On November 3, 2026 (U.S. time), AMD announced results for the third quarter of fiscal 2026, which ended September 26. Revenue rose 45% year over year to a record $13.377 billion. The data center business drove the growth, and its revenue increase alone exceeded the increase for the company as a whole.
According to AMD's earnings release, both the server business, fueled by strong demand for 5th Gen EPYC processors, and the data center AI business, which launched the AI GPU Instinct MI450 series, posted record revenue. The company expects adoption to broaden in the fourth quarter for "Helios," an AI rack built around the 6th Gen EPYC "Venice" and the Instinct MI455X. Meanwhile, a large inventory build in preparation for data center demand is already putting pressure on cash flow.
Data center growth exceeded the company-wide increase
Data center revenue nearly doubled, from $4.341 billion a year earlier to $8.546 billion, an increase of $4.205 billion. That exceeds the company-wide increase of $4.131 billion, meaning that revenue from all other businesses combined declined.
| Segment / business (US$ millions) | Q3 FY2025 | Q3 FY2026 |
|---|---|---|
| Data Center | 4,341 | 8,546 |
| Client | 2,750 | 2,949 |
| Gaming | 1,298 | 773 |
| Embedded | 857 | 1,109 |
| Total | 9,246 | 13,377 |
Client and Gaming belong to the same reporting segment. Client revenue rose 7% year over year, helped by strong sales of Ryzen for notebook PCs, and Embedded revenue grew 29%. Gaming revenue, however, fell 40% on lower sales of semi-custom chips for game consoles and graphics products. As a result, revenue outside the data center business fell by a combined $74 million.
The data center business's share of total revenue rose from about 47% a year earlier to about 64%. According to AMD's earnings presentation, the segment's operating income increased from $1.074 billion to $2.882 billion, and its operating margin improved from 25% to 34%. Company-wide GAAP operating income rose 123% year over year to $2.832 billion. AMD said it recorded record revenue for both EPYC and Instinct, but it did not disclose the revenue increase for CPUs and GPUs separately.
The spread of AI agents is boosting EPYC demand
Growth in the data center business is not driven by AI GPUs alone. In systems that use AI agents, the CPU handles more work in addition to the model inference performed by GPUs, including data retrieval, execution of external tools, and integration with business applications. In environments that run multiple agents in parallel, the capacity for this surrounding processing affects overall system performance, which is raising the importance of server CPUs. Oracle's cloud service "OCI Compute E7," which uses next-generation EPYC, is likewise intended for both the services that support AI models and general enterprise applications.
EPYC is also increasingly used as the host CPU that controls AI accelerators. According to AMD, OpenAI adopted the 5th Gen EPYC "Turin" for its first in-house AI accelerator infrastructure. In the strategic partnership with Meta, EPYC will power the dedicated computing environment for Meta's personal AI agent "Muse." Host CPUs that manage GPUs and CPUs that run the peripheral processing of AI agents play different roles, but both are areas that lead to growing EPYC demand.
According to AMD, third-quarter server revenue more than doubled from a year earlier for both cloud and enterprise customers. Cloud instances based on Turin now number more than 540 across major providers. Shipments of the next-generation Venice, which generated revenue, began in the third quarter; initial servers are planned for the fourth quarter, and cloud instance rollouts for early 2027. However, it is unclear how much AI-agent-related demand contributed to the overall growth in EPYC revenue.
Instinct MI455X enters volume production; Helios at early deployment stage
In the third quarter, AMD began volume shipments of the Instinct MI455X AI GPU. According to Oracle, the Helios AI rack holds 72 MI455X GPUs, placing four in each of 18 compute trays. It is designed to handle the training and inference of large AI models at the rack level, combining Venice processors, Pensando networking equipment, and ROCm, the software platform for GPUs. AMD has also prepared "Helion," a reference design for AI agent racks that allows customers to choose between CPU-only configurations and configurations that combine GPUs.
On the earnings call, it was revealed that early customers such as Meta and OpenAI have run AI models and other workloads on initial Helios racks. Meta had installed initial racks as of October 27 and was carrying out validation ahead of full-scale deployment. This is separate from the EPYC adoption for Muse.
AMD expects partners such as HPE and Dell to begin shipping Helios systems in the fourth quarter. However, volume shipments of the GPU itself, initial operation by early customers, and system shipments from server makers are distinct stages. In assessing future growth, the shipment scale of completed rack systems will matter as much as the track record of early deployments.
Q4 revenue guided to $17 billion; inventory build strains cash flow
AMD forecasts fourth-quarter revenue of $17 billion, plus or minus 3%. At the midpoint, that represents growth of about 66% year over year and about 27% quarter over quarter. By segment, data center is expected to grow about 50% sequentially and Embedded by double digits, while Client and Gaming are expected to decline by more than 20%. Client revenue also fell 4% sequentially in the third quarter, and AMD explained that rising memory prices weighed on PC demand.
While it anticipates a sharp rise in revenue, AMD's non-GAAP gross margin guidance of about 55% is below the third-quarter actual of 56%. Even if the expansion of AI-related products lifts revenue, margins will not necessarily improve at the same pace.
The impact of upfront investment to meet growing demand is also visible in cash flow. Third-quarter free cash flow was $554 million, a sharp drop from $1.558 billion in the previous quarter. This figure is operating cash flow from continuing operations of $1.202 billion less capital expenditures of $648 million.
According to Chief Financial Officer Jean Hu, AMD added about $3 billion to inventory from the end of the previous quarter, bringing it to about $11.5 billion, in preparation for data center demand. In addition to the funds directed to inventory, capital spending to secure supply capacity pushed down free cash flow, she said. The inventory increase is a move to prepare for expanded shipments, but its effect will become clear only as actual sales and cash collection progress.
AMD expects server revenue to roughly double year over year in 2027, with the data center business accounting for more than 80% of total revenue. The first important milestone toward that goal is the fourth quarter of fiscal 2026. How far expanded Helios shipments are reflected in actual revenue, and how the built-up inventory translates into sales and cash flow, will be key indicators for judging future growth.