Test report DSG-4551 · Rev F · tested November 6, 2026

AI Datacenter InfrastructureDevice under test

AMD Q3 2026: Revenue Up 44.7% to $13.38bn, Still No Match for Nvidia

AMD's Q3 2026 revenue rose 44.7% to $13.38bn with net income up 95.3%, but HBM supply limits cap its Instinct GPU challenge to Nvidia, whose lead sits at 6.79X.

Read
4 min
Words
784
Node
5nm
Operator
Marcus Bennett

Spec summary

  1. AMD Q3 2026 revenue up 44.7% to $13.38bn; net income up 95.3% to $2.43bn
  2. Data Center segment revenue $8.55bn, up 96.9%; modeled Instinct GPU sales $3.43bn, up 72.3%
  3. Nvidia's 2026 revenue expected around $354bn, a 6.79X gap over AMD's $52.2bn
  4. Cash hoard more than doubled to $15.78bn
  5. Su expects server revenue to double in 2027 and datacenter to exceed 80% of total revenue
AMD Cannot Catch Nvidia, But It Sure Can Make Some Money - The Next Platform
Fig. AAMD Cannot Catch Nvidia, But It Sure Can Make Some Money - The Next Platform — JacekAbramowicz / pixabay

AMD booked $13.38 billion in revenue in the third quarter of 2026, up 44.7 percent year on year, while operating income rose 2.2X to $2.83 billion. The numbers show a company making serious money in the AI era — but they also show why AMD still cannot catch Nvidia in datacenter GPUs.

Net income climbed 95.3 percent to $2.43 billion, an 18.2 percent net margin. AMD's cash hoard more than doubled to $15.78 billion, now more than one quarter of revenue in the bank.

Why can't AMD take GPU share from Nvidia?

The short answer is memory. AMD cannot get enough high-bandwidth memory (HBM) allocations from Samsung, SK Hynix, and Micron to build more Instinct GPUs. AMD also loads more HBM onto each Instinct MI400 card than Nvidia puts on its Blackwell B200s, B300s, and Rubin R200s, trading balance of compute and memory against raw unit volume.

The revenue gap tells the story. In 2021, Nvidia had $26.9 billion in sales against AMD's $16.4 billion — a 1.64X gap. In 2023, Nvidia's $60.9 billion was 2.69X AMD's $22.7 billion. This year, Wall Street expects Nvidia at roughly $354 billion versus $52.2 billion for AMD, a 6.79X gap, partly because AMD does not own the lucrative scale-across, scale-out, and scale-up AI networking layers. One consensus estimate puts Nvidia at $1,091 billion in calendar 2029 against $164 billion for AMD — 6.65X, essentially unchanged.

The harder truth: AMD is hurting Intel far more than Nvidia. It has a little more than a third of x86 server CPU shipments and just under half of x86 server CPU revenue for the first half of 2026. That share comes from a superior x86 design, careful process node management, and the packaging supply chain to grow shipments faster than Intel at higher average selling prices.

What did the Q3 segment numbers show?

Two thirds of quarterly revenue came from datacenter products — Epyc CPUs, Instinct GPUs, Pensando DPUs, Helios rack designs, and ROCm software:

  • Data Center: $8.55 billion, up 96.9 percent, beating the company's own "more than 80 percent" growth target. Operating income up 2.7X to $2.88 billion, 33.7 percent of revenue.
  • Client (PC CPUs and GPUs): $2.95 billion, up 7.2 percent; modeled operating income down 18.8 percent to $401 million.
  • Gaming (Radeon discrete GPUs): $773 million, down 40.4 percent; modeled operating income down 53.9 percent to $172 million — once a crisis metric, now noise.
  • Embedded (Xilinx FPGAs): $1.11 billion, up 29.4 percent; operating income up 60.1 percent to $453 million, roughly three times the profitability of the PC chip business.

For the third straight quarter, CPUs drove the majority of datacenter revenue. CEO Lisa Su said Epyc sales to hyperscalers and cloud builders "more than doubled," as did sales to enterprise, telco, service provider, government, and academic buyers. Modeling those statements yields roughly $3.53 billion from hyperscalers (up 2.2X), $1.15 billion from enterprise and others (up 105 percent), and total Epyc sales of $4.68 billion, up 116 percent.

Back out Pensando DPUs, NICs, FPGAs, and board components, and Instinct GPU sales land near $3.43 billion, up 72.3 percent. The counterfactual is stark: with 5X the memory allocations and packaging capacity, AMD could almost certainly sell that much more. Possibly 10X. The missing revenue is a direct read on how HBM allocations are shaking out.

What does this change for AMD's roadmap?

The "Venice" Zen6 Epyc processors started shipping for revenue in Q3. First OEM machines arrive before year-end, and OEMs and ODMs are building 240 different Venice platforms — a 20 percent jump over Turin design wins. Helios racks with "Altair" MI455X GPUs have started shipping to Meta Platforms, Microsoft, OpenAI, and Oracle.

Su raised datacenter guidance: 2026 revenue should more than double 2025's, with the AI portion (CPUs, GPUs, DPUs) growing 50 percent sequentially in Q4 as GPU sales surpass CPU sales on a bump of Helios production iron.

"Our visibility into 2027 also continues to strengthen," Su said. "Last quarter, we said we expected annual server revenue to grow more than 70 percent... We now expect server revenue to double year-over-year in 2027 and datacenter AI revenue to grow even faster than we projected 90 days ago." Datacenter should exceed 80 percent of total revenue in 2027.

The AI market is underserved rather than truly competitive, which preserves pricing power for vendors who can ship. AMD will keep making money on what it can manufacture. Catching Nvidia is another matter.

via Google News: GPU datacenter (Source)

Filed under

  • amd
  • nvidia
  • hbm
  • ai-gpus
  • epyc
Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

News editor covering regulation, policy and their impact on hardware layer of artificial intelligence at Die Signal.

681 articles

Same lot · LOT-360D

« Previous articleNext article »