This article is from the free weekly Barron’s Tech email newsletter. Sign up here to get it delivered directly to your inbox.
Lost Year. Hi everyone. Advanced Micro Devices is facing trouble in 2025. Double trouble, in fact.
On Tuesday, AMD reported better-than-expected fourth-quarter earnings results and gave a slightly better-than-expected outlook for revenue in its current quarter.
Still, the chip stock tumbled as much as 10% on the news. The culprits were some underlying numbers in the company’s data center business, along with weak commentary about its outlook for all-important AI chips.
AMD’s December-quarter data center revenue came in at $3.86 billion, which was below the $4.14 billion estimate. The miss suggests the strong growth for AMD’s core x86 franchise, where it has been gaining share from Intel, could be slowing.
AMD also faces a rising threat from Arm in the server CPU market. Microsoft, Amazon Web Services, and Nvidia have designed Arm-based server CPUs for cloud data centers that are increasingly taking share from the x86-based chips made by AMD and Intel.
The coming threat isn’t just on the server side. Last month, I predicted the next logical move for Nvidia would be to make an Arm-based CPU for consumer Windows laptops, which could further crimp AMD’s client PC x86 business. I believe that will happen some time this year.
But the commentary around AMD’s AI GPU business was the most concerning for investors. CEO Lisa Su suggested the company’s AI GPU revenue for the first half of 2025 would be flat compared with the second half of 2024, disappointing Wall Street, which was hoping for a stronger secular growth ramp. Su also said AMD would stop giving explicit numerical guidance for AI GPU sales, which also worried analysts. Less transparency is rarely a positive sign.
AMD lacks Nvidia’s ability to deliver an end-to-end data center solution. It won’t have a competitive offering against Nvidia’s Blackwell GB200 NVL72 this year. Nvidia’s AI server incorporates 72 GPUs all linked together inside one server rack that’s roughly two feet wide, four feet deep, and seven feet tall. These super clusters of networked GPU servers, which can power increasingly complex AI workloads, is where the demand is, and AMD is far behind Nvidia.
AMD isn’t alone in its predicament. Intel said last week that it wouldn’t be bringing its next GPU called Falcon Shores to market, instead using it for internal testing, while working on a data center AI system named Jaguar Shores to eventually compete with Nvidia.
Nvidia’s current advantage stems from its 2019 acquisition of networking leader Mellanox. At the time of the deal, CEO Jensen Huang presciently said that AI workloads in the future would run on tens of thousands of connected GPUs at data center scale with hardware, software, and networking all designed to work together. That’s exactly what is happening today at Microsoft, Google, Amazon, Meta, and Elon Musk’s xAI.
Nvidia managed to see the future several years ago. It will take a similar amount of time for AMD and Intel to catch up.