It’s no secret that Intel has been struggling on the AI front—its laggard status helped to fell CEO Pat Gelsinger two months ago—but the company made the extent of the problem crystal-clear on Thursday.
Intel was set to release a new AI chip called Falcon Shores late this year, to replace its Gaudi 3 accelerator chip. But interim co-CEO Michelle Johnston Holthaus said on an earnings call that this would no longer be happening, meaning customers will now need to wait for a later version called Jaguar Shores.
“Based on industry feedback, we have decided to leverage Falcon Shores as an internal test chip,” Holthaus told analysts. Instead of releasing Jaguar Shores, she said Intel would use the chip internally as way to figure out how to build better AI data-center products, including entire racks of AI-specific chips that are needed to run many AI applications in the cloud.
So how’s Gaudi doing in the meantime? Badly. “We’re not yet participating in the cloud-based AI data-center market in a meaningful way,” Holthaus admitted. “We have learned a lot as we have ramped Gaudi, and we are applying those learnings going forward.”
Intel also revealed that the upcoming Clearwater Forest generation of its Xeon data-center CPUs would only come out next year, rather than this year as the road map previously held.
Remarkably, investors seemed to shrug off Thursday’s bad news, which suggests that their expectations are already very low. Intel’s fourth-quarter revenue was down 7% year on year to $14.26 billion, but it still managed to beat analyst estimates, and the company’s share price rose by over 3% in aftermarket trading.
Intel’s legacy strength is in traditional central processing units (CPUs), but lately, much of the world’s attention, and a significant portion of the industry’s growth, has been focused on Nvidia’s AI-market-dominating graphics processing units (GPUs). These specialized chips are much better suited to the parallel processing that AI and other modern high-performance-computing applications require than CPUs. Falcon Shores and Jaguar Shores are GPUs.
Falcon Shores isn’t the first Intel GPU to be pulled from its road map. Indeed, it only became Intel’s great GPU hope after the company canceled its planned Rialto Bridge data-center GPU in early 2023. Before that, Falcon Bridge was supposed to be a novel CPU-GPU-memory combination called an XPU. But it became clear that the AI market wanted GPUs, so Intel’s XPU push was shelved.
Nvidia commands around 80% of the data-center AI-chip market, according to recent estimates. Intel and its big rival from the desktop CPU days, AMD, will have to do a lot of work to catch up, particularly when it comes to the developer software that is associated with their hardware—Nvidia’s CUDA software is seen as one of its great strengths.
Intel has previously conceded that hard-to-use software was something holding back its AI aspirations. “One of the things that we’ve learned from Gaudi is, it’s not enough to just deliver the silicon,” said Holthaus on Thursday.
AMD is due to release its next quarterly results on Tuesday, and all eyes will be on adoption indications for its MI300 AI accelerator.
As for the search for a new permanent CEO, Intel had little to share this week. “The board remains intensely focused on the search for a permanent CEO,” said interim co-CEO David Zinsner on the earnings call. “The search is progressing, but we have nothing new to report.”