AMD’s AI Chip Challenge To Nvidia Still An Uphill Battle Goes On

(Reuters) – Advanced Micro Devices has revealed new details about an artificial intelligence chip that will take on market leader Nvidia, but the company wanted to know Wall Street and who plans to buy it. I omitted whether.

Santa Clara, Calif.-based AMD said the next chip will ship in small chunks in the third quarter, followed by mass production in the fourth quarter, with 192 gigabytes of memory.

AMD Chief Executive Lisa Su said in an interview with Reuters that this could allow tech companies to manage the skyrocketing costs of offering a service similar to ChatGPT. Speaking following her keynote address in San Francisco, Su showed an AI system on the MI300X chip writing a poem about the city.

“The more memory, the bigger set of models the chip can handle,” Su said. “We’ve seen it run much faster on customer workloads. We really think this is a differentiator.”

However, unlike previous presentations where AMD discussed key customers for the new chip, AMD didn’t reveal who would adopt the smaller versions, dubbed MI300X and MI300A. The company didn’t provide details on how much the chip would cost or how it would boost sales.

AMD shares have more than doubled since the beginning of the year, hitting a 16-month high on Tuesday, but closed 3.6% after the company’s presentation on its AI strategy. Nvidia shares rose 3.9% to close at $410.22, making it the first chipmaker to close with a market capitalization above $1 trillion.

Kevin Crewell, principal analyst at TIRIAS Research, said, “[Big customers]may have disappointed Street by not stating they would use the MI300 A or X. They told AMD that some I want them to say they’ve been superseded by Nvidia in design.”

Nvidia’s stock has soared 170% so far this year and dominates the AI ​​computing market with 80% to 95% market share, analysts said.

Nvidia has few competitors working on a large scale. Intel Corp and several start-ups such as Cerebras Systems and SambaNova Systems have competing products, but Nvidia’s biggest revenue threat so far is Alphabet Inc’s internal growth in Google and Amazon.com’s cloud divisions. It’s a chip initiative, with the companies renting out custom chips to outside developers.

Aside from the AI ​​market, AMD announced that it has begun mass-shipping a general-purpose central processor chip called “Bergamo” to companies such as Metaplatform.

Alexis Black Bjorlin, who oversees computing infrastructure at Facebook’s parent company Meta, said the company chose Bergamo chips as part of AMD’s data center business serving cloud computing providers and other large chip buyers. He said he was targeting parts of

But investors were looking for news about AI. Nvidia is leading there not only with its own chips, but for more than a decade it has provided AI researchers with software tools to predict what they will need for chips that take years to design. It comes from what you have learned.

AMD provided an update to its Rocm software on Tuesday, which competes with Nvidia’s Cuda software platform.

In a presentation, Soumis Chintara, vice president of Meta, which has contributed to the development of open source software for artificial intelligence, said in a presentation that AI developers can easily get free tools from the “single dominant vendor” of AI. It said it has been working closely with AMD to make it available. They provide their chips to other products such as AMD.

“You don’t really have to do a lot, or often very little, to move from one platform to another,” Chintala said.

But analysts say just because sophisticated companies like Meta can extract superior speeds from AMD chips doesn’t promise widespread market traction by unsophisticated buyers. said.

“People are still convinced that AMD’s software solutions will be competitive with NVIDIA’s, even if they are competitive in terms of hardware performance,” said Moor Insights & Strategy analyst Anshel Sag. No,” he said.