Demand
AMD Data Center Revenue
Reported quarterly AMD Data Center segment revenue in USD billions—actuals only, no guidance. A proxy for AI infrastructure spending converted into AMD compute, not a pure AI-revenue line. Calendar-like quarters (Q1 is January–March), unlike NVIDIA and Broadcom fiscal calendars.
AMD Data Center Revenue
- AMD
API access · Pro coming soon
As of 2026-04-01, AMD Data Center revenue was $6.72B in Q2 2026, up 107% year over year.
Methodology
Values are AMD-reported Data Center segment revenue from quarterly filings and earnings releases, converted from millions of dollars to billions. They are recognized actuals, not guidance, TAM, or management forecasts. Something generally has to be shipped or delivered and meet AMD’s revenue-recognition criteria before it appears here.
AMD’s historical segment appendix provides comparable recast figures for 2021 through Q2 2022 after segment presentation changes. Subsequent filings continue that recast series. Full-year checks: 2023 Data Center was $6.496B, including $2.282B in Q4; 2025 was $16.635B, including $4.341B in Q3 and $5.380B in Q4.
AMD defines Data Center as AI accelerators, server CPUs, GPUs, APUs, DPUs, AI NICs, FPGAs, and adaptive SoCs. Instinct accelerators (MI300X, MI325X, MI350-series) are the most AI-sensitive piece, but EPYC host CPUs, Pensando DPUs, and networking also sit in the same line. AMD does not publish a clean quarterly split of Instinct vs EPYC vs networking, so this is a proxy for AMD’s monetization of the AI data-center buildout—not “AMD AI revenue.”
QoQ and YoY growth are stored on each point because they make the AI-cycle signal clearer than the dollar level alone. The series was roughly $1.3–1.7B a quarter from 2021 through Q3 2023, then changed character after the late-2023 MI300X launch.
AMD’s fiscal year ends on the last Saturday of December, so Q1 is January–March. Labels follow AMD’s own quarters (Q1 2021, Q2 2026). NVIDIA’s year ends in late January and Broadcom’s in late October. Compare by calendar date, not by matching Q1 to Q1.
Frequently asked questions
Is this AMD AI revenue?
No. Do not say AMD had $6.7B of AI revenue in Q2 2026. The correct statement is that AMD had $6.7B of Data Center revenue, a significant portion of which is exposed to AI infrastructure demand. EPYC CPUs still serve enterprise databases, virtualization, cloud, web servers, HPC, storage, and general-purpose compute. FPGAs and networking are not necessarily AI-specific either. AMD does not split Instinct GPU revenue from EPYC or networking on a quarterly basis, so Data Center is a proxy, not a direct measurement of AI accelerator sales.
Why is this useful as an AI-growth indicator?
It is recognized revenue, not announcements, purchase intentions, theoretical TAM, or forecasts. When the series moves from about $1.3B to $2.3B to $3.9B to $5.4B to $6.7B a quarter, you are seeing money converted into shipped AMD data-center hardware. Conceptually: hyperscaler capex → data centers → servers, accelerators, CPUs, and networking → AMD revenue. That makes it a downstream check on whether enormous AI capex plans are actually turning into semiconductor demand.
How is this different from NVIDIA, Broadcom, and TSMC?
NVIDIA Data Center revenue is the best public proxy for merchant accelerated computing at the market leader. Broadcom AI semiconductor revenue captures custom XPUs and AI cluster networking. TSMC HPC is vendor-agnostic foundry production of high-end compute silicon. AMD Data Center tells you whether AI demand is large enough that a meaningful second accelerator supplier can also grow extremely rapidly, while also picking up EPYC CPUs and other AI-server attach. If NVIDIA is growing and AMD Data Center is also accelerating, the implication is stronger than NVIDIA alone: the AI compute pie is expanding rather than AMD merely stealing a fixed slice.
Why does the acceleration starting in late 2023 matter?
Before Q4 2023, quarterly Data Center revenue had been roughly $1.3–1.7B for most of two years. AMD launched MI300X into the generative-AI accelerator market in late 2023 and highlighted deployments at Microsoft, Meta, Oracle, and others. Q4 2023 jumped 43% sequentially to $2.28B. By Q2 2026 the segment was $6.72B—more than 4× Q3 2023 in less than three years. AMD says recent growth has been driven by both EPYC processors and Instinct GPUs; in Q2 2026 it attributed a 107% YoY increase to strong EPYC demand and the Instinct MI350 ramp. That is a fairly clean view of the boom moving from announcements to purchase orders to deployments to AMD revenue.
What four things is this one number a proxy for?
First, AMD AI accelerator adoption: if Instinct use at Microsoft, Meta, Oracle, OpenAI, labs, and sovereign projects expands, it eventually shows up here—and sustained outperformance versus ordinary server growth is evidence that NVIDIA’s accelerator monopoly is broadening into a multi-vendor market. Second, hyperscaler AI capex conversion: capex says companies intend to spend; AMD revenue says equipment was delivered. Third, AI compute demand outside NVIDIA: a second supplier growing this fast is a stronger signal that the overall pie is expanding. Fourth, the broader AI server ecosystem: Instinct, EPYC, Pensando DPUs, AI NICs, and FPGAs mean some non-GPU Data Center revenue can still be AI-related attach around GPU clusters.
Why keep quarter-on-quarter and year-over-year growth in the data?
The dollar level mixes AI accelerators with traditional EPYC workloads. Growth rates make the cycle easier to read. Examples: Q4 2023 was +42.8% QoQ as MI300X ramped; Q2–Q3 2024 were +114.5% and +122.1% YoY; Q1–Q2 2025 slowed sequentially (−4.8%, −11.8%) even while remaining positive YoY; Q3 2025 reaccelerated +34.0% QoQ; Q2 2026 was +16.3% QoQ and +107.3% YoY. A sequence of fading YoY while QoQ stays negative would be an early warning. Reacceleration alongside NVIDIA, Broadcom, and TSMC is stronger evidence the infrastructure cycle itself is strengthening.
Why are quarters labeled Q1 2021 instead of Q1 FY21?
AMD’s fiscal year ends on the last Saturday of December, so its quarters line up with the calendar: Q1 is January–March. We keep AMD’s own labels (Q1 2021 through Q2 2026). NVIDIA’s fiscal year ends in late January (Q1 is February–April) and Broadcom’s ends in late October (Q1 is November–January). Matching Q1 to Q1 across those charts misaligns the periods. Compare AMD with NVIDIA, Broadcom, and TSMC by calendar date, not by quarter number.