Intel Corp is rolling out a new version of its artificial intelligence chip, aiming to challenge Nvidia Corp. in one of the fastest-growing parts of the semiconductor industry.
The updated processor, called Gaudi 3, will be widely available in the third quarter, Intel said at a company event Tuesday. The chip is designed to boost performance in two key areas: helping train AI systems — a process that involves bombarding them with data — and running the finished software.
Intel CEO Patrick Gelsinger is looking to a new version of its Gaudi chip lineup to help the company catch up with Nvidia.
Booming demand for AI services has sent tech companies scrambling for these so-called accelerator chips, but Nvidia has seen most of the benefit. Earlier versions of Gaudi failed to achieve the market share gains that Intel had been hoping for, Chief Executive Officer Pat Gelsinger has said. He expects the new model to have a bigger impact.
“They’re a good competitor, but people want an alternative,” he said Tuesday of Nvidia. “The world needs more suppliers, and we’re quite dedicated to providing that choice.”
Intel shares gained were up less than 1% at US$38 as of 1:33 p.m. in New York on Tuesday, reversing an earlier decline. They had been down 24% this year through Monday’s close.
Gelsinger declined to give pricing but said that his chips would be “a lot below” the cost of Nvidia’s current and future chips. They’ll provide an “extremely good” total cost of ownership, he said.
Challenging Nvidia won’t be easy. The runaway success of that company’s H100 accelerator helped more than double revenue and sent its market valuation over US$2 trillion. And now Nvidia is looking to build on its lead with a just-announced chip platform called Blackwell. Systems based on that product will be available later this year, the company said in March.
According to Intel’s assessment, Gaudi 3 will be faster and more power-efficient than the H100. It will train certain types of AI models 1.7 times more quickly and be 1.5 times better at running the software, the chipmaker claims. The product will be roughly equal with Nvidia’s newer H200, Intel said, performing slightly better in some areas and a bit behind in others.
See also: Anthropic CEO says mandatory safety tests needed for AI models
Intel, based in Santa Clara, California, said it can’t provide comparisons with Nvidia’s upcoming Blackwell line until those products are publicly available. Intel rival Advanced Micro Devices Inc. — its longtime competitor in personal computer processors — also is pushing into the field. It unveiled an accelerator lineup called MI300 in December.
Intel’s Gelsinger has said that he’s not just trying to catch up with Nvidia. He expects AI to bring a bigger windfall for the industry — especially as the technology spreads beyond its current concentration in the data centres of companies such as Microsoft Corp. and Alphabet Inc.’s Google. Personal computers, mobile phones and networking gear will require chips that are able to handle AI tasks and provide users with instantaneous feedback — something that’s not always possible with remote server farms.
When Intel announced its fourth-quarter results in late January, Gelsinger said he was increasing the supply of Gaudi to meet growing orders and that the company has a “pipeline” for 2024 of “above US$2 billion and growing.” The broader market for corporate spending on generative AI gear will increase from US$40 billion in 2024 to US$151 billion in 2027, Intel said, citing market research.
But that just underscores how much of a lead Nvidia has. The company had data centre revenue of more than US$47 billion during the 12 months ended in January. In its current fiscal year, that total will top US$95 billion, according to analysts’ estimates.
