The firm’s first quarter 2025 results saw data centre revenue dip slightly quarter-on-quarter by 5%, but surge 57% year-on-year, maintaining momentum after a record $3.9 billion in Q4 2024.
Looking to capitalise on strong GPU demand and Nvidia’s stumbles with Blackwell delays, AMD has pulled forward the launch of its next-gen MI350 series to mid-2025.
AMD claims the MI350, unveiled last June during Computex, delivers up to 35 times the inference performance of its predecessor and is expected to be a key growth catalyst in the second half of the year.
“We’re right on track for that launching midyear,” CEO Dr. Lisa Su told investors. “I would say customer interest has been very high.”
She added that Q1 performance of its data centre GPU offerings was “maybe a little bit better than expected,” having projected double-digit growth for the segment for 2025.
The company’s Instinct accelerators are now deployed across more than a dozen cloud providers, with Meta, Microsoft, and IBM among the early adopters. The earnings call suggested that a major AI foundation model developer is now using its MI325X line to handle daily inference workloads.
Su confirmed that AMD has “built some inventory” to support what she described as “very strong” client and server ramp.
“The lead time is really long to build for the Q3, Q4 ramp. We really need to start wafers right now. That’s why the inventory has increased,” Su added.
AMD also closed its acquisition of ZT Systems in Q1 to further bolster its ability to deliver rack-level AI solutions and co-design platforms based on upcoming Instinct GPUs.
Beyond GPUs, AMD’s EPYC server CPUs are playing a central role in its data centre expansion.
The fifth-gen Turin chips, revealed last October, are seeing ever-increasing deployments ramping across hyperscalers and large enterprises.
The firm said that its data centre revenue was driven by a gain in server CPU share amid increased demand for its EPYC line.
“Hyperscaler demand remained strong as cloud providers expanded EPYC deployments to power critical infrastructure and public services,” Su told investors.
On the software side, AMD continues to push ROCm as an open alternative to Nvidia’s CUDA, with recent updates improving inferencing speeds by nearly threefold.
Not all of AMD’s business segments are firing, however.
Embedded revenue fell 3% year-on-year to $823 million, with AMD citing a “mixed” demand environment and sluggish recovery in industrial and communications markets.
Its embedded arm continues to underperform relative to the rest of the business, dragging on overall growth.
AMD’s Q2 outlook was also shaped by new US export controls, which led to a one-time $800 million inventory-related charge, temporarily dragging non-GAAP gross margin guidance down to 43% — a notable dip from the typical 54% baseline.
Su told investors: “Despite the dynamic macro and regulatory environment, our first quarter results and second quarter outlook highlight the strength of our differentiated product portfolio and consistent execution, positioning us well for strong growth in 2025.”
RELATED STORIES
AMD completes £4.9bn ZT Systems acquisition to boost data centre solutions
AMD unveils Embedded EPYC CPUs to tackle AI network traffic demands
Jio, Cisco, Nokia, AMD launch AI platform to transform telecoms
