AMD Q3 2026: data center $8.5B, up 97%, Q4 guide $17B

AMD's Q3 2026 revenue reached a record $13.4 billion, with data center sales of $8.5 billion, up 97% year over year, the company reported after the close on November 3, and guided Q4 to about $17 billion. The number that matters for anyone paying for AI tokens is not the revenue line. It is that Anthropic, Meta, Microsoft, OpenAI and Oracle now run workloads on AMD Helios racks. The second GPU source is shipping.

What actually happened

From the press release for the quarter ended September 26:

• Revenue: $13.4 billion, up 45% year over year and 16% from Q2.
• Data center: $8.5 billion, up 97%, "driven by strong demand for AMD EPYC processors and the launch of AMD Instinct MI450 Series GPUs." The segment was 64% of revenue, up from 47% a year ago.
• Margins: GAAP gross margin 54%, non-GAAP 56%. Non-GAAP EPS $2.09, up 74%.
• Helios: the first rackscale systems, built on MI450 GPUs and 6th Gen EPYC, are running models at Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud Infrastructure.
• Guidance: Q4 revenue of approximately $17 billion, plus or minus 3%, about 66% growth at the midpoint. Lisa Su said growth should "accelerate further in the fourth quarter" as Helios and "Venice" production scale.
• The cost of the ramp: inventory rose to $11.5 billion from $7.9 billion at the end of 2025. Free cash flow fell to $554 million from $1.6 billion in Q2.

Client and gaming fell 8% to $3.7 billion. GAAP net income of $2.4 billion includes a $1.1 billion gain on customer warrants, so read the non-GAAP line.

Same evening, server maker Supermicro reported quarterly revenue of $19.9 billion, up from $5.0 billion a year earlier, and raised its full-year outlook to $72 billion to $82 billion. The build-out is accelerating.

Why it matters for your business

You do not buy GPUs. You buy tokens, and token prices are set by how much compute the labs can rent and from whom. Until this year, that supply came from one vendor, and your per-token price carried that premium.

Expect the price curve to keep bending. A second supplier at scale gives the labs leverage on their biggest cost. That reaches your API bill over quarters, not days. Do not lock a two-year rate card on today's prices.

Watch AMD-backed inference endpoints. Oracle and Vultr are standing up Helios capacity, and open-weights models served on MI450 will be priced to win. Keep your model layer swappable so you can take that lane when it opens.

Read the inventory line as a warning. $11.5 billion of parts on the shelf is a bet on sustained orders. If demand softens, that is where the correction starts. Price your AI roadmap on what you can measure, not on supplier confidence.

Sources: AMD Q3 2026 press release, Supermicro Q1 FY2027 press release.