Advanced Micro Devices (AMD) — Q3 FY2026 Earnings Briefing

Fiscal period: Q3 FY2026 · Report date: November 3, 2026

Executive Summary

  • Record quarter: revenue of $13.4 billion rose 45% YoY and beat the high end of guidance, led by a data center segment that nearly doubled.
  • Non-GAAP EPS of $2.09 grew 74% YoY — faster than revenue — as operating margin expanded on a richer mix.
  • Data Center now dominates the mix, and Q4 guides to the first quarter in which data center AI revenue surpasses server revenue in dollars.
  • Management raised 2027 markers across total revenue, server, and data center AI, citing expanded customer deployment plans and secured supply.
  • The tension: memory-cost inflation is dragging the PC business into a planned down 2027 even as data center and embedded more than offset it.

Highlights and Lowlights

Highlights

  1. Data Center revenue reached a record $8.5 billion, up 97% YoY, with segment operating margin expanding to 34% as initial MI450-series sales ramped alongside share-gaining EPYC sales.
  2. Management now expects total revenue to grow more than 80% in 2027 — more than twice the long-term financial model outlined at last November’s Financial Analyst Day — with data center above 80% of the mix.
  3. The 2027 server revenue marker was raised to approximately double YoY, up from the 70%-plus growth expectation given one quarter earlier.
  4. AMD shipped MI455X in volume and put its first Helios racks into production use, with Anthropic, Meta, Microsoft, OpenAI, and Oracle bringing first racks online quickly.
  5. EPYC Venice entered revenue shipment with more than 240 platforms committed by OEM and ODM partners, 20% more than Turin.
  6. Embedded revenue grew 29% YoY, its fourth consecutive quarter of accelerating growth, led by aerospace and defense, automotive, communications, and industrial demand.
  7. AMD ended the quarter with $15.8 billion of cash and short-term investments against $7.1 billion of total debt.

Lowlights

  1. Management is planning for 2027 PC market unit shipments to decline by a double-digit percentage, after sell-through slowed late in the quarter on rising memory and component costs.
  2. Client and Gaming revenue is guided down more than 20% sequentially in Q4 as the PC correction and console decline bite.
  3. Gaming revenue fell 40% YoY to $773 million on lower semi-custom sales at this stage of the console cycle.
  4. Free cash flow narrowed sharply as working-capital build and capacity capex absorbed cash ahead of the data center ramp.
  5. Q4 non-GAAP gross margin guided to approximately 55%, below this quarter’s rate, as faster-growing data center AI revenue mixes up.
  6. Client and Gaming operating income fell to $573 million, 15% of revenue, from $867 million, or 21%, a year ago.
  7. Client revenue declined 4% sequentially as memory costs weighed on PC unit demand, partially offset by a richer high-end mix.

1. The Quarter in Numbers

AMD called it the largest and most profitable quarter in its history, with data center growth accelerating sequentially for the third consecutive quarter and management characterizing execution as “on plan” across the Venice, MI450-series, and Helios ramps. The P&L:

Metric Q3 FY2026 YoY QoQ
Revenue $13.4B (record) +45% +16%
Non-GAAP gross margin 56% +190 bps —
Operating expenses $3.6B +29% —
Operating income $3.9B (record) +75% —
Operating margin 29% +5 pp —
Non-GAAP diluted EPS $2.09 +74% —
GAAP diluted EPS $1.46 — —

The gap between GAAP and non-GAAP EPS is the customer-warrant accounting that began this quarter: GAAP results absorbed a $1.1 billion non-cash charge in other income and expense, primarily reflecting the increase in AMD’s share price and changes in the estimated probability of meeting the warrant conditions as MI450-series GPUs shipped. Management expects these quarterly fair-value remeasurements to increase as the agreements progress, varying with AMD’s stock price and other valuation assumptions; they do not affect non-GAAP results or free cash flow (the full three-part warrant mechanics are covered in the outlook section below).

On cash: operating cash flow from continuing operations was $1.2 billion, and free cash flow came in at $554 million (see Lowlights) as AMD invested in inventory and capital equipment — including back-end capacity dedicated to data center demand — ahead of the ramp. Inventory rose approximately $3 billion sequentially to $11.5 billion to fulfill data center demand, and management expects continued inventory and capacity investment into 2027. The quarter’s $4.75 billion senior notes offering (August) sits within the total debt figure shown in the Highlights.

2. Data Center: EPYC, Venice, and the Agentic CPU Flywheel

Segment results for the quarter:

Segment Revenue YoY QoQ Op income Op margin
Data Center $8.5B (record) +97% +27% $2.9B (+168%) 34% (vs 25%)
Client and Gaming $3.7B −8% — $573M (vs $867M) 15% (vs 21%)
— Client $2.9B +7% −4% — —
— Gaming $773M −40% ~flat — —
Embedded $1.1B +29% +14% $453M (vs $283M) 41% (vs 33%)

Data Center contributed approximately 64% of total revenue, up from 58% last quarter and 47% a year ago — a mix shift management expects to keep pushing higher (see the outlook section). Within server, both cloud and enterprise sales more than doubled YoY, significantly exceeding the prior outlook of more than 80% growth, and AMD says it gained revenue share in the process.

The demand narrative centers on agentic workloads becoming “the next killer app for CPUs in the cloud”: Meta’s Muse and OpenAI’s Dots personal agents illustrate the pace, with more than 10 million people activating Muse agents in their first six weeks. Agents run continuously and in parallel across applications, adding to accelerator demand while creating a substantial CPU demand increase of their own. Turin is now the most widely deployed EPYC generation in the public cloud, with more than 540 instance types available globally. In head nodes, AMD won major deployments with several of the largest model developers and cloud providers, ramping over coming quarters — and OpenAI selected Turin EPYC processors as the host CPUs for its first custom accelerator platforms. Enterprise sell-through was a record, with dozens of first-time EPYC customers added across financial services, communications, retail, and healthcare, and existing customers extending EPYC from specific workloads to a much larger share of their infrastructure.

Venice is the next leg: revenue shipments began in Q3, first OEM systems arrive this quarter, and early demand is the strongest AMD has seen for any new EPYC generation (see the Highlights for the platform-commitment count). Cloud providers are deploying Venice this quarter ahead of instance rollouts planned for early 2027. The complementary Helion reference design targets rack-scale agentic infrastructure — an open design contributed to OCP — whose all-CPU configuration packs up to 144 Venice processors and nearly 74,000 threads into a single rack, which AMD claims is the industry’s highest agent density; every major server OEM and ODM plans to offer it, with first systems shipping this quarter. Management frames the opportunity against a server TAM of approximately $220 billion by 2030, growing at roughly a 50% CAGR over the next three to four years across three areas — traditional cloud and enterprise, agentic CPUs and workloads, and head nodes — with AMD claiming share gains in each.

3. Data Center AI: MI455X, Helios, and the Rack-Scale Ramp

Data center AI revenue set a record on the strength of frontier-lab, hyperscaler, and cloud-provider Instinct deployments for both training and inferencing. Q3 marked the first quarter of MI455X volume shipments and the first Helios rack revenue — a milestone management called “very major” — and the lead customers named in the Highlights brought their first racks online quickly with strong initial performance. Q4 deployments are expected to increase substantially as HPE, Dell, Lenovo, Supermicro, and other OEM and ODM partners begin shipping systems. On positioning, AMD says MI455X leads in compute flops, memory capacity, and scale-out bandwidth, with Helios adding an even larger memory-bandwidth lead and higher scale-up bandwidth than the competition.

Two proof points stand out. In supercomputing, EuroHPC selected Instinct MI430X GPUs and 6th-gen EPYC for LUMI-AI, Europe’s next flagship supercomputer, planned online in the second half of 2027 — MI430X delivers nearly 9x the double-precision performance of the competition, critical for scientific simulations. In software, the new ROCm 10 adds faster start-up across large GPU clusters, greater resilience for long-running workloads, and ROCm.AI, which profiles workloads, identifies bottlenecks, and validates improvements — turning weeks of engineering optimization into hours. Anthropic provided the showcase: one engineer used Claude to bring up their latest model on Instinct MI355X racks over a single weekend.

AMD also announced two acquisitions to be closed by year-end: Taalas, whose team and specialized silicon target low-latency, efficient inference for integration across the solutions portfolio, and World Labs — led by Dr. Fei-Fei Li, who joins as Chief Scientist — bringing spatial-intelligence research that management says will shape future hardware, software, and systems roadmaps.

4. Client and Gaming: Managing Through a Memory-Cost Squeeze

Client revenue grew on strong Ryzen mobile sales, and commercial is becoming a larger part of the business: eight of the 10 largest banks and seven of the 10 largest healthcare and automotive companies now use Ryzen Pro PCs, with HP, Lenovo, and others launching the first commercial notebooks and small-form-factor desktops on the new Ryzen AI Max Pro 400 series — systems that run models with up to 300 billion parameters locally. But the market turned late in the quarter: higher memory and component costs are feeding into system prices, consumer demand is proving sensitive, and AMD is aligning sell-in with sell-through to keep channel inventory healthy. The 2027 planning consequence is in the Lowlights; management expects to outperform that down market on the strength of the Ryzen roadmap and a growing commercial footprint, focused deliberately on enterprise, premium consumer, and desktop channel — the segments with the highest value and, in the low end, AMD’s least exposure. Supply is fungible across the PC and data center businesses, letting AMD move capacity toward data center demand.

Gaming declined on lower semi-custom sales, as expected at this stage of the console cycle; gaming graphics revenue fell YoY while Radeon GPU channel sell-through rose sequentially. The bright spot is Radeon AI Pro, which grew significantly from the prior quarter as developers run and fine-tune models locally, with strong growth expected on performance-per-dollar leadership and an open software stack optimized for the latest open-source models.

5. Embedded: Fourth Straight Acceleration

Embedded delivered its fourth consecutive quarter of accelerating growth (see the Highlights for the growth rate and the segment table for the quarter’s figures), with very strong demand across all end markets and data center buildouts now pulling embedded CPUs and adaptive SoCs into control-plane and optical-networking applications. The design-win engine behind it: AMD remains on track to close more than $18 billion of design wins in 2026, a third consecutive record year, with adaptive design wins increasing every year since the Xilinx acquisition and embedded x86 and semi-custom wins growing even faster. As those wins ramp over the next few years, management expects strong double-digit annual revenue growth and continued share of market outgrowth.

6. Outlook: Q4 Guide and Raised 2027 Markers

Q4 FY2026 guidance:

Q4 FY2026 guide Value
Revenue $17B ± 3% (+66% YoY; ~+27% QoQ)
Data Center ~+50% QoQ
Client and Gaming down >20% QoQ
Embedded double-digit % QoQ growth
Non-GAAP gross margin ~55%
Non-GAAP opex ~$3.75B
Net interest & other ~$4M gain
Non-GAAP effective tax rate 13%
Diluted shares ~1.67B, including the estimated impact of closing the Taalas and World Labs acquisitions; not impacted by customer warrants

The Q4 data center guide carries the quarter’s most important mix event: data center AI revenue should surpass server revenue in dollars for the first time, on what the CFO called a “very significant” sequential MI450 step-up — server is growing strongly, but data center AI will be “much higher” in dollars. For the full year, management also expects FY2026 data center segment revenue to more than double from 2025.

The 2027 markers, raised from 90 days earlier:

2027 marker Commitment
Total revenue growth of more than 80% (see Highlights for the context)
Server revenue ~double YoY (raised from more than 70% growth)
Data center AI revenue “significantly more than double” — grows even faster than projected 90 days ago
Data center mix more than 80% of total revenue, driving the majority of revenue and earnings growth
OpEx growth less than half the revenue growth rate → very significant operating-margin expansion
Embedded strengthens through the year, with typical Q1 seasonality
PC market unit shipments down a double-digit percentage (planning assumption)

Three warrant mechanics, per the CFO, frame how customer-deal accounting layers into the model: stated revenue is always net of the warrant revenue reduction, fixed at grant-date fair value; shares enter the diluted count only once all vesting and exercise conditions are met, with explicit share-count guidance promised in 2027; and the quarterly fair-value remeasurement described in section 1 is non-cash. Beneath the markers, management points to strengthened visibility — customers expanded EPYC and Instinct deployment plans and AMD secured more supply — while acknowledging demand still exceeds supply and supply-chain work continues. MI500 engagements with multiple frontier labs are progressing toward multi-gigawatt deployments, R&D investment is increasing across silicon, software, and systems, and AMD is investing alongside supply-chain partners to secure long-term capacity.

Conclusion

This was a quarter where the mix shift became the story: AMD is now substantially a data center company whose growth is governed by supply rather than demand, and management’s raised 2027 markers — set against a deliberately conservative PC planning assumption — read as a bet that rack-scale execution, not customer appetite, is the binding constraint. The lowlights are real but mostly offsetting: the client correction is a demand-side drag AMD says it can route around via fungible supply, while free cash flow and gross margin are the price of the ramp, not signs of weak end demand. The genuine open question is profitability through the mix shift — management declined to commit to a 2027 gross-margin floor, leaving the Q4 rate as the only benchmark. For the investment view, the quarter strengthens the case that AMD’s data center trajectory is compounding into next year; the watch items are crossover magnitude, OEM Helios shipments, and margin discipline.

Artifacts and Follow-Ups

Key Artifacts — highest priority

  1. Q4 data center AI vs. server crossover. The guide implies Data Center up roughly 50% QoQ (see outlook table) with data center AI surpassing server in dollars; the crossover’s magnitude — “much higher,” not “barely,” per the CFO — sets the entrance rate into 2027 and should be confirmed or refuted by the Q4 segment disclosure.
  2. Helios/OEM ramp. Q4 should show substantially increased Helios deployments as HPE, Dell, Lenovo, and Supermicro begin shipping; baseline: five lead customers’ first racks online in Q3 and volume MI455X shipments started.
  3. Gross margin through the mix shift. Q4 non-GAAP gross margin is guided to approximately 55% (see outlook table) with data center AI slightly dilutive to the corporate average; 2027 carries no floor commitment, so each quarter’s rate versus the Q4 benchmark tests whether mix dilution is contained.
  4. Client correction depth. Client and Gaming is guided down more than 20% QoQ (see outlook table) against a 2027 planning assumption of double-digit PC unit declines; watch sell-through versus sell-in alignment and whether memory-cost pass-through deepens the downturn or stabilizes.
  5. Venice ramp milestones. First OEM systems ship this quarter and cloud providers deploy Venice ahead of early-2027 instance rollouts; baseline: the platform commitments in the Highlights and the strongest early demand of any EPYC generation.

Secondary Tracking Items

  • Taalas and World Labs acquisition closings, embedded in the Q4 share-count guide.
  • Warrant share-count guidance, promised for 2027 once vesting and exercise conditions clarify.
  • MI500 optimization with frontier labs ahead of planned multi-gigawatt deployments.
  • LUMI-AI build progress toward a second-half 2027 online date.
  • Embedded design-win pace toward the 2026 record.
  • AMD’s conference appearances: UBS Technology & AI Conference (December 1) and Barclays Global Technology Conference (December 9).