the Gaudi 3 AI accelerator for AI inferencing. This new technology is crucial for Intel's success in the AI server and data center market. The company has already unveiled several new graphics cards at Computex, and the Q2 preview is expected to showcase its capabilities.

Intel's Gaudi 3 AI accelerator is designed to handle AI inference workloads, which involve using pre-trained AI models. This is a growing market, with Cloudflare predicting that inference could eventually be larger than training. The Gaudi 3 features 8x HBM2E with 128GB capacity, which is superior to Nvidia's H100 Tensor core GPU architecture in terms of memory capacity [2]. While Intel's AI accelerator is still lagging behind Nvidia's H200 and B200, or AMD's MI300 in overall performance, the company is making significant progress in its GPU technology.

In addition to the Gaudi 3, Intel has also announced the ARC B50 and ARC Pro B60 graphics cards, which are designed for workstations and AI inference, respectively. Both new graphics cards will be available in Q3 FY25. Intel is partnering with several industry players, including Adobe, Ansys, Autodesk, Bentley, and Siemens, to build the ecosystem for its ARC chips.

Intel's Q2 preview is expected to provide more details about the company's progress in its 18A process node. The management is expected to disclose more information about the ongoing progress of this technology, which could accelerate the adoption of Panther Lake CPUs.

The company is also facing several challenges, including a shortage of capacity for its Intel 7 process node due to surging demand for 7nm chips. Additionally, trade uncertainties and geopolitical tensions between the U.S. and China are expected to pose growth challenges for Intel's consumer-related chips and foundry business.

Despite these challenges, Intel's new management has done a great job controlling its costs, lowering its full-year operating expenses to $17 billion in FY25 and anticipating $16 billion in FY26. The company is expected to deliver flat revenue growth for FY25, assuming its 18A will start full production later this year. From FY26 onwards, Intel anticipates a 6% organic revenue growth, with Client Computing growing by 5%, Data Center and AI growing at double-digit rates, and Foundry adding an additional 2% to the topline growth.

Intel's shares repurchase program has been halted since FY22, and its total shares outstanding have grown by 2% annually due to stock-based compensations. The company is expected to continue growing its shares outstanding by 2% annually.

References:

[1] https://www.theglobeandmail.com/investing/markets/stocks/AMD-Q/pressreleases/33436569/intel-might-be-quitting-the-ai-training-market-for-good/
[2] https://seekingalpha.com/article/4801670-intel-q2-preview-availability-of-gaudi-3-ai-accelerator-for-ai-inferencing-is-key