AMD Data Center Revenue Nearly Doubles as MI450 Ramps | Zeniteq
Loading
Loading article
Zeniteq is now available on mobile for iPhone, iPad and Apple Watch.Download nowLearn more
AMD Data Center Revenue Nearly Doubles as MI450 Ramps
AMD’s record $8.5 billion data-center quarter reflects growth in both AI accelerators and server CPUs, with Helios racks and Venice processors central to its next production push.
AMD reported record data-center revenue of $8.5 billion for the third quarter of 2026, a 97% year-over-year increase driven by strong demand for EPYC processors and the launch of its latest AI accelerators.
The company confirmed that volume shipments of the Instinct MI455X GPU began during the quarter, alongside the initial deployment of Helios rack-scale systems by customers including Meta, Microsoft, and OpenAI. CEO Lisa Su projects growth to accelerate in the fourth quarter as production scales for both Helios and new Venice server processors.
AMD reported record data-center revenue of $8.5 billion for the third quarter of 2026, up 97% year over year, as EPYC server processor demand and the launch of Instinct MI450-series GPUs lifted its largest business. Total company revenue reached $13.4 billion, up 45%, according to AMD’s November 3 earnings announcement.
The product detail matters: AMD’s prepared remarks identify MI455X as the GPU that began shipping in volume during the quarter. MI450 is the broader series name. The same remarks say major customers brought their first Helios racks online, moving the story beyond a chip launch to operating AI infrastructure.
CEO Lisa Su expects growth to accelerate further in the fourth quarter as Helios and Venice production scales. That is AMD’s outlook, separate from its reported third-quarter results.
Data Center Now Accounts for Nearly Two-Thirds of AMD Sales
AMD’s segment figures show how heavily its growth now depends on infrastructure computing. Data Center contributed 64% of total revenue, up from 47% a year earlier, according to the company’s prepared earnings remarks.
A calculation makes the shift clearer. Dividing the rounded $8.5 billion data-center figure by 1.97 implies approximately $4.3 billion in the year-earlier quarter, or an increase of roughly $4.2 billion. AMD’s companywide revenue increased by about $4.1 billion over the same period.
Those rounded figures indicate that , with changes elsewhere broadly offsetting one another. Embedded grew, while the combined Client and Gaming segment declined.
Data Center accounted for roughly all of AMD’s net revenue growth
The $8.5 billion figure covers both EPYC CPUs and Instinct GPUs. AMD has not supplied a separate MI450-series revenue figure in these materials. Su also said cloud and enterprise server sales each more than doubled year over year, making the CPU contribution substantial enough to warrant attention alongside the accelerator launch.
That distinction changes how I’d read the quarter. AMD is benefiting from multiple parts of infrastructure spending: GPUs for model computation, plus CPUs supporting cloud services, enterprise applications and the surrounding work involved in running AI systems. Treating the entire segment as accelerator revenue would obscure that broader demand.
Profitability improved too. AMD reported GAAP operating income of $2.832 billion and net income of $2.428 billion. Its non-GAAP figures were higher, at $3.923 billion and $3.471 billion respectively. The financial tables show GAAP operating margin reaching 21%, up from 14% a year earlier. These accounting measures should remain distinct when comparing earnings growth.
MI450 Is the Family; MI455X Defines the Reported Ramp
AMD’s segment summary credits the launch of Instinct MI450 Series GPUs as one driver of data-center growth. Su’s prepared remarks provide the more specific shipment detail:
“We started shipping MI455X in volume in the quarter,” she said in the prepared remarks.
AMD named Anthropic, Meta, Microsoft, OpenAI and Oracle as lead customers that brought their first Helios racks online. The company said customers were running their latest models and workloads on those systems.
The wording supports a precise account of the launch: the MI450 family contributed to the quarter, MI455X entered volume shipments, and initial Helios deployments became operational. It gives no basis for assuming every MI450-series model had reached the same shipping stage.
Helios is AMD’s rack-scale solution combining MI450-series accelerators with sixth-generation EPYC CPUs. That makes it a different commercial milestone from shipping standalone processors. A rack brings multiple computing components together into infrastructure customers can deploy for workloads.
For readers following the AI industry, the useful signal is the progression from a product announcement to volume shipments and named operating deployments. AMD’s statements provide evidence of that progression, although they do not disclose shipment quantities or customer-by-customer revenue.
The financial result and deployment statements also measure different things. Revenue captures sales recognized during the quarter; a customer bringing racks online describes deployment progress. The announcement does not provide a conversion between the two.
Helios and Venice Give the Q4 Plan Two Production Tracks
Su linked expected fourth-quarter acceleration to scaling Helios and Venice production. The documents describe two related product ramps: rack-scale AI systems and a new generation of server CPUs.
AMD’s published timeline separates the completed milestones from the planned ones:
Helios: AMD reported initial customer racks operating in Q3 and expects deployments to increase substantially in Q4 as production expands.
Venice revenue shipments: AMD said shipments of the new EPYC processors began in Q3.
Venice OEM systems: The first systems from original equipment manufacturers were on track for Q4 availability.
Venice cloud instances: AMD said cloud providers were deploying processors in Q4 ahead of planned public instance rollouts in early 2027.
These stages are described in the November 3 prepared remarks. The distinction is practical: a processor can generate revenue before an OEM system becomes available, and a cloud provider can install hardware before customers can rent an instance built around it.
AMD said its OEM and original design manufacturer partners had committed more than 240 Venice platforms, 20% more than for Turin, the previous generation. That is a measure of planned platform breadth, distinct from the number of servers shipped or customer purchases.
Su characterized early Venice demand as the strongest AMD had seen for a new EPYC generation. That remains management’s assessment, without a disclosed order tally that would allow readers to quantify the comparison independently.
My view is that the staggered rollout is the most useful detail for infrastructure buyers. A company evaluating Venice needs to distinguish processor availability, a supported server configuration and availability through its preferred cloud provider. Those dates can differ by a quarter or more within AMD’s own schedule.
The production plans also leave an important question open. AMD’s expectation of faster growth depends on converting production capacity into shipments and deployments. The published remarks do not quantify how much of the expected increase comes from additional system volume, product mix or other factors. They also do not identify a single production bottleneck.
What the Results Say About Competition With NVIDIA
AMD’s quarter offers a concrete competitive signal: its infrastructure business is growing rapidly while a new accelerator generation reaches volume shipments and named customers operate rack-scale systems.
That strengthens the case for taking AMD seriously in AI infrastructure procurement. It does not provide a like-for-like measurement against NVIDIA’s accelerators. A GPU market-share comparison would need AMD’s accelerator-only revenue and comparable NVIDIA results over aligned reporting periods.
Performance and economics require another set of evidence. The earnings materials discussed here do not supply matched workload measurements covering model quality, throughput, latency, power consumption and total system cost. Revenue growth cannot answer which platform offers the best economics for a particular deployment.
I’d put the most weight on the combination of recognized sales and operational milestones. The $8.5 billion segment result establishes the scale of AMD’s data-center business; MI455X volume shipments and initial Helios deployments explain part of the product transition behind it. Fourth-quarter reporting will show how much of the planned Helios and Venice expansion becomes additional revenue.
Frequently Asked Questions
3 questions
1
How Much Did AMD’s Data Center Revenue Grow?
AMD reported $8.5 billion in Q3 2026 data-center revenue, up 97% year over year. The total includes EPYC server CPUs and Instinct GPUs.
2
What Is the Difference Between MI450 and MI455X?
MI450 is the GPU series name used in AMD’s earnings release. MI455X is the specific model AMD said began shipping in volume during Q3.
3
Are Helios and Venice Growth Figures Completed Results?
AMD reported initial Helios deployments and Venice revenue shipments in Q3. Further production scaling, wider system availability and faster Q4 growth were company expectations when the results were announced.
