AMD has been fighting to increase its share of the market for sophisticated AI processors sold mostly by Nvidia to heavyweight tech customers including Amazon and Microsoft.
AMD reported fourth-quarter data centre revenue of $3.9 billion, which missed the consensus estimate of $4.15 billion. The company’s data centre segment is a proxy for its AI revenue as it contains revenue from its line of processors that compete with Nvidia’s chips.
For 2024, AMD said it generated more than $5 billion of AI chip revenue.
During a conference call with analysts, AMD Chief Executive Lisa Su said the company’s data centre sales in the current quarter will be down about 7% from the just-ended quarter, in line with an overall expected decline in AMD’s revenue.
Discover the stories of your interest
Su declined to give a forecast for the company’s AI chips, but said AMD expects “tens of billions” of dollars in sales “in the next couple of years.” Summit Insight analyst Kinngai Chan said: “AMD’s AI GPU is probably not tracking to investors’ expectations. We continue to believe Nvidia is opening a gap against AMD in AI GPU performance and value.”
‘Bespoke’ AI chips
AMD’s forecast comes as tech giants including Microsoft and Meta increasingly dedicate resources to developing their own silicon for processing large data volumes required by GenAI.
Microsoft and Meta unveiled new AI chip versions last year, casting doubt on AMD’s ability to compete in a race to dominate AI infrastructure.
Marvell and Broadcom help other companies design their own custom chips, contrary to off-the-shelf processors provided by AMD and Nvidia and have contracts with various cloud computing companies. Su told analysts that AMD is working to compete against Broadcom and Marvell in collaborating with its customers to create custom AI chips. Reuters last year reported that Nvidia has also entered that business.
“We are very involved in a number of (custom chip) discussions with our customers as well,” Su said. “They’d like to take our baseline (intellectual property) and really innovate on top of that.”
Despite AMD’s efforts, Nvidia maintains nearly an 80% share of the AI chip market, in part because of its proprietary CUDA software, which is a standard among some developers. Computer code written using the CUDA language does not easily run on AMD hardware. Customers have to incur additional, high costs to switch chip providers.
The company expects revenue of about $7.1 billion, plus or minus $300 million in the first quarter, compared with analysts’ average estimate of $6.99 billion, according to data compiled by LSEG.
AMD – one of the largest providers of personal computer chips – is also likely to benefit from improving demand from a wave of consumers and businesses buying new PCs that can handle generative AI tasks, after a prolonged slump.
Adjusted for stock-based compensation, among other things, AMD reported fourth-quarter per-share profit of $1.09, compared with a consensus estimate of $1.08. AMD reported revenue jumped 24% to $7.66 billion for the fourth quarter, compared with estimates of $7.53 billion.